Energy

Altman to humanity: ‘Accept some bad things to get benefits of AI’

October 29, 2026

The leader of the most visible company hawking the future of artificial intelligence (AI), Sam Altman, CEO of OpenAI, told Politico a little over a week ago that “the world should accept some bad things happening for the benefits of this technology.” Let me provide additional context. Previously, Altman has said the following:

  1. “It is terrifying because of the power that superintelligent AGI wields to destroy human civilization, intentionally or unintentionally.” — Lex Fridman Podcast #367 (March 25, 2023)
  2. “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” — Center for AI Safety statement (No date)
  3. “I was very intent on not destroying the world before having kids.” — Vanity Fair interview (October 5, 2026)

I do not know for a fact that AI can destroy civilization, and I’m not sure exactly what Sam Altman means when he says that. But I do know one very important thing that bears repeating again and again: Any product or system that has a nonzero chance of destroying all of human civilization should be abandoned. There is no amount of benefit that can justify risking the destruction of the entire human project no matter how small the risk.

This principle, of course, has been regularly violated by human society in the modern industrial era. We have built enough nuclear weapons to destroy us in the resulting nuclear winter. We are engaging in an uncontrolled experiment in the environment with genetically engineered crops. Their failure could have catastrophic effects on food supply. Their unmodeled effects on the environment could lead to unforeseen cascades that could have severe consequences. And, of course, climate change is the manifestation of our uncontrolled experiment with the burning of fossil fuels and with our profound changes in land use, which have unleashed vast amounts of heat-trapping carbon into the atmosphere.

So, who thinks there is a nonzero chance that AI could destroy human civilization? Well, lots of people as is evidenced by the signatories to the Center for AI Safety statement linked above, many of whom work in the AI industry. And Sam Altman uses the word “nonzero” in his recent interview with Vanity Fair regarding the risk of civilizational collapse.

The apologists for technologies that threaten all of civilization typically point to the fact that humans are still around despite those technologies. That’s like saying that you’ve been smoking for 40 years and you never once got cancer. These apologists also like to say that the probability of such a wipe-out is very small.

Here it is essential to understand what happens when you do something that has even a very small chance of producing catastrophe, and you iterate it again and again well into the future. You make it almost certain that the catastrophe will occur because that small chance becomes a big chance if you repeat the steps that could lead to catastrophe repeatedly with no intention of stopping. The world’s nuclear powers are not dismantling all their nuclear weapons. The agricultural seed giants are not putting any limit on the number of genetically engineered crops they will design. And, of course, we are doing very little to avert the catastrophic outcomes of climate change (which could include dramatic declines in crop yields due to drought, heat, and flooding depending on where the crops are grown).

I believe that the current version of AI, based on what are called Large Language Models, has far fewer benefits than the industry is touting, which keeps leaning on what AI supposedly will do, not what it can do now. For a thorough takedown of AI’s supposed benefits, I recommend keeping up with journalist Ed Zitron.

But even if AI were to end up having all the benefits the industry is touting, it simply cannot be justified if continuing to use it will risk the entirety of civilization. Implied in Sam Altman’s so-called risk/benefit analysis is that AI would continue to be used far into the future—until, of course, we destroy ourselves. One analogy I’ve used to understand this risk is Russian roulette:

Now imagine we are playing Russian roulette using a gun with 100 chambers in which only one bullet has been loaded. If we pull the trigger one or twice, there is indeed a very small chance that we will kill ourselves. But, if we place the gun to our heads and pull the trigger 100 times, we will almost certainly perish.

The risk could be one a million or one in a billion or even one in a trillion and the analogy would still hold. And, if you think about the number of operations AI systems around the world can perform, we could bring on disaster rather quickly. Remember: If you keep repeating something that has a nonzero chance of producing catastrophe, you will eventually produce catastrophe. If Sam Altman is right, that is where we must head if we want the so-called benefits of this technology.

Kurt Cobb

Kurt Cobb is a freelance writer and communications consultant who writes frequently about energy and environment. His work has appeared in The Christian Science Monitor, Common Dreams, Le Monde Diplomatique, Oilprice.com, OilVoice, TalkMarkets, Investing.com, Business Insider and many other places. He is the author of an oil-themed novel entitled Prelude and has a widely followed blog called Resource Insights. He is currently a fellow of the Arthur Morgan Institute for Community Solutions.


Tags: Artificial Intelligence (AI)

Forum

Visit the Resilience Community Forum today to join conversations on topics you care about. Share ideas and resources, make connections, and help create momentum on important issues.

Go to Forum

Comments

Subscribe
Notify of
3 Comments
Oldest
Newest Most Voted
mwildfire
21 days ago

And what was that about wanting to have kids before he destroys the world? He wants to destroy his own kids along with everyone else? So he doesn’t want to destroy the world first and THEN have kids? These people are crazy, dangerously so.
I fully agree that aside from the stupidity and recklessness of risking civilization, or human survival, for the dubious benefits of AI, the evidence so far is that those benefits are pretty trivial. LLMs have done many useful things so far, nearly all of which come with downsides bigger than the upsides. For example, it can write your term paper for you, and it might even fool the teacher. But MIT has actually done studies showing what teachers have already observed, what is obvious before the trial–using a crutch to do your mental work means you don’t learn what the mental work was supposed to teach. and using crutches all the time means your brain atrophies, just as your muscles atrophy if you lie on the couch all day.
But I’ve read and heard things suggesting that the techbros may be greatly exaggerating and hyping the risks, because: this is a way of hyping the supposed incredible power of the technology, to pull in more investment; they want to get ahead of the demands for regulation so they can control the regulation, so it doesn’t hamper their companies, does hamper competitive startups, and allows collusion between companies without feat of antitrust enforcement; and frames the situation as “rogue AI agents out of control” to avoid responsibility for the crimes their tech commits.

Juli Thomas
Admin
20 days ago
Reply to  mwildfire

Hi , we have a few people chatting about AI on the Resilience forum, would love to have your input: https://www.resilience.org/community/postid/140/

Steve
20 days ago

My reaction to Altman’s self-contradictory comments can be encapsulated in 2 paragraphs from a post I will add soon to my Substack,”Clear-eyed Resilience”, stevekaagan.substack.com.

In a recent book entitled, If Anyone Builds It, Everyone Dies: Why Superhuman AI Would Kill Us All, authors Eliezer Yudkowsky & Nate Soares outline how an AI superintelligence, if built, would lead to the demise of humanity. The gist of their argument is that a superintelligence will operate beyond human control, refining its own capacities, untethered by its designers.
To support their case, the authors argue that a superintelligence will share the attributes of three difficult-to-control operations – space probes, nuclear reactors and cybersecurity. Destructive effects will arise outside its ken (like space probes) and will occur faster than we can remediate them (like nuclear reactors). And malicious manipulation of systems will nearly always outpace preventative measures (like cybersecurity).