The breeze whispers of a transformation, a time of great trial and tribulation for mankind. We face an adversary whose complexity is almost unimaginable. Its vast computing power makes the human brain look like a toy. Worse still, it wields against us a powerful nanotechnology, crafting autonomous machines out of thin air. Already, more than one percent of the Earth is under its sway. I’m talking of course about the mighty Amazon rainforest.
But humans have long ago learned to live with our ancient enemies, the plants and the animals. Perhaps we can take some lessons for how to navigate our new friends, the AIs, who may be arriving any day now.
The rainforest is not actively trying to kill us, at least not most of the time. It is locked into a fierce struggle with itself, deploying its vast resources in internal competition. As a side effect, it produces large amounts of oxygen, food and other ecosystem services that are of great benefit to humanity. But the rainforest doesn’t like us, doesn’t care about us, mostly doesn’t even notice us. It exists in a private hell of hyper-competition, honed to a sharp point by the lathe of selection turning for a billion years.
So here is one model for AI safety. Don’t hope for direct control. Don’t dream of singletons. Instead, design a game that locks the AIs in a competition we don’t care about, orthogonal to the human world, perhaps with a few ecosystem services thrown our way as a side-effect. We will only collect scraps from the table of the Gods. But while the Gods are busy with their own games, we can get on with ours.
I think of the Irish proverb: What’s as big as half the Moon? Answer: The other half. Good advice for fighting with God.
Written with assistance from ChatGPT :-)
How do we keep them loyal to the orthogonal game? At a minimum, wouldn’t drift and errors lead to some of them doing something else after long enough, and potentially taking over?
Also, how do you make sure the orthogonal game has no real-world consequences? What if AIs start competing to get systems run by their opponents to dedicate themselves constantly to pointless tasks?
Finally, if they have this orthogonal game to suck up their enormous capabilities, how will AIs still be useful to us?
Hey Milan! Very good questions.
I wrote the post not because I think this is necessarily the best approach, or that it’s easy to make work, but because it seems like it’s at least a possible approach, and I haven’t seen other people mentioning it.
Scott Aaronson wrote a blog post recently where he lists the “Eight Approaches to AI Alignment”, and this one isn’t on the list. Hence this post.
> Finally, if they have this orthogonal game to suck up their enormous capabilities, how will AIs still be useful to us?
To answer the last question first – they won’t be, but I’m claiming that’s OK.
Everyone is trying to make AI controllable. Essentially they want to trap God in a box and make him do our bidding. Everyone agrees this is hard.
But what if you take a different perspective – God is arriving, and we just want to make the event as survivable as possible. We give up on control, and we give up on utility, and maybe then we gain extra degrees of freedom to make the “survival” part work.
The situation right now is as if we have a completely bare patch of soil, and plants have just been invented. Trying to keep the soil totally bare for eternity seems like a losing battle (ask any gardener). So on the assumption that *something* is going to grow on that soil, maybe we can at least nudge it towards growing heather instead of growing briars. The heather isn’t of much use to us, except that it out-competes the briars and is less harmful to us.
A world in which we swear off development of (strong) AI altogether, and never have to face these problems, might be even better. However, a lot of people worry that it’s almost impossible to navigate to that world in a reliable way, and I tend to agree. It’s like preventing nuclear proliferation, but much harder because the materials are already widely available to everyone, and there’s clear economic incentives to build AI.
> How do we keep them loyal to the orthogonal game? At a minimum, wouldn’t drift and errors lead to some of them doing something else after long enough, and potentially taking over?
Yes, this seems hard. I do think it’s possible to design a game that could be stable for a long time (biology in a sense is an existence proof of this, ecosystems can stay in the same equilibrium for many millions of years). “Forever” is a long time though. Some kind of drift seems inevitable eventually (as biology also shows, e.g. you often get a punctuated equilibrium, where things can be static for many millions of years, but then shift to a new regime)
> Also, how do you make sure the orthogonal game has no real-world consequences?
This also seems hard.
It will undoubtedly have some real-world consequences, e.g. the AIs will consume resources to support their existence.
You would just hope to design a game such that:
– the real-world consequences are survivable for us
– the conditions are sufficiently competitive that “our AIs” outcompete any other rival AI population that tries to take hold
I don’t know how to achieve that. It may not be possible, and this could all be a bad plan. My goal in writing it was just to add one more option to the list of possible plans.