I go to a lot of conferences. And I've sat through a lot of AI keynotes that follow the same arc: first, a slide showing how much of your job AI can theoretically do. Then, a beat of silence while the audience absorbs that. Then, a pivot to "but don't worry, humans are still essential" — without much evidence for why. The talk ends, people applaud, and walk out more anxious than when they sat down. That is not a useful talk.
I lead the Ipsos Global AI Monitor, which tracks AI attitudes across 32 countries. I also oversee the US Consumer Tracker, which goes into the field every two weeks. I have actual data on this. Let me tell you what it says.
The Numbers Most Speakers Won't Show You
Across 32 countries, 62% of people say AI will change their job in the next five years. Far more agree than disagree. One in four — 25% globally, around 25% in the US too — say AI will replace their job in the next five years. That is a striking number. And it hasn't meaningfully improved year-over-year.
Why do people believe this? Partly because the literal people building these tools keep saying so. Every conference I attended in the last year had two slides in almost every session: one showing how much of various job categories AI could currently do, and one showing how much it could do. The red bar of current capability against the blue bar of theoretical potential. Finance and business professions sit in an uncomfortable spot on that chart.
What almost none of those presentations included was a third slide: okay, but then what? If 25% of the workforce loses their jobs or gets significantly displaced — what's the plan? If you're familiar with the Underpants Gnomes from South Park, you know the problem. Phase 1: collect underpants. Phase 3: profit. Phase 2 is a question mark. The AI version of this is: Phase 1: AI takes your job. Phase 3: somehow, the economy is better. Phase 2 remains a question mark, and I find "universal basic income" an unsatisfying answer in a country where it's deeply culturally anathema.
America Is a Global Outlier
Here's the part that most AI keynotes miss entirely: the United States is not typical of the world in its relationship with AI. Globally, we see a productive tension between wonder and worry — people who are excited about what AI can do for them and nervous about what it might do to them. The US sits in an unusual spot: far more worried than excited, more so than almost any other country in our survey.
Part of the reason is trust — specifically, trust in regulation. Fewer than 6% of Democrats and 12% of Republicans think the government has no role in regulating AI. The bipartisan consensus is that some oversight is needed. Yet the stated policy position of the US government has been, essentially, no regulation for a decade. When you combine widespread anxiety about AI with deep distrust of its governance, you get the uniquely American skepticism that shows up in our data.
We're also starting to see the signals of something beyond passive worry. People are getting sick of AI-generated content — they can increasingly tell, and they don't like it. There is genuine public opposition to data center development. And at the extreme edge, there's been actual targeted anger at AI company leadership. These aren't mainstream phenomena yet. But they are signals, and my job is to notice signals before they become trends.
The Bartender Tab Problem
I want to be direct about something that matters for understanding AI's limitations. One of my colleagues suggested I ask AI systems to predict the results of our survey data. The results were off by 30 percentage points on a question about something as mundane as whether Gen Z runs bar tabs or pays per round. (This was, I'll note, a New York Times trend story that traced back, on click-through, to a tweet from a British bartender and a TikTok reaction video with 20,000 views.)
AI can read the New York Times. AI does not have access to 1,000 real people to ask. There is a category of knowledge — what Americans actually believe, how they actually feel, what they actually do versus what the media says they do — where real research still outperforms AI inference by a significant margin. People say AI provides consistent and repeatable results. Our data shows it doesn't. That gap matters.
It matters because people are now using AI as a tool, not a toy. In our work with Google tracking AI attitudes over three years, we saw a meaningful shift: AI went from something people used to generate cat memes to something they use to research complicated financial decisions and navigate their professional lives. When people use AI to research "how should I invest my money" or "should I fire my financial advisor," they are making consequential decisions based on output that may be significantly miscalibrated to real human behavior.
What Keeps Humans in the Game
To illustrate this point in a talk, I ran an experiment. I asked an AI to act as my financial advisor — I told it I had a Northwestern Mutual advisor, but was thinking about switching. Without much prompting, it laid out a detailed plan for exactly how to do that. It wrote the emails. It suggested fee-free fiduciary alternatives. It outlined an investment strategy. It performed what it called a "sunk cost audit" of my entire financial life. It was thorough and confident.
But when I pushed it on why I shouldn't just let it run my finances entirely, it told me something revealing: it couldn't be my sole financial co-pilot because someone still needed to sign legal documents. And then, unprompted, it said this: if my agent wants to keep my business, he needs to stop being a product salesperson and start being a wealth strategist.
That is not a threat to human advisors. That is a job description. The value of the human in the loop is not just legal necessity — it's the trust that comes from knowing a person is reading the data, questioning the output, and taking responsibility for the recommendation. AI doesn't build that trust. AI depends on humans to build it.
The thing that helps AI trust is keeping humans in it — not laying them off. The companies getting this right are the ones where people understand how AI is being used in their relationship. They assume their advisor uses AI tools. They trust that the advisor is also reading the output critically and making judgment calls. The black box version, where AI is making decisions that affect you and you don't know how, is where trust breaks down.
What a Useful AI Keynote Actually Does
A good AI talk doesn't just show the capability curve and leave people anxious. It shows the capability curve and then explains what we know — from actual research — about where human judgment still outperforms AI inference. It shows where trust is won and lost. It names what the data says about American skepticism versus global norms. And it gives the audience a specific, credible path for what to do next — not vague reassurance that "humans are still important," but a concrete description of how to be important in an AI-augmented environment.
If you want that talk — grounded in Ipsos data, calibrated to your industry, and built to leave your audience with clarity instead of dread — I'm happy to have that conversation.
Why this particular AI keynote
There are a lot of AI speakers. Most of them are drawing from the same pool of published research, the same McKinsey slides, the same OpenAI capability announcements. The reason to book someone who leads the Ipsos Global AI Monitor across 32 countries is that the data hasn't been in a TED talk yet. When I show American AI anxiety compared to every other major economy, that's a number your audience hasn't seen. When I show the shift from AI-as-toy to AI-as-tool over three years of Google-partnered research, that's a trend line, not a snapshot.
I also have an unusual background for this subject. I spent most of my career as a journalist — at Advertising Age, at Crain's Chicago Business, as a music critic. Journalists are professionally skeptical of hype. When I say AI won't replace your job as quickly as the headlines suggest, it's not because I'm trying to make you feel better. It's because I've looked at the data from 32 countries and that's what it shows. And when I say the trust problem is real and consequential, same source.
The goal of a useful AI keynote is to leave the room better calibrated — neither panicked nor complacent — with a clear-eyed picture of where the data actually points and what that means for their specific industry. That's what I build.