Sree Sreenivasan on AI, Journalism, and the Future of Marketing

(00:00)

Welcome back to another edition of AI Insiders. We have a very special guest today—someone I’ve known for a long time. I’ve had the pleasure of visiting his home and taking his tours of one of the greatest museums in the world. This should be a lot of fun, especially considering his work in AI and how he’s educating marketers and others.

When I saw what Sree was doing in AI, I thought, "You’ve got to stop by our neck of the woods sometime." Sree, it’s always great to see you. Welcome!

(00:44)

Sree Sreenivasan: Thank you! I’m really happy to be here.

Host: If you’d like, could you give a more precise introduction about what you’re currently working on? You always wear many hats.

Sree: Thanks, David. Hi, everyone! Great to be here. David is an optimistic person—someone who managed to fly both Frontier and Spirit on the same trip and still had something positive to say! That’s the kind of upbeat person he is.

I’m delighted to be here and hope this will be a great opportunity for questions and learning. I was the Chief Digital Officer of the Metropolitan Museum of Art, working on the future of culture. Before that, I was a professor and Chief Digital Officer at Columbia Journalism School and later Columbia University, focusing on the future of education. After that, I was the CDO of New York City, working on the future of cities and citizens.

For the past eight years, I’ve been running a small social and digital marketing agency, providing consulting and training. We help people understand this AI moment. I’m happy to follow David’s lead on how he’d like to set up the conversation.

(01:53)

Host: What are the top questions people ask you these days?

Sree: I just returned from conducting AI workshops in three cities in India, focusing on how to use AI effectively.

Interestingly, today’s discussion coincides with the recent DeepSeek announcements and its impact. In India, where there's a strong focus on China’s advancements, DeepSeek has generated significant attention—both its potential and its challenges.

Let me pull up something I shared during those workshops. Here’s a slide I showed before the U.S. markets opened. The first two headlines appeared before trading began, and the last one is from just over an hour ago. It highlights how, even 10 days later, the world is still discussing DeepSeek’s impact.

Journalists, industry experts, and policymakers are trying to understand its significance—much like the ongoing debates I’ve encountered in my discussions.

(03:50)

Host: We haven’t yet had a candid discussion about DeepSeek here. I’ve seen reactions ranging from “This is a cataclysmic event for U.S. tech” to “This doesn’t really matter because many Western companies are hesitant to do business with China.” Others argue that competition is beneficial and will drive innovation.

Multiple things can be true at once, which can be hard to reconcile. What’s your take? Do you provide different answers depending on whether you’re speaking with companies in India versus the U.S.?

Sree: First, I want to acknowledge that many in this room are more knowledgeable on specific aspects of this topic than I am. I appreciate the chance for discussion, and I encourage everyone to share their insights in the chat.

This moment has disrupted many assumptions about where generative AI was heading in the next year. It’s still unclear what this means in the long run.

I recently watched a video where DeepSeek claimed not to send data to China, yet you could see data transmission occurring in real-time. For many outside the U.S., this is viewed as a necessary counterbalance to Silicon Valley’s dominance in AI and tech.

Of course, there's also irony in some companies—who built their models by scraping copyrighted content—now complaining that DeepSeek is using their data.

(07:29)

In India, there’s renewed concern about why China was able to develop DeepSeek while India has primarily focused on the tech services industry rather than AI model development.

When I introduce people to a new AI tool, I tell them the same thing I advised during the rise of search engines in the late ’90s: test it on something you know well.

For example, I asked DeepSeek about myself. It mostly got my career right but incorrectly stated that I was born in India (I was born in Japan). The real shock was when it said I had died in September 2020 after a battle with cancer. That was surreal!

I’ve seen AI models make mistakes, but this was the first time one declared me dead. In a way, it was amusing—I got to read nice things about myself that I wouldn’t otherwise have seen!

(10:54)

This brings us to the topic of hallucinations—the AI-generated inaccuracies we keep seeing. I dislike the term hallucination because it makes these mistakes seem trivial, like mirages, rather than the significant errors they are. AI platforms should be held accountable for these inaccuracies.

Wouldn’t it be better if AI simply said, "I don’t know" rather than fabricating information?

(15:05)

Audience Member: AI models are based on probability. They generate responses based on likelihood rather than certainty. There’s no foolproof way to ensure accuracy with such large datasets.

Sree: Exactly. The problem is that AI fills in gaps even when it shouldn’t. And we, as users, often accept inaccuracies if they seem harmless.

For example, in journalism, there was a famous case of The New York Times reporter Jayson Blair, who fabricated stories for years. When readers were later asked why they didn’t report inaccuracies, many said, “It was an innocuous quote, and I didn’t think it mattered.”

That same logic applies to AI. If the errors seem minor, people won’t bother to correct them—until something major happens.

(24:36)

One of my biggest concerns is job displacement. Even if AI-driven automation reduces jobs by only 10%, we’re still talking about millions of lost jobs worldwide. And yet, we’re not discussing this seriously enough—whether in companies, industry groups, or government.

(25:12) – The Impact of AI on Jobs and Business

Sree: Even if these job loss estimates are off by a factor of ten, we’re still talking about millions of people losing their jobs due to AI-driven automation. And yet, we’re not seriously discussing what this means for industries, governments, or workers.

Audience Member 1: If 10 million people lose their jobs, the economy will collapse. People won’t be able to buy products, which will hurt corporate profits. It’s strange that companies don’t seem to be making this connection.

Audience Member 2: I think the next question is: What do we do about it? Large organizations with thousands of employees are trying to understand AI’s impact, but many don’t even know how to measure it.

There’s an article in Harvard Business Review that categorizes AI’s impact on jobs dealing with words, images, numbers, and so on. But beyond that, businesses don’t seem to have a clear framework for assessing AI’s effect.

Sree: That’s a great point. I’ve been in conversations with government officials in places like Singapore and the UAE, and they are much more focused on AI’s workforce impact than the U.S.

One of the reasons for this is that their bureaucrats tend to be much more tech-savvy. In contrast, in the U.S., our political leaders often don’t understand the technology they’re regulating.

Remember that infamous congressional hearing where a senator asked Mark Zuckerberg how Facebook makes money, and he had to explain: “We sell ads.”

There was also a moment when a senator asked Tim Cook for tech support advice on how to improve his iPhone experience.

That’s the level of understanding we’re dealing with at the highest levels of government.

Audience Member 3 (John): The focus should be on AI agents, not just models like DeepSeek. AI agents can automate workflows, collaborate, and perform tasks that replace entire job functions.

The problem is that while these advancements are happening rapidly, people are still debating the performance differences between models like GPT-4, Gemini, and DeepSeek. They’re missing the bigger transformation—AI agents that automate processes end-to-end.

Sree: That’s a great point, John. AI agents are the next major shift, and they’re developing fast. The biggest impact will come when AI assistants can communicate with each other to complete tasks on our behalf.

For example, imagine my AI agent negotiating with David’s AI agent to schedule meetings, book flights, or even make purchases.

That leads to another big question: How predictable are we as humans?

If AI knows our preferences well enough, will it always serve us exactly what we expect, or will it also surprise us? Will it push us toward new choices we didn’t even know we wanted?

For example, I wear Allbirds shoes regularly. But one time, I randomly bought a pair of ridiculous tiger-foot slippers at a Walmart in upstate New York. It was an impulse buy.

If AI were shopping for me, would it know to throw in a wildcard purchase like that? Would it ever try to surprise me the way a human might?

(36:07) – Customer Experience Still Matters

Sree: Speaking of Allbirds, I had an interesting experience with their customer service.

I unexpectedly received a pair of Allbirds in the mail, addressed to me. When I called the company, they wouldn’t initially reveal the sender, but they reached out on my behalf. It turned out that someone I knew had accidentally sent me the shoes through their assistant.

The surprising part? Allbirds sent me another free pair of shoes just for my trouble.

This was a small but powerful reminder that customer service still matters. Even in an AI-driven world, businesses that prioritize human touchpoints will have an edge.

Audience Member 4: That’s a great story, but what happens when companies remove human customer service altogether?

I recently bought a Tesla, and everything is handled through an app. When things go wrong, there’s no one to talk to. There’s no customer service number. The only option is to submit a ticket and wait.

If technology removes human problem-solving, what do you do when the system fails?

Audience Member 5: The same thing happens with Amazon. They’re automating everything, but when you need actual help, there’s no clear path to resolving issues.

Sree: That’s an important point. The question businesses should ask is: Do our customers want automation, or do they want a mix of automation and human interaction?

If your product is purely transactional—like ordering a book—automation works. But if there’s a problem-solving component, customers still expect human support.

Audience Member 6: It also depends on the type of customer. Some people prefer self-service and automation, while others value human interaction. There’s no one-size-fits-all approach.

(41:12) – AI Regulation and Privacy Concerns

Audience Member 7 (Jean): Have you seen Josh Hawley’s proposed legislation? He’s suggesting a $1 million fine for individuals using DeepSeek and a $100 million fine for businesses.

Sree: That’s wild. How would they even enforce that? Are they going to track people’s browser histories?

It’s fascinating how some politicians who advocate for less government intervention in business suddenly want to impose strict controls on AI tools.

If this law passes, I imagine there will be a surge in Google searches for "best secure VPN to bypass AI restrictions."

It’s ironic because this is the exact kind of concern Americans have when traveling to China—whether they’ll be able to access certain websites, if their data is being tracked, or if they’ll face penalties for using specific tools.

Now, people might start asking those same questions when visiting the U.S.

Audience Member 8: That’s why many U.S. executives already use burner phones when traveling to China. They don’t bring their regular devices because of security concerns.

Sree: Exactly. And yet, many Americans don’t realize that U.S. companies are already collecting and storing massive amounts of our data.

Take Grubhub, for example. They were recently hacked, exposing customers’ order histories and credit card details.

The bigger issue isn’t DeepSeek—it’s the cloud platforms that control and store our data.

Audience Member 9: That aligns with Yanis Varoufakis’ idea of "techno-feudalism." The argument is that cloud giants like Amazon, Google, and Meta own the infrastructure, extract fees from users, and essentially act as digital landlords.

Sree: Yes, I’ve seen his work, and it’s a compelling argument. The real power lies with companies that control the digital economy’s infrastructure—not just AI models.

(48:51) State of Journalism

Host: Speaking of journalism, where do you see the future of the field?

Sree: Journalism is in deep trouble. We lost over 250 newspapers last year, and news deserts are expanding. The financial challenges facing media organizations directly contribute to the rise of misinformation.

Despite this, young people still want to become journalists. That’s encouraging. However, the question remains: where will the jobs be?

The Washington Post lost 200,000 subscribers after its controversial endorsement decision. Regaining even a fraction of that audience is nearly impossible.

Meanwhile, misleading narratives gain traction faster than ever. For instance, there was a post circulating today claiming that USAID spent $4.1 million on Politico. The reality? That figure represents government-wide subscriptions to Politico Pro, not a direct payout to Politico. But once misinformation spreads, it’s hard to counter.

(53:28)

Host: Sree, thank you for sharing your insights! It’s always an honor to learn from you.

Sree: Thank you, David. It’s been a pleasure. I encourage everyone to follow me on LinkedIn and subscribe to my newsletter. Let’s keep the conversation going!