Post

It's also based on observations on the ground, my own & those of numerous educators & scholars who are seeing clear signs of these harmful effects on their students, children and peers, unfolding in real time. Still, some major caveats to the analogy are necessary: 4/n

First caveat: the asbestos analogy does not apply to AI or machine learning generally, nor all AI uses. Not all AI/ML is GenAI, and AI does not equal 'chatbot' or even 'LLM'. If the analogy applies, its clearest application is to commercial GenAI chatbots used to automate cognitive tasks. 5/n

Second caveat: the asbestos analogy does not support the view that GenAI is 'useless' or uniformly harmful. Asbestos had many beneficial use cases, which is how it ended up everywhere! It is debatable *how* useful GenAI is, and where, but the idea that it has zero beneficial uses is silly. 6/n

Third caveat: on balance, asbestos ended up being seen to do more harm than good, hence bans on its use and manufacture in the EU, UK and USA. The jury is still out on the total impact of GenAI, but even if its effects are net negative (my current educated guess), the solution is far less clear. 7/n

Fourth and final caveat (for now, I'm sure others will add to the list): Asbestos does irreversible and often fatal structural damage to human lung tissue. Even if GenAI is bad for human brains, it is not the same kind of damage, and it is plausibly reversible (though how easily is less clear). 8/n

OK, then how IS GenAI like 'asbestos for the mind'? First, any damage asbestos does is initially unperceived. You don't feel your lungs being damaged until much later. GenAI actually makes most people feel cognitively stronger at first, even when they are instead becoming cognitively dependent. 9/n

Second, and most obviously, by the time we realized the severity of asbestos' harms, it was already in all the walls. The comparison with GenAI is obvious. It's already in the walls -- of our schools, our governments, our media, our courts -- before we have measured the severity of its harms 10/n

Third, because it's in the walls, so to speak, it's hard to avoid. You often don't know when you're in a house that still has asbestos. It's also increasingly hard to see where GenAI is present in our collective cognitive endeavours. I know I'm not using it, but anyone I write with might be. 11/n

Fourth, the cost of getting GenAI out of the walls will disproportionately harm those with the least resources & power to do so. It was only the wealthiest countries in the world--including those that first mined and marketed asbestos insulation--that stopped using it and paid for remediation. 12/n

Likewise, some elite private schools and universities are already testing 'AI-free' models of education for the wealthy (including children of AI developers!) while cash-strapped public schools go all-in for AI to deliver 'efficiencies' (reduce headcount). Same dynamics in media, health, etc. 13/n

nytimes.comSilicon Valley Executives Are Tech Fans. Just Not for Their Own Kids.At home, moderation and restraint are the rule for tech parents. Boredom is exalted. Maybe they’re onto something.

Likewise, working people like my grandfather, who died from asbestos exposure at work, didn't have a choice about whether to be harmed by it. Many people don't feel they have the power to say 'no' to using GenAI at work or to avoid it for schoolwork, if the cost is being seen as 'unproductive.' 14/n

Fifth: asbestos harms the lungs by stiffening them, making them harder to expand. Eventually it just requires too much physical effort to breathe. We don't yet know exactly how GenAI affects the mind, but early studies suggest it increases the perceived mental effort of thinking without it. 15/n

Again, this may be reversible! But it's a hell of a gamble, and frankly an inexcusable one, that we are making our kids think with it *before we know* how severe is the harm to their cognitive flexibility and motivation, or how to reverse its harmful effects. Nor do we know how to prevent harm. 16/n

In the early days of asbestos mitigation, and still in places like Russia today, people were/are told they'll be fine if they just 'use it correctly.' (sound familiar?) "Just don't breathe in clouds of it! Wear a paper mask!" Oops, sorry, doesn't work, now it's in your lungs. 17/n

With GenAI, we roll our eyes and say 'it's fine, just don't let AI do your thinking for you!' Yet there's no evidence for what that amounts to, and studies on cognitive surrender (Shaw and Nave 2026) make that prophylactic look shaky as hell. You don't always notice when it's thinking for you. 18/n

And while asbestos stiffens the lungs physically, we have reason to worry that GenAI might stiffen our minds intellectually, creatively and morally. AI-generated outputs are, for the most part, variations on the dominant mathematical pattern in the training corpus that is linked to your input. 19/n

Even when their outputs are weird or disturbing, they can be oddly homogenous (see the freaky AI-generated food photos/ads used by local restaurants). I just read an AI-written manuscript that had consistently beautiful prose. By the end it felt like listening to one song on repeat for a week. 20/n

Many now notice people they know starting to sound--in person--like chatbots, speaking and reasoning in AI's stiffened tropes. I warned in The AI Mirror against becoming entranced and captivated by these mirror reflections of our aggregated minds. Worse is if we also begin to resemble them. 21/n

OK, last point, because I have a job to do that is not Internet. There is one place where the analogy really breaks down. In the US, when the true cost of asbestos became clear, we didn't ask the manufacturers to save us. We didn't invite them to summits. We sued them into oblivion. fin. 22/22

This may yet happen though. It took a long time for the asbestos hazard to be sufficiently "clear" to enough people. We're still in that period wrt to genAI. Thanks for the thread: another small step towards sufficiently widespread clarity!

asbestos is such a depressing case, indeed: "Looking back in the light of present knowledge, it is impossible not to feel that opportunities for discovery and prevention of asbestos disease were badly missed." Thomas Legge, ex Chief Medical Inspector of Factories, in Industrial maladies, 1934

useful on this point with case studies: "Late lessons from early warnings: the precautionary principle 1896-2000" www.eea.europa.eu/en/analysis/...

10 reposts28 likes

Yeah, if you haven't yet, I DON'T recommend looking up Wittenoom where iconic pictures of children playing in piles of the stuff were taken. Hancock mining who owned the site is also where Gina Rinehart, the Australian billionaire who funds a far-right Trumpian campaign here, also got her money.