What's the big idea?
Why refusing to use AI won't stop the race, and why AI companies hire philosophers
In this issue:
If everyone would just stop using AI... - International coordination is more feasible, and AI companies don’t need you to buy their AI products directly
Why AI companies hire philosophers - The money is big, and the reasons are complicated
Dispatches from Mitch
If everyone would just stop using AI...
International coordination is more feasible, and AI companies don’t need you to buy their AI products directly
_02.jpg){kind=link}
Here at AI StopWatch, we don’t try to pretend that today’s AIs aren’t useful. As a rule, we avoid pretending anything at all.
It is the AIs clever enough to outmaneuver humanity that we have to worry about; if we build those with anything like the approaches in use today, we’re probably toast.
In contrast, the strongest AIs already released, like Claude Mythos and Fable, are unlikely to be our undoing — though they create their own risks and complications that must be dealt with.
I imagine this stance could be confusing to readers who think everyone should keep their distance from the technology entirely and so deprive the AI companies of revenue. I respect that position and anyone living by it. Unfortunately, it’s an “if everyone would just” position, and as the meme goes, history isn’t exactly full of stories where “everyone just” ended a chronic problem by independently acting against their short-term interests.
The AI problem is in the class of problems we have to solve through structural changes: laws, agreements, penalties. As a rule, companies do not “just” decide not to dump toxic waste into rivers; it took laws and lawsuits to change that behavior. As a rule, people do not “just” fund public services out of the kindness of their hearts; they set up a system where we all pay taxes instead.
So I think a durable international treaty banning the creation of superintelligence is a lot more achievable than convincing everyone to “just” not use AI — just as during the Cold War, treaties limiting nuclear stockpiles proved more achievable than every country “just” deciding not to deploy as many warheads as they could build.
Also, I don’t think AI companies would stop racing even if everyone who didn’t work for them refused to use AI. I think the companies would instead use their AIs in-house to create goods, services, and discoveries they would sell to you indirectly. We see signs companies are already moving in this direction in response to the recent banning and unbanning of Mythos and Fable — events that indicate companies may face White House obstacles to future releases. A headline last week announced that Anthropic is starting its own drug discovery program. I’m willing to bet the company’s executives decided that bringing drug researchers in-house was easier than perfecting Claude’s guardrails to ensure friendly biologists could continue to make discoveries while refusing the prompts of would-be bioterrorists.
So I don’t think you should feel guilty about using the AI tools while they’re still tools. If this were 1998, and you were concerned about what the internet would lead to (say, social media), I wouldn’t tell you to not use the internet. The internet itself isn’t the problem, and abstaining from it would hobble your work.
In the spirit of not pretending, I acknowledge the intellectual property concerns around AI training. I think these need to be settled by the courts, and I hope to see creators compensated appropriately, where practical.
I also acknowledge the resource consumption of AI. But I think we should always place this in context and compare it to the alternative. Everything we do in life consumes resources, and the best numbers out there suggest AI generally does the work we ask it to do using fewer resources than a human would need to do the same. That’s not because AIs are more energy efficient than brains — they’re not — but because humans aren’t just brains. Keeping us fed, clothed, housed, transported, climate controlled... it adds up. I don’t find resource arguments against AI compelling right now, at least not for tasks that are worth doing at all. If you want to cut down on your resource footprint, AI is probably not one of your lower-hanging fruits.
Finally, I acknowledge that in using the latest AI models for the non-writing, non-image, non-podcast-recording parts of our work, we are at risk of being “seduced” by the technology, unable to oppose it with a whole heart. To this, I personally plead guilty, because I will not pretend that the technology isn’t fascinating and full of potential. It’s useful now. If or when a way is found to ensure that superintelligence promotes human flourishing rather than human extinction or oppression, I’ll be the first to cheer it on.
Ultimately, I think my attachment cuts both ways for my work. It might be the reason I’m able to sift through dozens of AI articles and podcasts day after day, week after week. And it means you know I’m not opposing the race to superintelligence out of some generalized dislike of technology.
But yes, it means I always feel just a little bit sick to my stomach when I call for a halt. I imagine it’s like asking animal control to take your beloved family alligator away before it gets too big. It hurts, but you do it anyway, because you know how that story ends.
Not that I’ve ever had an alligator, or ever would. I’m not really a pets guy.
So you see, I can pretend. But on AI StopWatch, I try very hard not to.
Why AI companies hire philosophers
The money is big, and the reasons are complicated
You know we’re out of our depth when we bring in the philosophers.
That’s kind of the field’s whole schtick. It’s only philosophy until we have a handle on it, at which point we call it something else: mathematics, physics, economics... Even computer science was once philosophy, growing out of formal logic, a key part of the philosopher’s toolkit.
The New York Times ran a piece by Benjamin Wallace today called “The Revenge of the Philosophy Majors,” about the work philosophers are doing at AI companies.
Except that this piece ran in the Business section, so it’s even more about how much they’re getting paid to do it — up to $429,000, in the cases Wallace could confirm — and about how they got to be in that position.
As a practical matter, everyone quoted here still cautions against a career plan of specializing only in philosophy. The thinkers making the big bucks have had one foot in the AI camp for a while, exploring questions of machine intelligence and its implications.
How should we feel about their work?
Well, readers of the article might reasonably wonder why the companies would seek answers to questions about topics like AI consciousness and suffering when the answers could prove very inconvenient to their business models. It’s complicated, and not very clearly spelled out in the article. So here’s my best sense of the situation:
Many of the leaders at AI companies, especially Anthropic, have real ethical concerns about their work. They are probably looking into the welfare of their models at least in part out of an actual sense of moral obligation.
Many top figures at these same AI companies are also concerned about the risks of superhuman AIs pursuing interests not aligned with our own. They are looking to philosophers for ideas on how to approach the problem that aren’t just “stop making smarter AIs.” Most famously, part of Claude’s training involves a virtue-ethics themed Constitution written principally by Anthropic’s top philosopher, Amanda Askell.
Philosophers spend a lot of time thinking about thinking. Companies making thinking machines also spend a lot of time thinking about thinking. Philosophers can gain insight playing with these new kinds of minds, and those insights can potentially translate into performance improvements for said minds.
So there’s really no easy narrative I can give you about how philosophers taking AI money are selling their souls or fighting a good fight. Depending on the situation, it could be one or the other, or a bit of both.
But reading pieces like this one, my impression of the philosophers as people is almost always positive. It makes me wish the AI companies were coming at this new science of AI in the traditional way — philosophy first — instead of racing at full speed with a few philosophers strapped in for the ride.
The analyses and opinions expressed on AI StopWatch reflect the views of the individual contributors and the sources they cover, and should not be taken as official positions of the Machine Intelligence Research Institute.