The Seven Cardinal Sins of AI
Marco Andre · October 9, 2024
hallucinationai dependenceai bias
**(00:03)**
We have a special guest: Marco Andre. I first connected with Marco last summer, just as we were getting ready to launch this community, and he shared some brilliant insights on AI through his work at Novartis. His perspectives seemed like the perfect fit for sparking inspiration and discussion here.
Before Marco dives in, just a reminder: this is a community conversation, not a webinar. Marco will present his “Seven Cardinal Sins of AI,” but we encourage you to ask questions and jump into the conversation, especially since this is a fitting time for reflection on topics like these!
**(01:57)**
Marco, please go ahead.
**(02:44)**
Thank you! So, I’m skipping a formal intro because I have a confession to make: I have “sinned” in the world of AI—I’ve used it! Despite all the concerns we hear, I’ve continued using AI regularly. If you can, raise your hand—who here has used AI to write a work email? Quite a few. Who planned their last trip using ChatGPT or another AI tool? And lastly, who brainstormed an idea with AI before discussing it with a human? You see, I’m not alone; we’re all “sinners.” So, let me warn you about the “Seven Cardinal Sins of AI.”
**(03:37)**
The first sin: *Hallucination—the Fabricator.* AI generates information so convincingly that it often presents falsehoods as credible truths, which can be risky. For instance, when Google launched AI-generated overviews, one result advised smoking during pregnancy! While some AI-generated content can be funny, the implications of hallucination in business contexts are serious, with potential legal, trust, and misinformation risks. On the flip side, hallucination is also where creativity lies—where AI can help us see things from new perspectives, like bringing origami to life. So, hallucination can mislead, but it can also inspire.
**(06:26)**
Second, *Dependence—the Crutch.* AI reliance could make us overly dependent, so much so that we’d struggle to function without it. Take spell check, for example—most of us use it regularly. Could we go back to a world without it? And then there’s navigation. Remember when we used paper maps? Now, we have GPS, and we rely on it so much that some people have driven right into the ocean following its directions. So, AI can make us reliant, but we can also choose to use it selectively.
**(09:56)**
Third, *Bias—the Unfair Judge.* AI amplifies biases present in its training data, reflecting societal inequities. For instance, an AI might associate “nurse” with women and “programmer” with men due to biased data inputs. But bias isn’t exclusive to AI; in the past, biases affected even simple systems like photo finishes at the Olympics, where human judges determined race outcomes. With AI, we can either allow biases to proliferate or use technology to reveal and address them.
**(12:47)**
Fourth, *Loss—the Soul Stealer.* AI’s impact on the workforce is real; jobs will inevitably change or be eliminated. When cars replaced horses, certain jobs—like cleaning the streets of horse manure—disappeared, but new roles, like mechanics, were created. As Jensen Huang from NVIDIA pointed out, AI helps companies become more productive, which can lead to more innovation and job creation. So AI can eliminate some roles, but it also opens the door to new opportunities.
**(15:11)**
Fifth, *Greed—the Resource Hog.* We’re pouring massive investments into AI without always seeing an immediate return. Amazon wasn’t profitable for its first eight years because it focused on building infrastructure for the future. Today, we take Amazon’s services for granted, but that infrastructure investment was once mocked. Even if AI doesn’t revolutionize every aspect of business, it has already yielded benefits like early detection of diseases. Greed might push us to keep investing, but we’re already seeing positive returns from AI.
**(18:38)**
Sixth, *Deception—the Masked Trickster.* AI’s ability to deceive us—by creating realistic, fake content—is perhaps the most concerning risk. Today, anyone can impersonate someone like Elon Musk through synthesized voice and video. There isn’t much of an upside here; it’s a serious issue that requires swift action and regulation to prevent misuse and protect trust.
**(19:21)**
Lastly, *Control—the Final Sin.* People fear that AI will evolve beyond our control. This reminds me of the Y2K scare—people thought the world would shut down at midnight on January 1, 2000. Thanks to preparation and coordination, nothing happened, but at the time, it felt terrifying. We’re now faced with a similar fear of the unknown with AI, and the question is: do we limit it, or do we let it grow unrestricted?
**(21:30)**
So, these are the “Seven Cardinal Sins of AI.” While we in this community may already know some of this, we’re surrounded by colleagues, managers, and others who are asking these questions daily. Each “sin” presents a choice—whether to focus on the risks or the opportunities AI brings. As the singer Pink once said, “Just because it burns doesn’t mean you’re gonna die. You’ve got to get up and try, try, try.” AI presents us with challenges, but also with potential. It’s up to us to try and make the best of it.
**(23:18)**
Thank you, Marco! I love the way you framed this. You’ve added historical context that really brings these points to life.
**(24:07)**
Janet asks, "Is deception the only real concern here?"
Marco: Good question. Yes, I’m most concerned about deception because it’s already challenging to distinguish truth in today’s polarized world. Without regulation, deepfakes and other deceptions could erode public trust and exacerbate misinformation.
**(25:51)**
Jay-Z: Marco, I love how you balance fear with practicality in your delivery—it’s engaging and reassuring. Is there a story behind how you crafted this presentation style?
Marco: Thank you! I wanted a relatable approach. I started with “The Five Stages of Grief” last year because that’s how I personally felt about AI’s impact. This year, I created a new framework to show people that history repeats itself. AI’s challenges are serious, but we can learn from past tech fears, like Y2K, and apply similar solutions.
**(28:22)**
Kent: You mentioned dependence and the importance of critical thinking. With AI, the quality of our output often depends on the quality of our questions, which can promote a more thoughtful interaction compared to social media.
Marco: Absolutely. Education is the key here. In the past, we learned to memorize and regurgitate information, but with AI, the focus is shifting to asking the right questions. Soft skills like leadership, communication, and critical thinking will be our biggest assets in the AI era.
**(29:52)**
Kent: With tools like RAG GPT, could editorial control help mitigate issues like hallucinations?
Marco: Yes, giving users editorial control is a powerful way to prevent AI from veering off track. AI tools that allow user-defined parameters could help ensure quality and accuracy.
**(32:16)**
Jean asks, "Will AI become like the internet, just part of the background?"
Marco: I believe so. In two years, we may stop talking about “using AI” and just take it for granted, like we do with PowerPoint today.
**(33:21)**
David: At Novartis, do you have a Chief AI Officer, or how is AI currently structured?
Marco: AI at Novartis is often managed through IT or corporate strategy, which can make it challenging for business units to fully leverage AI. Ideally, business units should take ownership of AI applications to meet specific needs, as centralized control can be limiting.
**(34:28)**
Kate: I consult in this space, and I find that people often shift from asking about automation to realizing they need a new communication strategy. This is especially relevant when you mention Y2K—it was real and required collaboration and coordination. What should we be doing to prepare for AI’s risks?
Marco: The answer is education. We need to invest in educating everyone, from senior executives to entry-level employees. Singapore, for example, offers subsidies for AI retraining. Companies need to adopt a similar approach by prioritizing skills development as much as tech investments.
**(38:35)**
Gan: Coding is becoming easier with AI’s help. Does that free up resources for more complex problems?
Marco: Absolutely. With the basics handled by AI, people are freer to pursue more creative and complex ideas, particularly in fields like marketing where we can go from idea to prototype very quickly. It's one of the most exciting times to be a marketer.
**(41:48)**
David: So, what do you do at Novartis on a daily basis?
Marco: I train senior executives on AI, focusing on showing its potential and how it can apply to their roles. These trainings often lead to tangible outcomes—executives start asking more challenging questions, sometimes even pushing to develop or purchase the tools they need. The aim is to get people thinking strategically, not just reactively.
**(46:03)**
Jay-Z: In large companies, who is accountable for AI decisions? When AI
tools are used, especially in high-stakes fields like finance, what happens if something goes wrong?
Marco: Accountability should rest with business owners. They benefit from AI’s productivity gains, so they should also be responsible for overseeing its use and ensuring its proper implementation, with IT and compliance in supporting roles.
**(47:55)**
Karan: Do you record these sessions? You must get similar questions repeatedly, so would a knowledge base or GPT trained on your materials help?
Marco: Yes, I often use FAQs and examples from other teams to preempt common concerns, especially for those who might be skeptical. But nothing replaces real-time engagement—seeing people’s reactions and addressing their questions live is invaluable.
**(50:23)**
Kate: Who are your biggest detractors at Novartis?
Marco: There are generally two types—those who fear AI (often IT or compliance) and those who feel they “own” the AI initiative. But I find that skeptics can become strong champions once they see AI’s potential, even if it takes patience and protection from higher-ups to create that space for change.