We Should be Optimists

It’s been a while since I’ve written on anything this existential. I think the time is right, though, to revisit it. This week my dad texted me, asking about the near term risks of AI.

My gut reaction was too flee. I spent several weeks two summers ago swallowed by dread. I was consumed by the thought that I, more likely than not, would be dead in ten years. So I decided to push these thoughts to the horizon. I chose to rationalize my ignorance, and abstracted any perceived danger into folklore.

I mimicked contribution by researching alignment. But this of course, was just another roundabout attempt to compartmentalize. And so, I have lived for a while now dissociated from this topic. I think that is exactly the problem. And I also think that as someone who at least knows a bit about it, I have an obligation to speak.

First, I do not think that this sudden resurgence in discussion around the AI threat changes anything. We are exactly where we were two years ago. Two things explain the resurgence.

One reason is what’s called the overton window. The overton window is the range of ideas that exists in the mainstream at a point in time. Two years ago, the idea of AI as an existential threat sat squarely outside of this window. Only people with way too much time on their hands, like myself, had any reason to be concerned. So, the uptick in discourse on this issue is more likely a product of a shifting window, than a different threat. In other words, the threat of AI has always been as real as it is now, it is just that previously the general public did not recognize it as such.

The second reason is that we now have concrete examples of this threat. Incidents like the Hugging Face breach demonstrated the same kind of threat AI researchers have been discussing for years. Crucially, my point here is that we do not have a new, previously unknown danger, but the same one in clearer terms.

Both of these reasons get at a similar idea. We are seeing a shift in societal recognition around the danger of AI. The overton window describes that shift, and the examples describe how it happens.

The broader idea here is that there is a gulf between that which is real, and that which is perceived as real. Unfortunately, humans compute in a way where we only act on that which is perceived as real. And so, it seems that action would only follow from perceiving something as real. This is good news for us, and part of the way I segway this paper into a more prescriptive piece.

I think that AI is an existential threat. I also think that it is not the first existential threat humanity has faced. Nuclear war and terrorism (in a more narrow sense of existential) both come to mind. I was not alive during the cold war, but I would guess that it shares some taxonomy to what we are seeing now.

For instance, I would assume that before the general public felt threatened by the notion of nuclear fallout, a smaller subset of highly educated people felt it. I would also assume that over time, and through concrete events related to nuclear weapons, this threat made its way into the mainstream. Furthermore, I would wager that once it existed in the mainstream, the threat felt more real, and thus prompted more actual by those with the power to do so.

It is true that the threat of nuclear war eventually dissolved with a bit of luck. Namely, the economic collapse of the USSR. Yet, through this analogy, perhaps we can begin to construct a more optimistic outlook on our current scenario.

For one, we avoided catastrophe. For another, we learned from it, which means now we have a reference point through which we can approach the next existential threat: in our case AI.

There are also dimensions of the cold war conflict that give us other reasons to be optimistic about our current endeavor. One is that here, there are no winners. In a very oversimplified sense, the cold war pitted humans against other humans. Now, it is humans against non-humans. I think it is at least reasonable to assume that we have, minimally, a better chance at coming together against an entirely non-human threat.

Another reason for optimism is that it showed us that public sentiment matters. This happens to map on very well to our current moment in time. That is, as public fear of nuclear war grew, so too did political pressure. Fortunately, I think that this relationship between fear and political pressure remains true today. But fear entails something else, and this brings us back to our earlier point.

Fear requires perceived reality. You cannot be afraid of something you don’t think is real. It is only once the reality sets in, that you might allow it to make you afraid. And when you are afraid of something, in this case a threat to your very existence, usually you are inclined to do something about it.

There is a more technical concept for this that might give us yet another reason for optimism. It is called the 1% doctrine. The rule emerged from the war on terror. Put simply, if there is even a 1% chance of some existential catastrophe, we must act as if it is a certainty, and thus do everything in our power to prevent it.

One might argue that if we still recognized this doctrine, we would already have acted on AI. But I do not think the sentiment of the 1% doctrine has been forgotten. Instead, I think it is just that at the time of its adoption, terrorism was a very real, and very pervasive threat. There was no doubt in the minds of the American people that we ought to act if a threat of 1% emerged. And this certainty was primarily a function of the reality of their circumstances. That is, 9/11 gave the American people a very real reason to care about terrorism. One could not see the towers fall, and turn a blind eye. And so too, the logic of the 1% doctrine seems to depend on some sense of shared perceived reality.

This is all to say that there are reasons to be optimistic, but these reasons may be strengthened if the collective agrees that the threat before us is, in fact, real.

Regardless of whether I am correct in speaking and arguing for optimism, we still might wonder what we ought to do. What can we do in all of this madness? I suggest two things.

First, is to talk about it. If it’s true that action follows from recognition, then we should do things that promote recognition. We ought to educate ourselves on this threat, and talk about it with those around us. And it is through this discourse that perhaps we can begin to make the whole project feel more real.

Another thing we might do is to decide where we stand. That is, we should make a conscious decision on how we feel about it. You may turn a blind eye like I once did, or embrace optimism as I do now. But regardless, I think it’s important that we know how we feel about something as important as this.

And lastly, if in fact you think there is any tinge of truth in the bleakest of probabilities, then live like it. Do the things you’ve always wanted to, spend time with those you love, and be kind to others. This, I think, is all we can do.