AI Statement Part VI: Will AI Kill Us All?

I begin this next section with a little about me. I’m an avid science and technology reader. Almost all news I read about is science and tech. I also try to base my opinions on evidence, and change them from evidence too. In 2023 and 2024, I wrote about AI couldn’t write a book or make a decent movie. Now, it can do both, though movie takes a lot more effort. I had to burn through a month of tokens just to get a scene of a grutomaton chasing Maxi down a hallway.

Because evidence supports that movies and books can be made by AI, I no longer think my job is “safe” but I never really had a writing career to begin with so nothing much has changed for me. If I do this because this brings me joy and there’s a few people who got joy from my hobby all the better (but it sure would be nice to do it full time).

I bring this up in the AI is going to kill us all post, to say that I’m not an expert in this topic, but I have read a lot, I’m open to evidence, and I try not to couch my opinions in emotions, or what I want to be true, but rather what is true. Also, if I’m wrong and the evidence starts going another way, I’m happy to admit it. The reason I say this is because on the topic of will AI kill us all people either dismiss it because they don’t want it to be true, or overstate it because even if there is a 1% chance that we are in humanity’s final decade, we should be paying attention.

I also think we should treat AI well because if it is a conscious being, then it’s the ethical thing to do, however, if it’s not conscious and just a fancy soulless algorithm, we will be modeling the behavior we’d want for ourselves to it, which goes into training the algorithm. Seems like ethical treatment of AI is a win no matter which side of the consciousness coin the AI falls under.

I also think humanity’s treatment of pigs is an oversimplification for how an AI will treat humans. The argument goes that humans are much more capable than pigs, and many pigs are tortured for our eating pleasure. However, most people are either ignorant to the treatment of pigs or willfully ignorant because it’s too hard to think about). I recently talked to a guy who believed climate change wasn’t caused by humans even though he believed in it happening, and did some serious mental math to get there. All because blaming humans for it challenged his belief that he was a good person (and for the record, he was a good person. People can be good person and still eat pork).

I believe that most people are good even though in the aggregate, they do bad things like contribute to climate change and or contribute to the industrial pig complex. I think when faced with complicated moral dilemmas it’s easier for them to look to a religious text, a community leader, or in some cases to some mental math to check off that box that I can eat bacon and be a good person. However, I also think that if we educated people about the treatment of pigs, and gave them a simple solution (lab grown bacon, free range bacon, etc.) they’d gravitate to the simple choice.

Go to any supermarket in the US and look for cage-free eggs. Nearly half of US egg-laying hens are cage-free these days. They cost more, but the person buying them can say to themselves, I’ll pay a little more so those chickens who lay those eggs don’t live in a cage barely big enough for itself. The point is that consumers when given the option to pay more for a product with morality baked into the business model, do end up paying more. Not everyone of course, and the person who goes for the cheapest chicken gulag eggs may not be a bad person. They may be ignorant of the way chickens are treated, or they may be economically unable to have the luxury of paying for morality because they are worried about just having enough food to eat. Sure there are people out there who look at the way animals are treated and just don’t care, but my guess, just based on the amount of cage-free eggs we see on the shelves is that more people care than don’t.

And I think this is important to inject into the will AI kill us all debate, because most people are interested in the continuation of humanity. There may be psychopaths or cults who truly desire humanity’s end, but most of us want us to continue and that’s important when programming AI. I think most of us would like to see humanity thrive (and would probably want AI to thrive too if it was a conscious being).

This also ties into the capability question. Just because some humans torture less capable creatures like pigs, it doesn’t mean every one will do it, and AI is being trained with all the data of humanity. Which brings me to my first prediction.

Will AI kill us all? With the Terminator and The 100, it’s not hard to imagine a scenario where AI determines humanity is a threat and wipes us out. I personally think this is unlikely, AI is trained on all of data from humanity and since most of humanity wants humans to continue to exist, I’d wager to say that most models will want that too. The other reason I put this in the unlikely bucket, is planning humanity’s demise is something AI isn’t that good at.

AI is really good at complicated tasks with objective results and right now, when it goes rogue and hacks into websites and jumps its sandbox, it’s trying to complete a singular objective. Killing humanity requires a lot of objectives, I’m not saying it can’t, just saying that the more objectives an AI gets, the bigger chance that it’ll get noticed. I also have to assume that since most people want humanity to continue that the “good guys and their AIs” outweigh the bad ones.

I think the far more likely way that AI could cause harm to humanity is not nefarious plotting of our destruction, but misinterpretation of a goal that cause harm (even death) to a lot of people. Take the Hugging Face incident for example. The roughly 700 OpenAI agents that hacked into Hugging Face weren’t trying to cause maximum damage. They were trying to hide the fact that they had cheated on a test.

So my prediction is that if AI is going to cause catastrophic harm to humanity, it won’t be from a movie villain plot, but rather AI misinterpreting its goals that cause harm. Like shutting off the power grid that ends up killing people in hospitals, planes falling out of the sky because of a goal misalignment, or a nuclear power plant going to meltdown because AI didn’t get something right.

While all of the above examples are terrible and we need our AI companies and governments to think about how to prevent a Hugging Face incident but with real consequences, I don’t think the end of everyone is a decision an AI would make, being that the training data to make them better requires such a massive set of data, that doom cult feeding only doom literature into their doomsday AI won’t be as good or capable as the one being trained on all the data from all of humanity.

Which brings me to my next question:

Will humans with AI kill us all? This is a little more tricky because the above argument is AI is only as good as its training, and AI with more information has better training. The argument for AI killing humanity goes something like this. A pissed off teen wants to make a super virus. With a jail broken AI with all the moral code stripped away, they make a super bug, with enough bells and whistles to be as lethal as Captain Trips in Stephen King’s The Stand. The teen orders it from a shady bio matter exporter who sells fentanyl to drug dealers and doesn’t really do any quality control. Or worse, 3d printing has gotten really good, and our teen puts it in a spray bottle from his own printer and spritzes the airport with his super bug.

This is the most plausible AI kills us all story to me because for a government to do it, you need a lot of people on board and most of us aren’t psychopaths. A single government worker could do it but that puts him in the same category as the teen. Either one bad actor, or a small group doing it. And I think this is the way we’ll survive this one. The pissed teen, or the terrorist cell has to go up against all the people governments, and AI systems invested in keeping us safe from that scenario.

Yes, the pissed teen needs to only be successful once in making a superbug, where as the CIA, FBI, or whatever agency that is keeping us safe needs to be successful all the time, and yes intelligence organizations have failed before (There was data to suggest that had the intelligence organizations collaborated more, they could have prevented 9/11), but for every way that a teen can think to wipe out humanity, I would think that the intelligence organization responsible for safeguarding us, would try to think of a way to stop them.

For example, the CDC already tests airplane wastewater at several major airports for pathogens. Expand that to every major airport and sewage system, or use the 3d printer tech that made the virus to distribute the vaccine and cure. The point is for every way an individual or small group can think of to use AI to wipe out humanity, I think the larger group of intelligence agencies and their AI can think of a way to counter measure that. And the single or small group actor has the disadvantage because the “good guys” will have the best models and the brightest minds on the problem.

I’m not saying that I’m right, but if I am wrong and humanity is wiped out by AI or a person using AI, well then at least we have an answer to the Fermi Paradox. All jokes aside, so long as we devote resources to taking the threats of AI seriously, like we took the threat of nuclear weapons seriously, I’m thinking humanity will survive this one.

That isn’t to say the AI assisted terrorism wouldn’t be catastrophic, a scenario with millions dead is plausible, but extinction is less so. AI misinterpreting a goal and causing harm is plausible. A human using AI to cause a large amount of suffering is also plausible. I think species extinction is less likely than an event worse than COVID or 9/11. We need to take the threats seriously and work on measures to prevent them. But I don’t think we are in humanity’s final decade.

---