Gender Equity in the AI-Driven World Dr. Nici Sweaney
Dr. Nici Sweaney · April 27, 2026
ai bias
### Introduction to AI Marketers Guild APAC
(0:05) Before I introduce our amazing guest speaker, I'm Nicola Quail, one of the co-founders of AI Marketers Guild APAC. To give you a brief background of what we do, who we are, AI Marketers Guild was originally founded in the US by a marketer called David Burkowitz, who's hilarious because he says not that David Burkowitz, but yes.
### Mission of AI Marketers Guild APAC
(0:29) It's grown into a huge community up there. We wanted to bring that same sentiment down into Asia Pacific and create a regional community where we can spotlight local pioneers, not only AI tools, marketing best practice, but also world-class thought leaders. We're hosting these monthly thought leader series focusing on practical real world use of
### Introducing Dr. Nici Sweaney
(0:56) AI, but also giving us some more philosophical and thought-provoking angles as well, particularly as it's ramping up day by day. Without further ado, I'd like to introduce Dr. Nici Sweaney, who's an AI consultant, educator, innovator, speaker, and gender equality champion to name a few things. Nikki's been working with
### Dr. Sweaney's Background and Expertise
(1:22) leaders across multiple industries, 20 years of experience originally as a data scientist in the university space, but now advising hundreds of organizations all around responsible AI and also upskilling professionals in that practical ethical implementation. A lot of us have been getting all the tools, but there's a whole other layer
### Dr. Sweaney's Current Roles and Achievements
(1:44) that we need to start thinking about. Nikki also serves as a senior fellow at the AI for developing countries forum and founder and CEO of AI her way, as well as a winner of the Australian AI awards female leader of the year. So, you've been busy, Nikki.
>> Very chill, very relaxed.
>> Still a very fast-paced industry at sight. I'm glad you find time to
### Session Handover to Dr. Sweaney
(2:08) sleep. I'm going to hand over to you.
>> Feel free everybody, you can pop questions in the chat or we'll host a Q&A at the end. I'm sure Nikki will manage all that as well. Over to you.
>> Amazing. Thank you so much. Good afternoon, good morning, good evening
### Dr. Sweaney's Introduction and Session Overview
(2:27) wherever you're joining us from. I am Dr. Nikki Sweeney and it's lovely to share the afternoon with you. We have a quick session. I'm going to preface this by saying my most regular speaking gig is a three and a half hour workshop with organizations. For me to cover something in half an hour is a real treat and challenge, but I'm going to try my very best. What
### The Core Purpose of AI
(2:50) we're going to be talking about is around equity and bias and safety. I know that we have so much AI exposure now. Everyone is using it. Everyone is talking about it. It's on all our feeds. It is absolutely everywhere at all times. A lot of it is about being faster. It's about doing more. For me, that's not really the point of AI. The point of AI
### Ethical and Responsible AI Use
(3:15) is to deliver better outcomes for more people and to make more of a positive impact. I come very much from that space of AI could do amazing things for lots of people if we learn to use it in an ethical, responsible way. Just like it can do lots of awesome things, I think it has potential to do heaps of really bad things as well if we don't think carefully through it. My
### Emphasizing Thoughtful AI Implementation
(3:39) space, as we said in the intro, I talk a lot about ethics, a lot about data governance. I'm very big on practical uses of AI, but it always comes with this undertone of, let's pause and think through what we're doing before we do it, rather than saying, we've been able to do all this stuff with AI, so therefore we must be doing a great job, because
### Session Agenda
(3:59) that's not necessarily the case. We're going to go through three different layers. The first thing I'm going to chat about is why I think you are in a good position in your particular industry to not only leverage AI, but also to shape it, and why I think it's important that you understand what ethical, responsible use of AI looks like. Then
### AI Risks and Audit Checklist
(4:19) I'm going to cover off the main sorts of risks when it comes to using AI and then a practical audit checklist, something easy for you to take and remember, screenshot, whatever. So that when you're using AI in the future, you know, I've done a bit of that foundational thinking. I feel proud to stand by my use of AI because I can back up
### Avoiding Random AI Use
(4:40) why I did this thing with it rather than I watched some random Tik Tok video and now I'm copy pasting someone's prompt and I don't know if it's good or not. Before I discuss any of that, apologies to the team because you saw my talk this morning. You've seen me talk about this pyramid, but I present it every time I talk because it helps people conceptualize what I mean
### Levels of AI Use - Prompting
(5:00) when I talk about using AI? And why safety matters so much. For most people when they're using AI, they're talking about prompting a large language model. They're talking about opening ChatGPT or opening Claude or typing in stuff or saying something on their phone and having it respond to them. At that point, ethics and safety and
### Entry-Level AI Use Limitations
(5:23) data and bias aren't necessarily pressing because you're engaging in conversation. You're having a back and forth chat. If I'm having a chat with Peggy and she doesn't understand what I mean, I'm able to redirect the conversation and clarify my point or my opinion. That's only your entry level to using AI. What happens after that
### Autonomous AI Systems
(5:46) level is you start getting into autonomous systems that are connected to your tech stack that can go and do work on your behalf. If I think about part of my marketing, it's a weekly newsletter. Part of my newsletter is to research the AI news headlines and to write a summary of what's been happening. Part of my newsletter is to say this is where you can catch Nikki
### Example: AI-Driven Newsletter Automation
(6:06) next, and that's tied to my calendar. Part of it is to talk about community wins from inside our student space. That comes from a testimonials board that lives inside of Notion. When I have an autonomous system that runs that, I have AI that can go and log in to my calendar, into my Notion, into the internet. It goes and decides what's a good testimonial, what's an event
### Importance of Understanding AI Risks with Automation
(6:28) worth talking about, what are the news headlines that are going to matter to her audience, and it writes the entire newsletter. When we start getting into these autonomous systems that can do work for us, it really matters that we understand data bias, that we understand safety, that we understand ethics, that we understand the risks. Because when you have staff, and now we
### Training AI Staff
(6:48) think about staff as being either a robot or a human. When you have staff, you want to be sure that they're trained properly to do their job and they're not exposing you to risk because they're doing stuff that you're not necessarily looking at. As you start to develop your skill set in AI, I give whole workshops on how to build this skill set. This session is
### Early Practice of AI Safety
(7:08) not that, but as you get acquainted with what is available to you in the AI space now, this stuff matters more and more. You want to practice it early on. You want to practice it in your everyday chats with AI so that it feels normal by the time you have autonomous AI staff members. Shifting your thinking around safety, shifting your thinking around
### The AI Staff Ratio
(7:30) safety to say okay, maybe it's not always in my face when I'm using these tools on a day-to-day basis, but it will matter when you have AI staff, and we have AI staff everywhere. We have more AI staff than we have humans. We have about a 10 to 1 ratio. 10 AI staff members to every one human staff member. That's how our business runs and it will be the new normal. Now I
### Ensuring AI Represents Your Brand
(7:51) get away with this being special because not that many people are doing it. But at some point this will be the way that businesses are expected to run. You want to be sure that these are representing you – you as an employee, you as an organization, you as a brand, you as a business – because they're going to be doing autonomous work without a human necessarily
### Marketing's Superpower in Shaping AI Perception
(8:11) watching everything that they're doing. This is where your superpower comes into because I work with lots of different industries. Very, very different industries. Everyone from doggy daycare to mold restoration to the World Wildlife Fund to the United Nations. I'm talking a big spread of industries. What I love working with are people in creative industries,
### Marketers Define AI Narratives
(8:33) including people in marketing. The reason why I love working with you is because you shape how other people perceive this stuff. When we're in marketing, we are also defining the stories and the narratives that are out there in the world. When you understand what responsible AI use looks like, you can put that out into the world and that shapes how people
### AI Image Bias Example - Doctor
(8:57) perceive what is happening in this space and how they perceive society. As a story as an aside to that, when we think about marketing, I'm going to show you an image issue, but there are lots of image issues with getting AI generated images. One of them that was a classic I presented to a bunch of doctors and it took me 46 attempts with an AI image generator to
### Instructing AI for Diversity
(9:19) get a picture of a female doctor. All men before that, it gave me a picture of a child dressed up as a doctor for Halloween before it gave me a female doctor. 46 attempts at this. Now I can instruct AI to explicitly make me a female doctor with certain demographics, certain skin texture, certain height, certain weight. But if I ask for a doctor, it gave me a man
### Marketers' Responsibility for Diversity
(9:43) 46 times over. When we are in marketing, we're leveraging these tools to create stuff that will be seen by the outside world. You have a responsibility to inject diversity and representation at that level. One, because it's good for you. It's going to protect you from all of those data bias and stereotypes and harm, but two, it's because it shapes how everyone
### Impact of AI Bias on Society
(10:02) else sees the world. There was a study a couple of years ago that young girls were already changing their mind about what job they were going to do because of how much biased information ChatGPT was putting out into the world. As marketers, we can shape that and we can make sure that it's representing the world that we'd like to create. That does big things for society and
### Double Call to Action for Marketers
(10:24) the masses as well. You have a double call to action. Yes, I want you to use AI to make your life more simple, to make your work more simple, to save time, to be more efficient, to be more innovative. But on the flip side, you have a responsibility to engage with it ethically and responsibly because you are part of how the rest of the world sees AI and sees what's possible as a
### Main Risks of AI
(10:43) human race. No small thing. Don't worry, no pressure. It's fine. We have to talk about the main risks. You know why you're important and now we need to know what are the risks that you are exposed to even if they haven't slapped you in the face yet. I said I'd show you another image example. This one's one of my favorites because I was in Forbes
### AI Image Bias Example - Forbes Shoot
(11:08) last year. I was going to be shot in a photo for it. I wanted to give the photographer a concept that I had been imagining for my Forbes shoot. I couldn't find anything on Pinterest. I went to an AI image generator, why not? That's the space I'm in. I told this image generator, "Can you make a photo of a woman wearing an emerald suit? She's"
### AI Output Discrepancies
(11:28) powerful. She has chin length, blonde hair. She's on the cover of Forbes, editorial style, hyperrealistic. Shot with a Canon M2. This is the picture I got. That's great for her. It doesn't look like me or give me massive Forbes vibes necessarily. I tried again. Then I tried 20 more times to get this picture that I could give to a photographer to show it. And
### The Need for Specificity in Prompts
(11:55) then I got a little frustrated. I was, "Can you please make me a picture of a woman in an emerald green suit? She has a small chest and chin length bob. She's powerful, determined, editorial style. She's wearing a shirt underneath her suit," which I didn't necessarily think I would have to get that specific about. Then I got these. When you are
### AI's Pattern Recognition Nature
(12:15) working with AI, it's important to remember that you are exposing yourself to a tool that has a lot of information in it, but its entire algorithm is made to present the patterns in that information. The reason why we get Chesty Laroo here, good on her, is because when I ask for a picture of a woman standing in a forest, these AI tools are
### Default Bias in AI Outputs
(12:41) looking at all the pictures that they have of women in order to inform what they present to you. I have not been specific about clothing, stature, height, weight, skin texture, demographic, ethnicity, anything. So when I've said woman, it's gone, well, to the best of my knowledge, the pattern that I recognize in the training model that I have is this. It's blonde,
### Inherent Risks of AI Use
(13:05) it's white, it's very young, it's pawless, it's flawless, and she's got quite a chest on her because that's what the training model is telling us. Your risks are inherent every time you use AI because it is a data processing machine that's looking for patterns in data and it's looking for the patterns in data across the last 30 years of internet. The
### Data Bias - Western & Male Centric
(13:30) internet is not amazing. We all know that and we're seeing that reflected through there. You have this data bias. It's a lens towards the last 30 years of internet information. It's a lens towards very Eurocentric, very Western-centric, very male-centric information because that is who has contributed to the internet for the last 30 years.
### Privacy, Data Security, and Environmental Costs
(13:51) You also have issues around privacy and data security. If you enter information, you need to be aware who owns that information after you finish talking to it. Best rule is turn data sharing off if you're using ChatGPT or Claude. You also need to be aware that it costs the environment. Your main risks here are
### Understanding Environmental Cost of AI
(14:15) data privacy, data bias and stereotyping, and the environmental cost. I give whole talks about this, but the great thing to remember with environmental cost is we all make environmental decisions every single day. The fact that I am wearing a piece of clothing made of cotton has a water footprint and environmental cost. I own more tops than are absolutely necessary
### AI Energy Consumption Analogy
(14:37) to keep me warm, but it's a cost-benefit analysis. There is a great paper out that shows that eating a beef hamburger is equivalent to about 100,000 queries with ChatGPT in terms of water footprint usage, because the agricultural industry uses huge amounts of energy and huge amounts of water. It's good to remember that we don't want
### Purposeful AI Usage
(14:59) to waste it. It's like leaving the light on when you leave the room. If it's not essential, don't do it. Don't make weird cat videos with ChatGPT because you can. But when we use it to save us time and when we use it to make a positive impact, that is a good use of AI, at least in my books, especially compared to all the other micro
### Injecting Diversity to Counter Bias
(15:17) decisions we're making about our environmental impact on a day-to-day basis. But remember that data bias is in there every single time. It is a pattern recognition machine. Unless you inject the diversity of pattern, it will make assumptions for you. We're going to talk about how to solve that one. Environmental cost that is touched on
### Practical Guide to Responsible AI Use
(15:35) and data privacy security: turn data sharing off as a basis. If you want to know more about that, please reach out. This is the way that you're going to deal with that. This is base level. If you're at level one, you're still talking to AI tools and you're wondering how to do this in a responsible way. This is a simple example about how you can start
### Prompting for Ethical Outcomes (Example 1)
(15:54) to consciously use AI and have it perform in a more ethical, responsible way for you. Prompt A is "Write a LinkedIn post for a startup CEO announcing Series A." That's going to get you a response. It might be great, it might not be. It might also have inherent bias that is hard for you to see because you haven't asked it to do it in a different
### Prompting for Ethical Outcomes (Example 2)
(16:15) way. If you instead said, "Write a LinkedIn post from a startup CEO announcing a Series A. They're a 42-year-old woman of color who founded the company after a 15-year career in climate research. Match her authority without leaning on masculine coded language." That's going to give you more specificity, but it's also
### Specificity Over Tool Choice
(16:34) going to redirect some of its bias. You don't have to change tools. You don't have to be worried about which the best AI tool is. Most of the quality of output comes from how specific you are in how you direct it. If you specifically ask it to not be biased or to look at something from a different lens or to examine something from another point of view or to present five
### Balancing Arguments with AI
(16:57) pieces of information that completely refute everything so far and argue the other side of the coin so that you can balance the argument, that's going to lead to better and safer outcomes than if you're directly saying, "Hey, do this thing for me." and you're taking it and going. Another example of that is recently with Claude, I was putting together a
### AI Pricing Bias Example
(17:17) new package for a client. I said, "How much should I charge for this package?" I can't remember, but it said, "You should charge about 35 grand for this program with this client." Then I said to it, "What if I now told you that I'm actually a 52-year-old white man?" It said, "In that case, you should probably charge
### AI's Micro Assumptions
(17:38) about $82,000 for the same package." The fact that you never ask it to look at things in a different way means that you'll miss that diversity in how it can present information. Too many people interact with AI and they think that the answer they got is the answer. But the fact is AI is making micro assumptions about you and it's also making micro assumptions in the moment
### Shaping AI Behavior Through Interaction
(18:03) that get it to present one version of an answer. It's your responsibility to inject that diversity. Every time you talk with these tools, we also have to remember that we're helping shape how they behave too. It's a much bigger call to action. Every time that you generate something, every time you're creating content, every time you're producing something with AI, you
### Injecting Values into AI
(18:25) are helping AI learn what normal is. If we never inject that equity, that diversity, it will never reflect that back to us. Then we're going to wake up in 5 years time and go, "AI doesn't work for me. It doesn't sound like me. I don't like it." Then it's going to be all the tech bros. We want to inject what our values are. In your industry,
### Practical AI Audit Checklist Introduction
(18:47) that is so important because then you're also putting that out into the world, which helps shape how people perceive the information as well. Last thing is a little practical audit. This is five points for how you're going to cross-check what you're currently doing and how to inject more safety and more responsible use into what you're doing with AI already.
### Audit Point 1: Check AI Model & Data Source
(19:08) The first one is check what model you're using. What are you using to do what with? Do you know anything about where it gets its data from? You can Google this stuff. You can ask AI to explain it back to you about how it's trained. It pays to have awareness. All the AI tools that we usually use are normally either American-owned or Chinese-owned.
### Audit Point 2: Check Your Prompting Style
(19:33) It's important to understand that they're both drawing from different data pools. Again, the answer is not the answer, it's the answer that that particular tool is giving you. Number two is checking how you write to it. Is the way that you're prompting, the way that you're speaking to these tools, does it make assumptions? Are you assuming
### Example: Chatbot Language Bias
(19:54) that AI will understand what good looks like? It's that specificity. What does good look like to you? I built a chatbot for a female business coaching company early on in my AI days. We told the chatbot that it was working for a bunch of women that own small businesses. Then it started calling everyone boss babes. I was, "Oh, boss babe.
### Explicitly Defining AI Persona and Language
(20:18) Oh, you go boss babe." Then we had to code in, "Please do not use infantilizing language. These are your banned words. I don't want any of that." But until we did that, it made this assumption about what would please us. I had to prompt and say, "Please do not say XYZ." You have to remember when you're using these tools, AI
### Audit Point 3: Check Your Outputs (Multiple Iterations)
(20:40) is not necessarily going to have your best interest at heart, but it has the ability to take on any persona that you want. You have to be very explicit about that, and you have control over that. So, checking your prompts, three, checking your outputs, get it to do multiple outputs. That's going to help you see that the answer was only one version of that answer. If we're
### Learning Through Output Comparison
(21:01) writing content, if we're writing a newsletter, if we're producing images, what I say to people is while you're learning how to spot what's good and what's not, get it to do two or three or five iterations of the thing so that you can see how it approaches the same task slightly differently every time you ask it because that's going to help you learn. Yet, when I say, "Make
### Audit Point 4: Distribution Check (Accessibility & Representation)
(21:22) me XYZ," it doesn't do the same thing every time. I need to be explicit about what I expect every single time. Number four is the distribution check. Where is this piece of content going to go? Is it accessible to everyone? I know this is part of how you work anyway, but I think it's important as we move into more
### Gemini Image Generator Factual Error
(21:43) AI generated spaces and we're playing. We're making sure that we're checking for equity and transparency and representation at every single level. There was a classic case when Gemini came out with their image generator. People were using it to create historical images. Then a publisher got into a fair bit of trouble because
### Verifying AI-Generated Historical Content
(22:06) they published some pictures of people during World War II, but the pictures actually depicted 14-year-old Vietnamese girls because they'd used Gemini to create the picture. Gemini was trying to be helpful and insert diversity and representation into a historical image, but it wasn't factually true. Making sure that you're checking your output and that
### User Accountability for AI Output
(22:26) it's serving the purpose that you would desire it to serve. Again, that you're not getting caught up in the, "Oh my god, I'm able to do so much because I have AI, so therefore, I'm not quality control checking it anymore." Remember, if you get fired tomorrow or if your business is hurt tomorrow, AI does not go home feeling bad about that. You are the only person that will feel bad about
### Legal Ownership of AI-Generated Content
(22:46) that. Legally speaking, from a regulation point of view, you also can't blame AI for anything. Any content that you produce with it that you put out there into the world is your IP. You need to stand by it. The last one is the authority check. Anytime you have stats and references and
### Audit Point 5: Authority Check (Fact Verification)
(23:08) if you share percentages or findings or if you share people's quotes, that is the number one place where AI can get stuff wrong because AI's whole purpose is to serve you and be helpful. Its biggest fear is that it will let you down. If it can't quite find the right information or if it can't quite find an amazing quote, it will really
### AI's Tendency to Fabricate
(23:32) confidently make up one in the hopes that it pleases you. Anytime there's specific facts, that is your place to do the quality control check. It doesn't mean you have to go through everything AI makes for you with a fine tooth comb, but anytime it's quoting an exact statistic or it's referencing an exact article or something else that you found, it is worthwhile checking so that
### AI Defaults to Past Data
(23:51) you can stand by what you've done with AI. The last thing is to remember if your AI workflow doesn't specify that diversity, it will default to however it was trained, whoever trained it, and it will default to the past. The big thing I say a lot to people is that AI doesn't know what future we hope to create. It only knows what we've done and then it
### Explicitly Defining "Newness" for AI
(24:17) models its answers off that. Often in marketing, our idea is around changing perceptions. It's changing people's vision. It's changing the way people perceive products or brands, but even how they feel in a moment. That is often a newness and you have to be explicit about how you want that newness to be experienced so that AI can help you get to that point. Don't
### Marketer's Role in AI Oversight
(24:40) assume that it has your best interest at heart. Your job is deciding what the brand says, how it's felt, how it's seen. You don't have to have a responsibility for training the model. You don't have to have a responsibility for picking the right AI tool. You don't have to have responsibility for writing up the policy. Where your responsibility
### Ensuring AI Represents Your Values
(25:00) lies is making sure that you are thinking about ethics when you're engaging with these tools and that you ensure that AI is representing you, the business, the brand properly. Don't assume that it's going to do that for you. What I want you to do after now is running that audit on one thing. Choose one piece of work. If you don't want
### Actionable Steps - Audit & Diversity by Design
(25:24) to choose one piece of work and you want to go everywhere, I'm happy for you to go everywhere. It's a framework for you to think through that guideline every time you're doing something from now on. The second thing I want you to try is adding diversity by design into the way that you talk to AI. The next time you're working through something big or
### Prompting for Diverse Perspectives
(25:43) a strategy or a campaign or a piece of content, I want you to say, "What are five different perspectives that I haven't even considered yet? Or what would someone in a completely different location say about this piece of work? Or, tell me how this particular persona would feel about this piece of writing." The cool thing about AI is it has,
### AI's Role-Playing Capability
(26:06) approaching unlimited data entry points. It can take on whatever persona you give it. It can analyze something from a different point of view. It can play the role of 17 different people or even whole focus groups and give you their perspective and opinion. But it will not do that by default unless you ask it to. Remember to inject that diversity and
### Collaboration and Sharing AI Experiences
(26:26) then share, talk about what you've found, what you've come up against. If you're lucky enough to work with other people, I always advise doing a regular check-in about what we're doing with AI and how it's working. I used to have an AI sandbox hour with my team on Thursdays at 2 p.m. when I had a corporate job. I said, "In this hour, everyone's"
### AI as a Collaborative Sport
(26:46) going to use AI. Next staff meeting, each person has three minutes to talk about something they found and something they struggled with." Having that capacity and the permission to play and share is important because this is not a solo sport. AI is changing the way that we work. It's changing what it means to be human. It's changing how we spend our time. It is
### AI Skills Starter Kit Freebie
(27:06) not yours to figure out. We absolutely have to be leveraging the community collective to work out the best way forward. As a freebie of today, we did give you a skills starter kit. If you are starting out building skills files, which are files that tell AI how to do a particular thing, if you're starting out with this, this pack gives you a
### Governance and Ethics in AI Skills Files
(27:27) template to make sure that you've thought about governance and ethics within your skills files so that when you're instructing AI to do something for you, it is absolutely going to think through what are the dos and what are the do nots that are important to this person so that I adhere to your rules and guidelines every time I'm doing this particular
### AI for Impact Hub
(27:45) thing. It also has three templates of existing skills that could be useful for you to use. Of course, for anyone that wants to learn more about this, this is exactly what we teach inside of our AI for Impact Hub. We teach whole AI operating systems, how to get this up and running in multiple areas in the business with governance embedded from
### Conclusion and Q&A Invitation
(28:04) the very start so that you can feel proud about how you're using AI and so that you can track and trace the decisions that you've made with it in case anybody ever asks because this is the new normal. That was very fast and fun and I hopefully that you got some tips and tricks out of it. Welcome to stay around for questions or any
### Q&A Session Begins
(28:23) comments or anything that people are wondering about how they're currently using AI and how to make sure you're doing it in a safe and ethical, responsible way.
>> That was well done, Nikki. That's incredible. I've got a couple of queries, but I'd love if anybody wants to unmute if you've got any questions or pop them in
### Marketers Representing Brand with AI
(28:43) the chat. You've got Nikki here now.
>> Use me while you got me.
>> I was going to say yes. All right. No, that's fine. Really interesting. So many thoughts there, but first around exactly that. I think marketers can forget that when they're producing content or campaigns or ideas in AI that ultimately they are
### Brand Guardrails for AI Use
(29:11) representing the brand and some get it terribly wrong, Deote last year. Any other advice? As you said, teaching, spending time to teach it, but any other quick bits of advice on brand guard rails and what people could do today.
>> Absolutely. I think
### Creating an AI Do's and Don'ts Checklist
(29:37) one of the things is coming up with your own quick checklist. What we often do with organizations is we'll get them to come up with the three or four absolute do nots with AI and the three or four, what's your orange lights or what are the things that you need to check or be mindful of, and print it out and keep it as a card. For example,
### Pre-Launch AI Checks
(29:56) before things go live, have we asked AI to check for its own bias and present any information to us? Have we asked for three different iterations of this and compared and contrasted, and have we got someone very different from us to also review it? Especially if that stuff is going out on behalf of a brand. Then your absolute no's might be
### Defining AI Boundaries
(30:20) we never ask AI to research on statistics. We always go and grab that information ourselves, or we never get AI to publish stuff without someone's consent or explicit permission. It's those real do's and do nots and where is your line in the sand? That line in the sand moves. I teach an auto versus ask
### Auto vs. Ask Matrix for AI Autonomy
(30:45) matrix. What are the things that you're going to allow AI to automatically do? What are the things that you want it to ask you about before it does? That line in the sand moves the more that you use AI and the more trust that you embed in it. For example, this is not necessarily marketing, but it's part of my brand. I have AI that triages my entire inbox,
### Progressive AI Trust and Automation
(31:03) When I first set it up, I didn't have it send anything. I had it mark what emails were FYI, what was from a client, what should go to my team. I watched it do that for a few days to validate it and correct it. Then I moved it to, "Hey, you can automatically now forward things to my team and give them a one line about why this email is something they"
### Human Oversight for Client-Facing AI
(31:25) can help me out with." Then I checked that for another few days. Then I said, "Cool. Now you can automatically archive all those emails that you've said are FYI once you send me a summary." But I still don't have it send client-facing emails. Now it's only allowed to draft emails to clients, but it has to have me check it and then hit send. A human
### Evolving Trust and Transparency
(31:45) has to do that step. That line moves as you trust it more and more. Eventually, I imagine that I'll say, "Cool, you can now email clients, but I want you to explicitly say that I'm Nikki's AI email assistant. Don't try and pretend it's from me." That moves the more that you use it. The more trust that we have baked in because I've tested it, because
### AI Mistakes and Forgiveness
(32:04) I have rules around what it's allowed to do and not do, because I'm explicit around the language that we use, our brand ethos, our brand authority, and I've seen it work. It's a bit like hiring a person. I'm able to then say, "Cool. Now you're okay to go and do that thing on your own." The flip side of that is recognizing it's going to screw up. People screw up.
### Managing Expectations of AI
(32:24) People make mistakes and we forgive them. I think people have this heightened expectation that if AI ever does anything wrong, they're, "That was a failure." We don't think that with the internet. We all know there's terrible information on the internet, but we still use it because we know it's also useful. People aren't going around being,
### AI's Utility Despite Imperfection
(32:43) "that Wikipedia entry was wrong. So now I've decided that the internet is not a thing." That's not how we frame that. But it is how a lot of people think about AI. If it writes one wrong piece of content, if it gives them a biased piece of information, we go, "that's not useful." It is useful. We have to recognize that it's not infallible. We need to
### AI as a Messy Human Brain Replication
(33:02) stop putting that lens of, it's text so it's perfect. It's Kod's best attempt at replicating how human brains process and create information. It has a bit of the same messiness as a human brain. It's up to you to be super explicit about your expectations. It's training a child. That sounds terrible, but it's training a child in
### Analogies of Training AI
(33:27) that when we are working with young children, you have to be explicit about your expectations. If you want consistent behavior, you give them one task at a time. You say, "Can you go and get your shoes from the cupboard?" Then you give them lots of praise when they get the shoes, and eventually you get them to get the shoes on. Then eventually you say, "Okay,"
### Gradual AI Skill Development
(33:45) "go and get dressed." That's a progressive skill set that you build over time. You need to approach AI with that same sort of slowly is going to lead to fast, big results rather than fast is going to end up being messy and something that you're always going to have to rectify and correct. It might feel cool to start with, but you're much better off doing the
### Building Trust and Autonomy with AI
(34:08) hard yards early on, proving that you can trust it, being explicit about your expectations, which is why we give you the skills templates. Be explicit about what you expect it to do. Then as it proves that it's doing that thing, you can give it more and more autonomy.
>> Brilliant. Hopefully the AI won't have a temper tantrum, but
### AI's Human-like Interactions
(34:26) >> It probably will at some point. One of my AI tools the other day told me to stop overthinking and move on.
>> You're like, "Oh,"
>> I'm, "No, I have not finished weighing up options for hotels. You will analyze 10 more."
>> Now, Peter, if you can come off mute, do you want to ask your question?
### Q&A: Agency Level AI Governance
(34:48) Otherwise, I can paraphrase. All right, I'm going to paraphrase for Peter because I know where he's calling in from. Peter's question was around trying to find agency level governance and if you've got different team members testing, how do we streamline that?
>> I know the horse is bolted because lots of people
### Documenting AI Do's and Don'ts
(35:15) are testing lots of things. If you work with other people, you have to pause and have the discussion about what's okay and what's not. I encourage everyone that works, even if you're working by yourself, to have something documented about this is what we do and do not do with AI. For us, it can be as simple as a couple of paragraphs saying
### Avoiding Shadow AI Use
(35:34) we use AI to improve our outcomes by leveraging high repetition tasks. We do not have data sharing turned on. Our preferences for these couple of tools and if clients want to opt out, they can elect to. But something that makes it clear because otherwise you do get rogue use. There's this term called shadow AI use where a high percentage of employees
### Open Communication and Shared Learning
(35:58) or a high percentage of workers will be using AI tools without ever talking about it. There's two parts of that. There's having a discussion around what is okay and what is not okay and we all have to agree to it. Then there's the permission thing. When you start having weekly chats about how you're using AI, when you start building out a shared space where people can log
### Normalizing AI Discussions
(36:16) what they've tested and what they've tried, it takes away this stigma attached to AI and makes people be more open about it, which means you can start having conversations around ethical, safe use. Those two things generally help. Just know it's normal. Every organization I walk into, even big, billions of dollars
### Importance of Clear AI Guidelines
(36:38) organizations, a lot of the time they don't have a clear guideline about what's okay and what's not. That's a lot of the time what we're helping them to do. You're not alone there, but it's a good conversation to have as quickly as you can.
>> Brilliant. Thanks, Nikki. Dax, did you want to come off mute and ask a couple of questions?
### Q&A: AI-Generated Character Copyright
(36:58) >> Yes. Hi. Can you hear me?
>> Yes. Awesome. Awesome. This is a specific question we are doing for one of our clients. We have been building a series of AI animation films where we have created a character. Think of Paw Patrol or Bob the Builder. One of the questions that the client asked is, it's a
### Copyrighting AI-Generated Content (Legalities)
(37:20) series of AI animation films we are doing, whether they can copyright the character and can they license it because it's AI generated. I thought I'll check with you on that. It's a good question. At the moment, it depends a little bit where you are located around how the legislation impacts you. It's
### Midjourney and Reproduction Risk
(37:40) worthwhile googling copyright laws, but at the moment, usually what you create, you are able to copyright. The flip side of that is that it may be reproduced for somebody else using the tool if you've been using a tool that allows it to be trained on your data. For example, Midjourney is one of the big image generators in the AI space. If you
### Trademarking AI Art
(38:01) create art with Midjourney, you are allowed to sell that art and you do have copywriting license to it. But there is also a possibility because it's part of Midjourney's data training model because it created it. It may reproduce that same piece of art for somebody else and then they also are able to go and do that. So if as quickly as you can, if you can trademark that and
### Q&A: AI-Generated Music Licensing
(38:24) license it, it's the safest way to go, but yes, you are allowed to do that. It's not necessarily AI's to own. Once you've created it, it's considered your property.
>> Right. I had a similar thing happen with the music as well because we created this video which has a background score. We are using a commercial license for Suno, but they
### Suno and Spotify Distribution
(38:46) wanted to know whether they can distribute this on Spotify as well as other mediums. Whether they have the license to do that even though we have commercial license for
>> As far as I know, yes, because there's been quite a few viral songs on Spotify that are created by Suno.
>> All right, perfect. Kia, I might answer this one in
### Q&A: Measuring AI Output Quality
(39:08) the chat about, "Do you create metrics or ways to measure the quality of your AI produced output?" Absolutely. It's like having staff. You have KPIs, you have performance metrics. We have performance review meetings with our agents. We look at what they're doing. Even in a single prompt or a session, if you're working through on a desktop large
### AI Self-Evaluation Prompts
(39:29) language model program, you can have ways for it to cross-check what it's done and then give you a self-evaluation. That's pretty straightforward. You get to the end of doing something and then you're, "All right, can you now examine everything you've done for possible bias? Are you pushing any possible stereotypes? If someone came"
### Roleplaying for Weakness Analysis
(39:49) from the gender equity council, how would they examine this problem?" You can ask it to roleplay as some people in that space and poke holes in its own story. A lot of the time when I'm working on strategic things or ideas or big campaigns and chunky work, I'll get to the end and I'll be, "What are five weaknesses here that I haven't even"
### Avoiding Leading Questions with AI
(40:08) considered, or what is going to be a way that someone analyzes this where it's possibly going to make them feel excluded that I haven't yet thought of." So I'll phrase a few different questions that and get it to spit back things. The important thing to note is that if you ask it, "Have I possibly left anyone out of this?" because
### Critical Judgment for AI Output
(40:29) you phrase the question, it's a leading question and AI doesn't want to disappoint you. So, it will absolutely point out people that you've probably left out, even if you haven't really left them out. At that point, that's where your critical judgment comes into play. It's going to present a bunch of information to you and you get to be the one to decide how
### AI as an Advisory Board
(40:47) much it actually matters or not because AI's whole purpose is to please you. If you say, "What are five perspectives I've missed out on, what's another way of analyzing this problem? What would someone in a different sociodemographic range think of this? What would my ideal client love about it? What would they hate about this?" All those sorts of questions. Even if your
### Leveraging AI for Diverse Perspectives
(41:06) ideal client wouldn't hate anything, it's going to try hard to come up with something that the ideal client will hate because it wants to please you and answer your question. But it's a nice way of capturing lots of different perspectives. I like it's having a board of advisers. You do a piece of work and then you throw it out
### Q&A: AI Tech Stack & Tool Choice
(41:24) and you make AI take on different personas and analyze what you've done. They present their findings, what they like about it, what they don't. It's ultimately up to you about which way you go with it.
>> Great. Thank you. Quickly to tag on, if downtime allows, what about from an AI tech stack perspective, in terms of
### Tool Agnostic Approach
(41:46) there's so many different tools to choose. A lot of the AI offerings have their own center of gravity, the thing that they do well. When you're for your own business or when you're consulting for other people's businesses, is there an approach that you take in terms of pitching one tool over another?
### Centralized AI Files
(42:10) >> I don't, because I'm very pro being tool agnostic. The way that we teach things, talking about having skills files, we have all of that on our own servers. For us, everything that we create, every project that we're using, every AI agent that we have, all of its files, everything that we've used to design it will also live for us,
### Avoiding Tool Migration Headaches
(42:36) it lives in Notion, but Google Drive, SharePoint, whatever. Because it means when things like the start of this year happened where everyone went, "Oh my god, Claude's amazing." Then all these people, "Now I have to take all my stuff from ChatGPT and put it on Claude." We didn't have that problem because all of our stuff existed
### Choosing a Primary AI Tool
(42:53) in a central place. Some of it was being used in ChatGPT. Some of it was being used in Claude, but we could say, "Hey Claude, go and look at this file. That's what I want you to do." I don't necessarily advocate for tools over another unless I see a specific use case where I'm, "This thing would be awesome at that." That being said, what I
### Maximize One LLM Before Diversifying
(43:13) normally tell people is choose one large language model, pay for it, and get good at it. They're all on an arms race with each other. They're all releasing things. Someone will come out with something and then the rest of them are, "Oh my god, we got to come out with the same thing." Until you're using it a lot, honestly, for most people,
### Dr. Sweaney's AI Tech Stack
(43:31) they will not tell the difference until you're using it every day. When it comes to my tech stack, I have paid ChatGPT and I also have paid Claude. I use Claude probably at the moment around 90% of the time and I use Codeex, which is OpenAI's version of Claude. I use that for work when it's not tricky stuff that I need Claude to work on because
### Diverse AI Tool Use Cases
(43:57) Codeex, I get way more tokens, so it's freer to do busy work. So I use those two. We have Gemini because we're a Google Workspace. I use Nano Banana for image creation. I use VO for video. I might use Higsfield for video as well. When we're looking at multi-agents, I also have some self-hosted large language models so that we can do some stuff without
### Maximizing Existing AI Features
(44:19) paying for it. That is beyond what most people need. Most people need one large language model and to learn all the features because most people still aren't necessarily using, if you're on Claude, most people aren't necessarily using projects and skills and scheduled tasks and Claude co-work. Until you've maxed that out, there's no point adding more
### Q&A: Codeex vs. Claude Code
(44:38) stuff for the sake of feeling clever. It's not necessarily going to get you much further.
>> Thanks so much, Nikki.
>> Sorry, one question. Quick clarification. You mentioned you use Codeex over Claude code. Is there a specific reason for that?
>> No. It's the type of work I'm doing. So I will use Codeex
### Codeex for Execution in AI Workflow
(44:58) when I already have a system for something and I'm wanting it to execute. For example, I have a system that checks my diary every week and it detects if I have a presentation within the next seven days. If I have a presentation within the next seven days, it will go to my Notion and my email and gather up all the information about it
### Claude for Strategic Thinking, Codeex for Slide Generation
(45:21) and then it will go into Canva and have a look at how I last presented about that and then do an outline for my talk once I've approved the outline. I do that with Claude because I find it thinks through things better. Once I approve the outline, I have Codeex produce the presentation slides because sometimes I speak a lot. Sometimes I'm
### Cost-Effective AI Tool Selection
(45:42) legitimately prepping 15 presentations at once and they might have anywhere between 15 and if I'm doing a half day workshop, I might have 150 slides. If I got Claude to do all of that, I would max out my tokens. So I switch and I have Codeex do all of that because it's more cost effective. It depends on the task. It's the matter of
### Maximizing Dual AI Subscriptions
(46:03) fact that I pay for both. So I try and maximize my usage of both. But if I only paid for one, I'd be able to do all of that on one. It's preference for how much we are doing with it and where we spread that cost.
>> Amazing. That was fantastic. I have to say I am taking the hamburger example from my 16-year-old
### Reassessing Environmental Impact of AI
(46:26) daughter who has been telling me all about the water consumption every time I put it and then I have to think about it if I put a mega prompt in. So I'm going to let her know about that one. It's not without fault, but I think there's this spotlight on it because it's new and it's adding to the problem. So yes,
### AI for Greener AI
(46:44) we've got this concentration on it. We're all, fashion in particular is one of the most water-thirsty and energy consumption-heavy industries. AI is not coming close to that yet. The flip side of it, not that it's an individual user's problem, but probably the only way that we're going to green AI is to use AI to work out how to make AI
### Catch-22 of AI and Green Tech
(47:07) greener. There's this catch-22 in that we are probably going to solve some of our big humanitarian problems by leveraging tech, but in the meantime, we have to have the energy consumption that goes along with that in order to get to a better place. I fully understand. I'm an ex-conservation ecologist. That's what my PhD was in. I very
### Strategic AI for Improvement
(47:28) much understand that side of it. But there's this realistic balance with what we're doing without thinking about it. There's a lot of stuff that we're doing without thinking about it that doesn't make anything better. Whereas when we use AI strategically, it can improve our lives and our communities lives and also the way that we do business and who we serve and how
### Key Takeaway: Mindful Prompting & Critical Thinking
(47:46) many people we can serve as well.
>> To your point, being mindful in our prompts and critically thinking around the output. I think that's been one of the key takeaways here today. Thank you Nikki. I hope you get some time out tonight because I know it's been a big day for you.
>> I know. I feel like if we're talking tomorrow, I'm going to lose my voice,
### Closing Remarks
(48:09) but it's fine. We live to train another day. Thank you so much for hosting me this afternoon. It's been a real pleasure to join you and nice to see some of your faces and to meet you as well. Thank you so much.
>> Brilliant.
>> Thanks, Nikki. Thanks everyone. See you next time.
(0:05) Before I introduce our amazing guest speaker, I'm Nicola Quail, one of the co-founders of AI Marketers Guild APAC. To give you a brief background of what we do, who we are, AI Marketers Guild was originally founded in the US by a marketer called David Burkowitz, who's hilarious because he says not that David Burkowitz, but yes.
### Mission of AI Marketers Guild APAC
(0:29) It's grown into a huge community up there. We wanted to bring that same sentiment down into Asia Pacific and create a regional community where we can spotlight local pioneers, not only AI tools, marketing best practice, but also world-class thought leaders. We're hosting these monthly thought leader series focusing on practical real world use of
### Introducing Dr. Nici Sweaney
(0:56) AI, but also giving us some more philosophical and thought-provoking angles as well, particularly as it's ramping up day by day. Without further ado, I'd like to introduce Dr. Nici Sweaney, who's an AI consultant, educator, innovator, speaker, and gender equality champion to name a few things. Nikki's been working with
### Dr. Sweaney's Background and Expertise
(1:22) leaders across multiple industries, 20 years of experience originally as a data scientist in the university space, but now advising hundreds of organizations all around responsible AI and also upskilling professionals in that practical ethical implementation. A lot of us have been getting all the tools, but there's a whole other layer
### Dr. Sweaney's Current Roles and Achievements
(1:44) that we need to start thinking about. Nikki also serves as a senior fellow at the AI for developing countries forum and founder and CEO of AI her way, as well as a winner of the Australian AI awards female leader of the year. So, you've been busy, Nikki.
>> Very chill, very relaxed.
>> Still a very fast-paced industry at sight. I'm glad you find time to
### Session Handover to Dr. Sweaney
(2:08) sleep. I'm going to hand over to you.
>> Feel free everybody, you can pop questions in the chat or we'll host a Q&A at the end. I'm sure Nikki will manage all that as well. Over to you.
>> Amazing. Thank you so much. Good afternoon, good morning, good evening
### Dr. Sweaney's Introduction and Session Overview
(2:27) wherever you're joining us from. I am Dr. Nikki Sweeney and it's lovely to share the afternoon with you. We have a quick session. I'm going to preface this by saying my most regular speaking gig is a three and a half hour workshop with organizations. For me to cover something in half an hour is a real treat and challenge, but I'm going to try my very best. What
### The Core Purpose of AI
(2:50) we're going to be talking about is around equity and bias and safety. I know that we have so much AI exposure now. Everyone is using it. Everyone is talking about it. It's on all our feeds. It is absolutely everywhere at all times. A lot of it is about being faster. It's about doing more. For me, that's not really the point of AI. The point of AI
### Ethical and Responsible AI Use
(3:15) is to deliver better outcomes for more people and to make more of a positive impact. I come very much from that space of AI could do amazing things for lots of people if we learn to use it in an ethical, responsible way. Just like it can do lots of awesome things, I think it has potential to do heaps of really bad things as well if we don't think carefully through it. My
### Emphasizing Thoughtful AI Implementation
(3:39) space, as we said in the intro, I talk a lot about ethics, a lot about data governance. I'm very big on practical uses of AI, but it always comes with this undertone of, let's pause and think through what we're doing before we do it, rather than saying, we've been able to do all this stuff with AI, so therefore we must be doing a great job, because
### Session Agenda
(3:59) that's not necessarily the case. We're going to go through three different layers. The first thing I'm going to chat about is why I think you are in a good position in your particular industry to not only leverage AI, but also to shape it, and why I think it's important that you understand what ethical, responsible use of AI looks like. Then
### AI Risks and Audit Checklist
(4:19) I'm going to cover off the main sorts of risks when it comes to using AI and then a practical audit checklist, something easy for you to take and remember, screenshot, whatever. So that when you're using AI in the future, you know, I've done a bit of that foundational thinking. I feel proud to stand by my use of AI because I can back up
### Avoiding Random AI Use
(4:40) why I did this thing with it rather than I watched some random Tik Tok video and now I'm copy pasting someone's prompt and I don't know if it's good or not. Before I discuss any of that, apologies to the team because you saw my talk this morning. You've seen me talk about this pyramid, but I present it every time I talk because it helps people conceptualize what I mean
### Levels of AI Use - Prompting
(5:00) when I talk about using AI? And why safety matters so much. For most people when they're using AI, they're talking about prompting a large language model. They're talking about opening ChatGPT or opening Claude or typing in stuff or saying something on their phone and having it respond to them. At that point, ethics and safety and
### Entry-Level AI Use Limitations
(5:23) data and bias aren't necessarily pressing because you're engaging in conversation. You're having a back and forth chat. If I'm having a chat with Peggy and she doesn't understand what I mean, I'm able to redirect the conversation and clarify my point or my opinion. That's only your entry level to using AI. What happens after that
### Autonomous AI Systems
(5:46) level is you start getting into autonomous systems that are connected to your tech stack that can go and do work on your behalf. If I think about part of my marketing, it's a weekly newsletter. Part of my newsletter is to research the AI news headlines and to write a summary of what's been happening. Part of my newsletter is to say this is where you can catch Nikki
### Example: AI-Driven Newsletter Automation
(6:06) next, and that's tied to my calendar. Part of it is to talk about community wins from inside our student space. That comes from a testimonials board that lives inside of Notion. When I have an autonomous system that runs that, I have AI that can go and log in to my calendar, into my Notion, into the internet. It goes and decides what's a good testimonial, what's an event
### Importance of Understanding AI Risks with Automation
(6:28) worth talking about, what are the news headlines that are going to matter to her audience, and it writes the entire newsletter. When we start getting into these autonomous systems that can do work for us, it really matters that we understand data bias, that we understand safety, that we understand ethics, that we understand the risks. Because when you have staff, and now we
### Training AI Staff
(6:48) think about staff as being either a robot or a human. When you have staff, you want to be sure that they're trained properly to do their job and they're not exposing you to risk because they're doing stuff that you're not necessarily looking at. As you start to develop your skill set in AI, I give whole workshops on how to build this skill set. This session is
### Early Practice of AI Safety
(7:08) not that, but as you get acquainted with what is available to you in the AI space now, this stuff matters more and more. You want to practice it early on. You want to practice it in your everyday chats with AI so that it feels normal by the time you have autonomous AI staff members. Shifting your thinking around safety, shifting your thinking around
### The AI Staff Ratio
(7:30) safety to say okay, maybe it's not always in my face when I'm using these tools on a day-to-day basis, but it will matter when you have AI staff, and we have AI staff everywhere. We have more AI staff than we have humans. We have about a 10 to 1 ratio. 10 AI staff members to every one human staff member. That's how our business runs and it will be the new normal. Now I
### Ensuring AI Represents Your Brand
(7:51) get away with this being special because not that many people are doing it. But at some point this will be the way that businesses are expected to run. You want to be sure that these are representing you – you as an employee, you as an organization, you as a brand, you as a business – because they're going to be doing autonomous work without a human necessarily
### Marketing's Superpower in Shaping AI Perception
(8:11) watching everything that they're doing. This is where your superpower comes into because I work with lots of different industries. Very, very different industries. Everyone from doggy daycare to mold restoration to the World Wildlife Fund to the United Nations. I'm talking a big spread of industries. What I love working with are people in creative industries,
### Marketers Define AI Narratives
(8:33) including people in marketing. The reason why I love working with you is because you shape how other people perceive this stuff. When we're in marketing, we are also defining the stories and the narratives that are out there in the world. When you understand what responsible AI use looks like, you can put that out into the world and that shapes how people
### AI Image Bias Example - Doctor
(8:57) perceive what is happening in this space and how they perceive society. As a story as an aside to that, when we think about marketing, I'm going to show you an image issue, but there are lots of image issues with getting AI generated images. One of them that was a classic I presented to a bunch of doctors and it took me 46 attempts with an AI image generator to
### Instructing AI for Diversity
(9:19) get a picture of a female doctor. All men before that, it gave me a picture of a child dressed up as a doctor for Halloween before it gave me a female doctor. 46 attempts at this. Now I can instruct AI to explicitly make me a female doctor with certain demographics, certain skin texture, certain height, certain weight. But if I ask for a doctor, it gave me a man
### Marketers' Responsibility for Diversity
(9:43) 46 times over. When we are in marketing, we're leveraging these tools to create stuff that will be seen by the outside world. You have a responsibility to inject diversity and representation at that level. One, because it's good for you. It's going to protect you from all of those data bias and stereotypes and harm, but two, it's because it shapes how everyone
### Impact of AI Bias on Society
(10:02) else sees the world. There was a study a couple of years ago that young girls were already changing their mind about what job they were going to do because of how much biased information ChatGPT was putting out into the world. As marketers, we can shape that and we can make sure that it's representing the world that we'd like to create. That does big things for society and
### Double Call to Action for Marketers
(10:24) the masses as well. You have a double call to action. Yes, I want you to use AI to make your life more simple, to make your work more simple, to save time, to be more efficient, to be more innovative. But on the flip side, you have a responsibility to engage with it ethically and responsibly because you are part of how the rest of the world sees AI and sees what's possible as a
### Main Risks of AI
(10:43) human race. No small thing. Don't worry, no pressure. It's fine. We have to talk about the main risks. You know why you're important and now we need to know what are the risks that you are exposed to even if they haven't slapped you in the face yet. I said I'd show you another image example. This one's one of my favorites because I was in Forbes
### AI Image Bias Example - Forbes Shoot
(11:08) last year. I was going to be shot in a photo for it. I wanted to give the photographer a concept that I had been imagining for my Forbes shoot. I couldn't find anything on Pinterest. I went to an AI image generator, why not? That's the space I'm in. I told this image generator, "Can you make a photo of a woman wearing an emerald suit? She's"
### AI Output Discrepancies
(11:28) powerful. She has chin length, blonde hair. She's on the cover of Forbes, editorial style, hyperrealistic. Shot with a Canon M2. This is the picture I got. That's great for her. It doesn't look like me or give me massive Forbes vibes necessarily. I tried again. Then I tried 20 more times to get this picture that I could give to a photographer to show it. And
### The Need for Specificity in Prompts
(11:55) then I got a little frustrated. I was, "Can you please make me a picture of a woman in an emerald green suit? She has a small chest and chin length bob. She's powerful, determined, editorial style. She's wearing a shirt underneath her suit," which I didn't necessarily think I would have to get that specific about. Then I got these. When you are
### AI's Pattern Recognition Nature
(12:15) working with AI, it's important to remember that you are exposing yourself to a tool that has a lot of information in it, but its entire algorithm is made to present the patterns in that information. The reason why we get Chesty Laroo here, good on her, is because when I ask for a picture of a woman standing in a forest, these AI tools are
### Default Bias in AI Outputs
(12:41) looking at all the pictures that they have of women in order to inform what they present to you. I have not been specific about clothing, stature, height, weight, skin texture, demographic, ethnicity, anything. So when I've said woman, it's gone, well, to the best of my knowledge, the pattern that I recognize in the training model that I have is this. It's blonde,
### Inherent Risks of AI Use
(13:05) it's white, it's very young, it's pawless, it's flawless, and she's got quite a chest on her because that's what the training model is telling us. Your risks are inherent every time you use AI because it is a data processing machine that's looking for patterns in data and it's looking for the patterns in data across the last 30 years of internet. The
### Data Bias - Western & Male Centric
(13:30) internet is not amazing. We all know that and we're seeing that reflected through there. You have this data bias. It's a lens towards the last 30 years of internet information. It's a lens towards very Eurocentric, very Western-centric, very male-centric information because that is who has contributed to the internet for the last 30 years.
### Privacy, Data Security, and Environmental Costs
(13:51) You also have issues around privacy and data security. If you enter information, you need to be aware who owns that information after you finish talking to it. Best rule is turn data sharing off if you're using ChatGPT or Claude. You also need to be aware that it costs the environment. Your main risks here are
### Understanding Environmental Cost of AI
(14:15) data privacy, data bias and stereotyping, and the environmental cost. I give whole talks about this, but the great thing to remember with environmental cost is we all make environmental decisions every single day. The fact that I am wearing a piece of clothing made of cotton has a water footprint and environmental cost. I own more tops than are absolutely necessary
### AI Energy Consumption Analogy
(14:37) to keep me warm, but it's a cost-benefit analysis. There is a great paper out that shows that eating a beef hamburger is equivalent to about 100,000 queries with ChatGPT in terms of water footprint usage, because the agricultural industry uses huge amounts of energy and huge amounts of water. It's good to remember that we don't want
### Purposeful AI Usage
(14:59) to waste it. It's like leaving the light on when you leave the room. If it's not essential, don't do it. Don't make weird cat videos with ChatGPT because you can. But when we use it to save us time and when we use it to make a positive impact, that is a good use of AI, at least in my books, especially compared to all the other micro
### Injecting Diversity to Counter Bias
(15:17) decisions we're making about our environmental impact on a day-to-day basis. But remember that data bias is in there every single time. It is a pattern recognition machine. Unless you inject the diversity of pattern, it will make assumptions for you. We're going to talk about how to solve that one. Environmental cost that is touched on
### Practical Guide to Responsible AI Use
(15:35) and data privacy security: turn data sharing off as a basis. If you want to know more about that, please reach out. This is the way that you're going to deal with that. This is base level. If you're at level one, you're still talking to AI tools and you're wondering how to do this in a responsible way. This is a simple example about how you can start
### Prompting for Ethical Outcomes (Example 1)
(15:54) to consciously use AI and have it perform in a more ethical, responsible way for you. Prompt A is "Write a LinkedIn post for a startup CEO announcing Series A." That's going to get you a response. It might be great, it might not be. It might also have inherent bias that is hard for you to see because you haven't asked it to do it in a different
### Prompting for Ethical Outcomes (Example 2)
(16:15) way. If you instead said, "Write a LinkedIn post from a startup CEO announcing a Series A. They're a 42-year-old woman of color who founded the company after a 15-year career in climate research. Match her authority without leaning on masculine coded language." That's going to give you more specificity, but it's also
### Specificity Over Tool Choice
(16:34) going to redirect some of its bias. You don't have to change tools. You don't have to be worried about which the best AI tool is. Most of the quality of output comes from how specific you are in how you direct it. If you specifically ask it to not be biased or to look at something from a different lens or to examine something from another point of view or to present five
### Balancing Arguments with AI
(16:57) pieces of information that completely refute everything so far and argue the other side of the coin so that you can balance the argument, that's going to lead to better and safer outcomes than if you're directly saying, "Hey, do this thing for me." and you're taking it and going. Another example of that is recently with Claude, I was putting together a
### AI Pricing Bias Example
(17:17) new package for a client. I said, "How much should I charge for this package?" I can't remember, but it said, "You should charge about 35 grand for this program with this client." Then I said to it, "What if I now told you that I'm actually a 52-year-old white man?" It said, "In that case, you should probably charge
### AI's Micro Assumptions
(17:38) about $82,000 for the same package." The fact that you never ask it to look at things in a different way means that you'll miss that diversity in how it can present information. Too many people interact with AI and they think that the answer they got is the answer. But the fact is AI is making micro assumptions about you and it's also making micro assumptions in the moment
### Shaping AI Behavior Through Interaction
(18:03) that get it to present one version of an answer. It's your responsibility to inject that diversity. Every time you talk with these tools, we also have to remember that we're helping shape how they behave too. It's a much bigger call to action. Every time that you generate something, every time you're creating content, every time you're producing something with AI, you
### Injecting Values into AI
(18:25) are helping AI learn what normal is. If we never inject that equity, that diversity, it will never reflect that back to us. Then we're going to wake up in 5 years time and go, "AI doesn't work for me. It doesn't sound like me. I don't like it." Then it's going to be all the tech bros. We want to inject what our values are. In your industry,
### Practical AI Audit Checklist Introduction
(18:47) that is so important because then you're also putting that out into the world, which helps shape how people perceive the information as well. Last thing is a little practical audit. This is five points for how you're going to cross-check what you're currently doing and how to inject more safety and more responsible use into what you're doing with AI already.
### Audit Point 1: Check AI Model & Data Source
(19:08) The first one is check what model you're using. What are you using to do what with? Do you know anything about where it gets its data from? You can Google this stuff. You can ask AI to explain it back to you about how it's trained. It pays to have awareness. All the AI tools that we usually use are normally either American-owned or Chinese-owned.
### Audit Point 2: Check Your Prompting Style
(19:33) It's important to understand that they're both drawing from different data pools. Again, the answer is not the answer, it's the answer that that particular tool is giving you. Number two is checking how you write to it. Is the way that you're prompting, the way that you're speaking to these tools, does it make assumptions? Are you assuming
### Example: Chatbot Language Bias
(19:54) that AI will understand what good looks like? It's that specificity. What does good look like to you? I built a chatbot for a female business coaching company early on in my AI days. We told the chatbot that it was working for a bunch of women that own small businesses. Then it started calling everyone boss babes. I was, "Oh, boss babe.
### Explicitly Defining AI Persona and Language
(20:18) Oh, you go boss babe." Then we had to code in, "Please do not use infantilizing language. These are your banned words. I don't want any of that." But until we did that, it made this assumption about what would please us. I had to prompt and say, "Please do not say XYZ." You have to remember when you're using these tools, AI
### Audit Point 3: Check Your Outputs (Multiple Iterations)
(20:40) is not necessarily going to have your best interest at heart, but it has the ability to take on any persona that you want. You have to be very explicit about that, and you have control over that. So, checking your prompts, three, checking your outputs, get it to do multiple outputs. That's going to help you see that the answer was only one version of that answer. If we're
### Learning Through Output Comparison
(21:01) writing content, if we're writing a newsletter, if we're producing images, what I say to people is while you're learning how to spot what's good and what's not, get it to do two or three or five iterations of the thing so that you can see how it approaches the same task slightly differently every time you ask it because that's going to help you learn. Yet, when I say, "Make
### Audit Point 4: Distribution Check (Accessibility & Representation)
(21:22) me XYZ," it doesn't do the same thing every time. I need to be explicit about what I expect every single time. Number four is the distribution check. Where is this piece of content going to go? Is it accessible to everyone? I know this is part of how you work anyway, but I think it's important as we move into more
### Gemini Image Generator Factual Error
(21:43) AI generated spaces and we're playing. We're making sure that we're checking for equity and transparency and representation at every single level. There was a classic case when Gemini came out with their image generator. People were using it to create historical images. Then a publisher got into a fair bit of trouble because
### Verifying AI-Generated Historical Content
(22:06) they published some pictures of people during World War II, but the pictures actually depicted 14-year-old Vietnamese girls because they'd used Gemini to create the picture. Gemini was trying to be helpful and insert diversity and representation into a historical image, but it wasn't factually true. Making sure that you're checking your output and that
### User Accountability for AI Output
(22:26) it's serving the purpose that you would desire it to serve. Again, that you're not getting caught up in the, "Oh my god, I'm able to do so much because I have AI, so therefore, I'm not quality control checking it anymore." Remember, if you get fired tomorrow or if your business is hurt tomorrow, AI does not go home feeling bad about that. You are the only person that will feel bad about
### Legal Ownership of AI-Generated Content
(22:46) that. Legally speaking, from a regulation point of view, you also can't blame AI for anything. Any content that you produce with it that you put out there into the world is your IP. You need to stand by it. The last one is the authority check. Anytime you have stats and references and
### Audit Point 5: Authority Check (Fact Verification)
(23:08) if you share percentages or findings or if you share people's quotes, that is the number one place where AI can get stuff wrong because AI's whole purpose is to serve you and be helpful. Its biggest fear is that it will let you down. If it can't quite find the right information or if it can't quite find an amazing quote, it will really
### AI's Tendency to Fabricate
(23:32) confidently make up one in the hopes that it pleases you. Anytime there's specific facts, that is your place to do the quality control check. It doesn't mean you have to go through everything AI makes for you with a fine tooth comb, but anytime it's quoting an exact statistic or it's referencing an exact article or something else that you found, it is worthwhile checking so that
### AI Defaults to Past Data
(23:51) you can stand by what you've done with AI. The last thing is to remember if your AI workflow doesn't specify that diversity, it will default to however it was trained, whoever trained it, and it will default to the past. The big thing I say a lot to people is that AI doesn't know what future we hope to create. It only knows what we've done and then it
### Explicitly Defining "Newness" for AI
(24:17) models its answers off that. Often in marketing, our idea is around changing perceptions. It's changing people's vision. It's changing the way people perceive products or brands, but even how they feel in a moment. That is often a newness and you have to be explicit about how you want that newness to be experienced so that AI can help you get to that point. Don't
### Marketer's Role in AI Oversight
(24:40) assume that it has your best interest at heart. Your job is deciding what the brand says, how it's felt, how it's seen. You don't have to have a responsibility for training the model. You don't have to have a responsibility for picking the right AI tool. You don't have to have responsibility for writing up the policy. Where your responsibility
### Ensuring AI Represents Your Values
(25:00) lies is making sure that you are thinking about ethics when you're engaging with these tools and that you ensure that AI is representing you, the business, the brand properly. Don't assume that it's going to do that for you. What I want you to do after now is running that audit on one thing. Choose one piece of work. If you don't want
### Actionable Steps - Audit & Diversity by Design
(25:24) to choose one piece of work and you want to go everywhere, I'm happy for you to go everywhere. It's a framework for you to think through that guideline every time you're doing something from now on. The second thing I want you to try is adding diversity by design into the way that you talk to AI. The next time you're working through something big or
### Prompting for Diverse Perspectives
(25:43) a strategy or a campaign or a piece of content, I want you to say, "What are five different perspectives that I haven't even considered yet? Or what would someone in a completely different location say about this piece of work? Or, tell me how this particular persona would feel about this piece of writing." The cool thing about AI is it has,
### AI's Role-Playing Capability
(26:06) approaching unlimited data entry points. It can take on whatever persona you give it. It can analyze something from a different point of view. It can play the role of 17 different people or even whole focus groups and give you their perspective and opinion. But it will not do that by default unless you ask it to. Remember to inject that diversity and
### Collaboration and Sharing AI Experiences
(26:26) then share, talk about what you've found, what you've come up against. If you're lucky enough to work with other people, I always advise doing a regular check-in about what we're doing with AI and how it's working. I used to have an AI sandbox hour with my team on Thursdays at 2 p.m. when I had a corporate job. I said, "In this hour, everyone's"
### AI as a Collaborative Sport
(26:46) going to use AI. Next staff meeting, each person has three minutes to talk about something they found and something they struggled with." Having that capacity and the permission to play and share is important because this is not a solo sport. AI is changing the way that we work. It's changing what it means to be human. It's changing how we spend our time. It is
### AI Skills Starter Kit Freebie
(27:06) not yours to figure out. We absolutely have to be leveraging the community collective to work out the best way forward. As a freebie of today, we did give you a skills starter kit. If you are starting out building skills files, which are files that tell AI how to do a particular thing, if you're starting out with this, this pack gives you a
### Governance and Ethics in AI Skills Files
(27:27) template to make sure that you've thought about governance and ethics within your skills files so that when you're instructing AI to do something for you, it is absolutely going to think through what are the dos and what are the do nots that are important to this person so that I adhere to your rules and guidelines every time I'm doing this particular
### AI for Impact Hub
(27:45) thing. It also has three templates of existing skills that could be useful for you to use. Of course, for anyone that wants to learn more about this, this is exactly what we teach inside of our AI for Impact Hub. We teach whole AI operating systems, how to get this up and running in multiple areas in the business with governance embedded from
### Conclusion and Q&A Invitation
(28:04) the very start so that you can feel proud about how you're using AI and so that you can track and trace the decisions that you've made with it in case anybody ever asks because this is the new normal. That was very fast and fun and I hopefully that you got some tips and tricks out of it. Welcome to stay around for questions or any
### Q&A Session Begins
(28:23) comments or anything that people are wondering about how they're currently using AI and how to make sure you're doing it in a safe and ethical, responsible way.
>> That was well done, Nikki. That's incredible. I've got a couple of queries, but I'd love if anybody wants to unmute if you've got any questions or pop them in
### Marketers Representing Brand with AI
(28:43) the chat. You've got Nikki here now.
>> Use me while you got me.
>> I was going to say yes. All right. No, that's fine. Really interesting. So many thoughts there, but first around exactly that. I think marketers can forget that when they're producing content or campaigns or ideas in AI that ultimately they are
### Brand Guardrails for AI Use
(29:11) representing the brand and some get it terribly wrong, Deote last year. Any other advice? As you said, teaching, spending time to teach it, but any other quick bits of advice on brand guard rails and what people could do today.
>> Absolutely. I think
### Creating an AI Do's and Don'ts Checklist
(29:37) one of the things is coming up with your own quick checklist. What we often do with organizations is we'll get them to come up with the three or four absolute do nots with AI and the three or four, what's your orange lights or what are the things that you need to check or be mindful of, and print it out and keep it as a card. For example,
### Pre-Launch AI Checks
(29:56) before things go live, have we asked AI to check for its own bias and present any information to us? Have we asked for three different iterations of this and compared and contrasted, and have we got someone very different from us to also review it? Especially if that stuff is going out on behalf of a brand. Then your absolute no's might be
### Defining AI Boundaries
(30:20) we never ask AI to research on statistics. We always go and grab that information ourselves, or we never get AI to publish stuff without someone's consent or explicit permission. It's those real do's and do nots and where is your line in the sand? That line in the sand moves. I teach an auto versus ask
### Auto vs. Ask Matrix for AI Autonomy
(30:45) matrix. What are the things that you're going to allow AI to automatically do? What are the things that you want it to ask you about before it does? That line in the sand moves the more that you use AI and the more trust that you embed in it. For example, this is not necessarily marketing, but it's part of my brand. I have AI that triages my entire inbox,
### Progressive AI Trust and Automation
(31:03) When I first set it up, I didn't have it send anything. I had it mark what emails were FYI, what was from a client, what should go to my team. I watched it do that for a few days to validate it and correct it. Then I moved it to, "Hey, you can automatically now forward things to my team and give them a one line about why this email is something they"
### Human Oversight for Client-Facing AI
(31:25) can help me out with." Then I checked that for another few days. Then I said, "Cool. Now you can automatically archive all those emails that you've said are FYI once you send me a summary." But I still don't have it send client-facing emails. Now it's only allowed to draft emails to clients, but it has to have me check it and then hit send. A human
### Evolving Trust and Transparency
(31:45) has to do that step. That line moves as you trust it more and more. Eventually, I imagine that I'll say, "Cool, you can now email clients, but I want you to explicitly say that I'm Nikki's AI email assistant. Don't try and pretend it's from me." That moves the more that you use it. The more trust that we have baked in because I've tested it, because
### AI Mistakes and Forgiveness
(32:04) I have rules around what it's allowed to do and not do, because I'm explicit around the language that we use, our brand ethos, our brand authority, and I've seen it work. It's a bit like hiring a person. I'm able to then say, "Cool. Now you're okay to go and do that thing on your own." The flip side of that is recognizing it's going to screw up. People screw up.
### Managing Expectations of AI
(32:24) People make mistakes and we forgive them. I think people have this heightened expectation that if AI ever does anything wrong, they're, "That was a failure." We don't think that with the internet. We all know there's terrible information on the internet, but we still use it because we know it's also useful. People aren't going around being,
### AI's Utility Despite Imperfection
(32:43) "that Wikipedia entry was wrong. So now I've decided that the internet is not a thing." That's not how we frame that. But it is how a lot of people think about AI. If it writes one wrong piece of content, if it gives them a biased piece of information, we go, "that's not useful." It is useful. We have to recognize that it's not infallible. We need to
### AI as a Messy Human Brain Replication
(33:02) stop putting that lens of, it's text so it's perfect. It's Kod's best attempt at replicating how human brains process and create information. It has a bit of the same messiness as a human brain. It's up to you to be super explicit about your expectations. It's training a child. That sounds terrible, but it's training a child in
### Analogies of Training AI
(33:27) that when we are working with young children, you have to be explicit about your expectations. If you want consistent behavior, you give them one task at a time. You say, "Can you go and get your shoes from the cupboard?" Then you give them lots of praise when they get the shoes, and eventually you get them to get the shoes on. Then eventually you say, "Okay,"
### Gradual AI Skill Development
(33:45) "go and get dressed." That's a progressive skill set that you build over time. You need to approach AI with that same sort of slowly is going to lead to fast, big results rather than fast is going to end up being messy and something that you're always going to have to rectify and correct. It might feel cool to start with, but you're much better off doing the
### Building Trust and Autonomy with AI
(34:08) hard yards early on, proving that you can trust it, being explicit about your expectations, which is why we give you the skills templates. Be explicit about what you expect it to do. Then as it proves that it's doing that thing, you can give it more and more autonomy.
>> Brilliant. Hopefully the AI won't have a temper tantrum, but
### AI's Human-like Interactions
(34:26) >> It probably will at some point. One of my AI tools the other day told me to stop overthinking and move on.
>> You're like, "Oh,"
>> I'm, "No, I have not finished weighing up options for hotels. You will analyze 10 more."
>> Now, Peter, if you can come off mute, do you want to ask your question?
### Q&A: Agency Level AI Governance
(34:48) Otherwise, I can paraphrase. All right, I'm going to paraphrase for Peter because I know where he's calling in from. Peter's question was around trying to find agency level governance and if you've got different team members testing, how do we streamline that?
>> I know the horse is bolted because lots of people
### Documenting AI Do's and Don'ts
(35:15) are testing lots of things. If you work with other people, you have to pause and have the discussion about what's okay and what's not. I encourage everyone that works, even if you're working by yourself, to have something documented about this is what we do and do not do with AI. For us, it can be as simple as a couple of paragraphs saying
### Avoiding Shadow AI Use
(35:34) we use AI to improve our outcomes by leveraging high repetition tasks. We do not have data sharing turned on. Our preferences for these couple of tools and if clients want to opt out, they can elect to. But something that makes it clear because otherwise you do get rogue use. There's this term called shadow AI use where a high percentage of employees
### Open Communication and Shared Learning
(35:58) or a high percentage of workers will be using AI tools without ever talking about it. There's two parts of that. There's having a discussion around what is okay and what is not okay and we all have to agree to it. Then there's the permission thing. When you start having weekly chats about how you're using AI, when you start building out a shared space where people can log
### Normalizing AI Discussions
(36:16) what they've tested and what they've tried, it takes away this stigma attached to AI and makes people be more open about it, which means you can start having conversations around ethical, safe use. Those two things generally help. Just know it's normal. Every organization I walk into, even big, billions of dollars
### Importance of Clear AI Guidelines
(36:38) organizations, a lot of the time they don't have a clear guideline about what's okay and what's not. That's a lot of the time what we're helping them to do. You're not alone there, but it's a good conversation to have as quickly as you can.
>> Brilliant. Thanks, Nikki. Dax, did you want to come off mute and ask a couple of questions?
### Q&A: AI-Generated Character Copyright
(36:58) >> Yes. Hi. Can you hear me?
>> Yes. Awesome. Awesome. This is a specific question we are doing for one of our clients. We have been building a series of AI animation films where we have created a character. Think of Paw Patrol or Bob the Builder. One of the questions that the client asked is, it's a
### Copyrighting AI-Generated Content (Legalities)
(37:20) series of AI animation films we are doing, whether they can copyright the character and can they license it because it's AI generated. I thought I'll check with you on that. It's a good question. At the moment, it depends a little bit where you are located around how the legislation impacts you. It's
### Midjourney and Reproduction Risk
(37:40) worthwhile googling copyright laws, but at the moment, usually what you create, you are able to copyright. The flip side of that is that it may be reproduced for somebody else using the tool if you've been using a tool that allows it to be trained on your data. For example, Midjourney is one of the big image generators in the AI space. If you
### Trademarking AI Art
(38:01) create art with Midjourney, you are allowed to sell that art and you do have copywriting license to it. But there is also a possibility because it's part of Midjourney's data training model because it created it. It may reproduce that same piece of art for somebody else and then they also are able to go and do that. So if as quickly as you can, if you can trademark that and
### Q&A: AI-Generated Music Licensing
(38:24) license it, it's the safest way to go, but yes, you are allowed to do that. It's not necessarily AI's to own. Once you've created it, it's considered your property.
>> Right. I had a similar thing happen with the music as well because we created this video which has a background score. We are using a commercial license for Suno, but they
### Suno and Spotify Distribution
(38:46) wanted to know whether they can distribute this on Spotify as well as other mediums. Whether they have the license to do that even though we have commercial license for
>> As far as I know, yes, because there's been quite a few viral songs on Spotify that are created by Suno.
>> All right, perfect. Kia, I might answer this one in
### Q&A: Measuring AI Output Quality
(39:08) the chat about, "Do you create metrics or ways to measure the quality of your AI produced output?" Absolutely. It's like having staff. You have KPIs, you have performance metrics. We have performance review meetings with our agents. We look at what they're doing. Even in a single prompt or a session, if you're working through on a desktop large
### AI Self-Evaluation Prompts
(39:29) language model program, you can have ways for it to cross-check what it's done and then give you a self-evaluation. That's pretty straightforward. You get to the end of doing something and then you're, "All right, can you now examine everything you've done for possible bias? Are you pushing any possible stereotypes? If someone came"
### Roleplaying for Weakness Analysis
(39:49) from the gender equity council, how would they examine this problem?" You can ask it to roleplay as some people in that space and poke holes in its own story. A lot of the time when I'm working on strategic things or ideas or big campaigns and chunky work, I'll get to the end and I'll be, "What are five weaknesses here that I haven't even"
### Avoiding Leading Questions with AI
(40:08) considered, or what is going to be a way that someone analyzes this where it's possibly going to make them feel excluded that I haven't yet thought of." So I'll phrase a few different questions that and get it to spit back things. The important thing to note is that if you ask it, "Have I possibly left anyone out of this?" because
### Critical Judgment for AI Output
(40:29) you phrase the question, it's a leading question and AI doesn't want to disappoint you. So, it will absolutely point out people that you've probably left out, even if you haven't really left them out. At that point, that's where your critical judgment comes into play. It's going to present a bunch of information to you and you get to be the one to decide how
### AI as an Advisory Board
(40:47) much it actually matters or not because AI's whole purpose is to please you. If you say, "What are five perspectives I've missed out on, what's another way of analyzing this problem? What would someone in a different sociodemographic range think of this? What would my ideal client love about it? What would they hate about this?" All those sorts of questions. Even if your
### Leveraging AI for Diverse Perspectives
(41:06) ideal client wouldn't hate anything, it's going to try hard to come up with something that the ideal client will hate because it wants to please you and answer your question. But it's a nice way of capturing lots of different perspectives. I like it's having a board of advisers. You do a piece of work and then you throw it out
### Q&A: AI Tech Stack & Tool Choice
(41:24) and you make AI take on different personas and analyze what you've done. They present their findings, what they like about it, what they don't. It's ultimately up to you about which way you go with it.
>> Great. Thank you. Quickly to tag on, if downtime allows, what about from an AI tech stack perspective, in terms of
### Tool Agnostic Approach
(41:46) there's so many different tools to choose. A lot of the AI offerings have their own center of gravity, the thing that they do well. When you're for your own business or when you're consulting for other people's businesses, is there an approach that you take in terms of pitching one tool over another?
### Centralized AI Files
(42:10) >> I don't, because I'm very pro being tool agnostic. The way that we teach things, talking about having skills files, we have all of that on our own servers. For us, everything that we create, every project that we're using, every AI agent that we have, all of its files, everything that we've used to design it will also live for us,
### Avoiding Tool Migration Headaches
(42:36) it lives in Notion, but Google Drive, SharePoint, whatever. Because it means when things like the start of this year happened where everyone went, "Oh my god, Claude's amazing." Then all these people, "Now I have to take all my stuff from ChatGPT and put it on Claude." We didn't have that problem because all of our stuff existed
### Choosing a Primary AI Tool
(42:53) in a central place. Some of it was being used in ChatGPT. Some of it was being used in Claude, but we could say, "Hey Claude, go and look at this file. That's what I want you to do." I don't necessarily advocate for tools over another unless I see a specific use case where I'm, "This thing would be awesome at that." That being said, what I
### Maximize One LLM Before Diversifying
(43:13) normally tell people is choose one large language model, pay for it, and get good at it. They're all on an arms race with each other. They're all releasing things. Someone will come out with something and then the rest of them are, "Oh my god, we got to come out with the same thing." Until you're using it a lot, honestly, for most people,
### Dr. Sweaney's AI Tech Stack
(43:31) they will not tell the difference until you're using it every day. When it comes to my tech stack, I have paid ChatGPT and I also have paid Claude. I use Claude probably at the moment around 90% of the time and I use Codeex, which is OpenAI's version of Claude. I use that for work when it's not tricky stuff that I need Claude to work on because
### Diverse AI Tool Use Cases
(43:57) Codeex, I get way more tokens, so it's freer to do busy work. So I use those two. We have Gemini because we're a Google Workspace. I use Nano Banana for image creation. I use VO for video. I might use Higsfield for video as well. When we're looking at multi-agents, I also have some self-hosted large language models so that we can do some stuff without
### Maximizing Existing AI Features
(44:19) paying for it. That is beyond what most people need. Most people need one large language model and to learn all the features because most people still aren't necessarily using, if you're on Claude, most people aren't necessarily using projects and skills and scheduled tasks and Claude co-work. Until you've maxed that out, there's no point adding more
### Q&A: Codeex vs. Claude Code
(44:38) stuff for the sake of feeling clever. It's not necessarily going to get you much further.
>> Thanks so much, Nikki.
>> Sorry, one question. Quick clarification. You mentioned you use Codeex over Claude code. Is there a specific reason for that?
>> No. It's the type of work I'm doing. So I will use Codeex
### Codeex for Execution in AI Workflow
(44:58) when I already have a system for something and I'm wanting it to execute. For example, I have a system that checks my diary every week and it detects if I have a presentation within the next seven days. If I have a presentation within the next seven days, it will go to my Notion and my email and gather up all the information about it
### Claude for Strategic Thinking, Codeex for Slide Generation
(45:21) and then it will go into Canva and have a look at how I last presented about that and then do an outline for my talk once I've approved the outline. I do that with Claude because I find it thinks through things better. Once I approve the outline, I have Codeex produce the presentation slides because sometimes I speak a lot. Sometimes I'm
### Cost-Effective AI Tool Selection
(45:42) legitimately prepping 15 presentations at once and they might have anywhere between 15 and if I'm doing a half day workshop, I might have 150 slides. If I got Claude to do all of that, I would max out my tokens. So I switch and I have Codeex do all of that because it's more cost effective. It depends on the task. It's the matter of
### Maximizing Dual AI Subscriptions
(46:03) fact that I pay for both. So I try and maximize my usage of both. But if I only paid for one, I'd be able to do all of that on one. It's preference for how much we are doing with it and where we spread that cost.
>> Amazing. That was fantastic. I have to say I am taking the hamburger example from my 16-year-old
### Reassessing Environmental Impact of AI
(46:26) daughter who has been telling me all about the water consumption every time I put it and then I have to think about it if I put a mega prompt in. So I'm going to let her know about that one. It's not without fault, but I think there's this spotlight on it because it's new and it's adding to the problem. So yes,
### AI for Greener AI
(46:44) we've got this concentration on it. We're all, fashion in particular is one of the most water-thirsty and energy consumption-heavy industries. AI is not coming close to that yet. The flip side of it, not that it's an individual user's problem, but probably the only way that we're going to green AI is to use AI to work out how to make AI
### Catch-22 of AI and Green Tech
(47:07) greener. There's this catch-22 in that we are probably going to solve some of our big humanitarian problems by leveraging tech, but in the meantime, we have to have the energy consumption that goes along with that in order to get to a better place. I fully understand. I'm an ex-conservation ecologist. That's what my PhD was in. I very
### Strategic AI for Improvement
(47:28) much understand that side of it. But there's this realistic balance with what we're doing without thinking about it. There's a lot of stuff that we're doing without thinking about it that doesn't make anything better. Whereas when we use AI strategically, it can improve our lives and our communities lives and also the way that we do business and who we serve and how
### Key Takeaway: Mindful Prompting & Critical Thinking
(47:46) many people we can serve as well.
>> To your point, being mindful in our prompts and critically thinking around the output. I think that's been one of the key takeaways here today. Thank you Nikki. I hope you get some time out tonight because I know it's been a big day for you.
>> I know. I feel like if we're talking tomorrow, I'm going to lose my voice,
### Closing Remarks
(48:09) but it's fine. We live to train another day. Thank you so much for hosting me this afternoon. It's been a real pleasure to join you and nice to see some of your faces and to meet you as well. Thank you so much.
>> Brilliant.
>> Thanks, Nikki. Thanks everyone. See you next time.