In my role as a volunteer for PauseAI, I regularly talk to college students and random people on the street about AI risk and how we might mitigate it. The #1 objection I receive is that AI won’t be very powerful. I think we have underinvested in effectively communicating the capabilities of existing AI systems.
(The #2 objection is that there is nothing we can do about it. These are also the two most common objections I see in YouTube comments on AI-risk-related videos.)
Additionally (very frequently online and only rarely in person), I hear the claim that everyone who claims AI could be existentially dangerous (and who would otherwise be credible) are lying in order to hype AI companies and make money. I don’t have any data (even anecdotal) about whether providing very strong evidence against this claim tends to change the interlocutor’s mind. However, I do think that most people who make this claim are not engaging in conspiratorial thinking, and are rather merely ignorant of facts that if accepted would force them to either change their mind or knowingly accept a conspiracy theory.[1]
At the same time, I should point out that I have noticed a vibe change over the course of this year, such that people tend to be more willing today to engage with x-risk ideas without dismissing them outright. X-risk has had some time to enter into the mainstream via various news stories, and more people have interacted with AI tools that impressed them. (This shift is relative rather than absolute, so YMMV.)
For many people it may be useful to present the facts of current LLM performance in math and coding competitions and on benchmarks with held-out test sets, but the general public has very little concept of what those things mean. (“Of course it can do math and write code; it’s a computer!”) The things I expect to be more compelling are things like: “AI has designed and created a working biological virus from scratch”[2] and “publicly available chatbots from last year outperform almost all expert virologists in their own subspecialties at troubleshooting wet lab steps relevant to creating a bioweapon.”[3] Let me know your thoughts, and if you are aware of other pieces of evidence that are simultaneously shocking, scary, and easy to communicate.
Facts like: “about half of all published AI researchers say there’s a significant chance of human extinction from AI this century”, “the 2 most cited scientists in the world helped invent modern AI, and they are very worried about this”, “independent and academic AI researchers are also concerned about this risk, along with industry insiders”, “whistleblowers have risked millions of dollars to warn the public about this,” and “all the CEOs of the top 4 AI companies talked about extinction risk from AI years before any of those companies existed.”
In my role as a volunteer for PauseAI, I regularly talk to college students and random people on the street about AI risk and how we might mitigate it. The #1 objection I receive is that AI won’t be very powerful. I think we have underinvested in effectively communicating the capabilities of existing AI systems.
(The #2 objection is that there is nothing we can do about it. These are also the two most common objections I see in YouTube comments on AI-risk-related videos.)
Additionally (very frequently online and only rarely in person), I hear the claim that everyone who claims AI could be existentially dangerous (and who would otherwise be credible) are lying in order to hype AI companies and make money. I don’t have any data (even anecdotal) about whether providing very strong evidence against this claim tends to change the interlocutor’s mind. However, I do think that most people who make this claim are not engaging in conspiratorial thinking, and are rather merely ignorant of facts that if accepted would force them to either change their mind or knowingly accept a conspiracy theory.[1]
At the same time, I should point out that I have noticed a vibe change over the course of this year, such that people tend to be more willing today to engage with x-risk ideas without dismissing them outright. X-risk has had some time to enter into the mainstream via various news stories, and more people have interacted with AI tools that impressed them. (This shift is relative rather than absolute, so YMMV.)
For many people it may be useful to present the facts of current LLM performance in math and coding competitions and on benchmarks with held-out test sets, but the general public has very little concept of what those things mean. (“Of course it can do math and write code; it’s a computer!”) The things I expect to be more compelling are things like: “AI has designed and created a working biological virus from scratch”[2] and “publicly available chatbots from last year outperform almost all expert virologists in their own subspecialties at troubleshooting wet lab steps relevant to creating a bioweapon.”[3] Let me know your thoughts, and if you are aware of other pieces of evidence that are simultaneously shocking, scary, and easy to communicate.
Facts like: “about half of all published AI researchers say there’s a significant chance of human extinction from AI this century”, “the 2 most cited scientists in the world helped invent modern AI, and they are very worried about this”, “independent and academic AI researchers are also concerned about this risk, along with industry insiders”, “whistleblowers have risked millions of dollars to warn the public about this,” and “all the CEOs of the top 4 AI companies talked about extinction risk from AI years before any of those companies existed.”
https://www.washingtonpost.com/opinions/2025/09/25/artificial-intelligence-advance-virus-created/
https://www.virologytest.ai/