Is there a ‘10%’ chance AI kills us all, or cures all diseases?

I have never seen a business advertise itself so heavily with the “Be careful, we might kill you” motto— but that is certainly the case for AI.

The more AI leaders get the public to believe the technology is so powerful it could wipe out all humans, the more it becomes believable that it can give you medical advice as well as your doctor, write code as well as a techie, or plan your vacation as well as a travel agent. It must be acknowledged that in a world desensitized to advertising— the “we will kill you all… maybe” is a stimulating message to overcome doom scroll, and convey our product is powerful.

“We might kill you” is also a great way to regulate your competitors out of existence before they launch and kill the industry ex-US.

Having said that, I genuinely believe that the people saying ‘AI might kill us’ believe what they are saying. I also think the people who think AI will cure cancer in 5 years also genuinely believe it. I have no doubt of the sincerity of their beliefs, but I have been around long enough to have heard many incorrect prophecies.

Francis Collins promised that by 2015, doctors would use genomics to pick the right blood pressure pill for you. Judah Folkman told us he would cure all cancer— and I was promised CAR-T would already have stopped most solid tumors by now (it currently has 0 solid tumor approvals in the US to date). I remember when bone marrow transplant cured RCC— and many other prophecies.

Will AI kill us all, or is this Y2K?

I suspect that AI will be able to hack into computers that haven’t yet been hacked, and quite possible reek havoc on financial systems and electrical grids. I could change all lights to green and cause car accidents. Total panic will ensue if any of this happens.

Can it kill us all? When people evoke this possibility they often say bioweapon. Just yesterday Anthropic revealed 5 vignettes of folks asking Claude, “Help me make an more potent virus or venom.”

I am someone who thinks studying very deadly diseases, and GOF research, is more likely to do harm than good. (We can’t afford another lab leak) And I am confident that AI could make viruses more deadly or more transmissable. But to really kill us all, AI would have to make a virus more deadly and more transmissible, and release it, and avoid selection pressure which would drive mutants that are less lethal and even more transmissible— and I think that is biologically quite hard.

Some people say imagine COVID but more deadly. But Omicron was precisely the more transmissible, less deadly variant that outcompeted ancestral strains. You have to also postulate— there is no omicron. AI must prevent selection from choosing less lethal variants.

What if AI launches all nuclear weapons? What if it releases small pox? Its hard to know if there is a 10% or 1% or <0.00001% chance of this happening.

And, what are the chances of any of this happening without AI? What are the chances of a terrorist group doing this? I think it is not zero. It is hard to know if I should be more worried about AI, or some fundamentalist zealot group who are indifferent to facts and reason and cling to religious beliefs with fervor— like the woke.

IF AI can solve Navier Stokes— AI can kill us all by making a better virus. And it can cure all cancer too. I am not sure this follows. Biological advancement and mathematical advancement don’t follow the same rules or patterns. I hear that Terence Tao believes the NS solution is brute force, and does not expand on fundamental mathematical principles— similarly, you can already study molecular biology and think of many ways to juice a coronavirus (add a furin cleavage site), but its still quite hard to “kill everyone.”

What about COVID? It is worth remembering that even covid— likely lab engineered—- didn’t kill 99+ % of people. It killed old people, and sick people, and overweight people, and some healthy people (but very few). Ironically life expectancy is partly up right now because covid killed off people with lower than avg life expectance in every age category.

COVID did more damage with the response. Closing schools, covering your face with a rag until your brain rotted, and you demonized folks who disagreed. Political polarization, vaccine mandates— COVID’s response did way more damage.

Instead of nukes or bioweapons, maybe AI would kill us all by getting us to hate each other more with more and more propaganda— like the observational kind published in nature medicine. But even here, propaganda has limits, and killing us all would require us to not stop after some of us are dead.

Is this possible or science fiction? Well, FWIW, I don’t lose sleep over AI. I could be wrong, and AI could kill us all, in which case I would be dead, so it will be hard to rub it in.