Live data from Hacker News

Underwriting Superintelligence

underwriting-superintelligence.com

31–36 of 36 posts

Re: Underwriting Superintelligence

#31
post #22

> As insurers accurately assess risk through technical testing If that’s not “the rest of the owl” I don’t know what is. Let’s swap out superintelligence for something more tangible. Say, a financial crash due to systemic instability. How would you insure against such a thing? I see a few problems, which are even more of an issue for AI. 1. The premium one should pay depends on the expected risk, which is damage from…

Thanks for the thoughtful response! Some replies: 1. Someone is always carrying the risk; the question is, who it should be? We suggest private markets should price and carry the first $10B+ before the government backstop. That incentivizes them to price and manage it. 2. Insurance has plenty of ways to manage moral hazard (e.g. copays). Pricing any new risk is hard, but at least with AI you can run simulations, red-…

What you are saying makes sense for conventional harms like non consensual deepfakes, hallucinations, Waymo running pedestrians over, etc.

However, those are a far cry from the much more severe damages that superintelligence could enable. All of the above are damages which already could exist with current technology. Are you saying we have superintelligence now?

If not, your idea of selling superintelligence insurance hinges on the ability of anyone to price this kind of risk: an infinitely large number multiplied by another infinitely small number.

(I realize my explanation was wrong above, and should be the product of two numbers.)

I think many readers will also take issue with your contention that the private market is able to price these kinds of existential risks. Theoretically, accurate pricing would enable bioweapons research. However, the potential fallout from a disaster is so catastrophic that the government simply bans the activity outright.

Re: Underwriting Superintelligence

#32
post #30

This article is a bizarre mix of center-right economic ideas and completely unfounded assumptions about the nature of AI technology, to the point where I'm genuinely not sure if this is intended as parody or not. > We’re navigating a tightrope as Superintelligence nears. There is no evidence we're anywhere near "superintelligence" or AGI. There is no evidence any AI tools are intelligent in any sense, yet alone "supe…

> The only reference for this, given much later, is to https://ai-2027.com/ which is no more than fan fiction.

There are certainly pretty gaping holes in its logic but it’s more than a fanfic. I’m a bit confused about the incentive of its authors to add their names to it, since it seems if they’re wrong they lose credibility and if they’re right I’m not sure they’ll be able to cash in on the upside.

Re: Underwriting Superintelligence

#33

> As insurers accurately assess risk through technical testing If that’s not “the rest of the owl” I don’t know what is. Let’s swap out superintelligence for something more tangible. Say, a financial crash due to systemic instability. How would you insure against such a thing? I see a few problems, which are even more of an issue for AI. 1. The premium one should pay depends on the expected risk, which is damage from…

Not to detract from your argument but, expected risk is the expectation of [loss x probability of said loss].

Re: Underwriting Superintelligence

#34

Earlier quoted context omitted.

If you financially penalize AI researchers, either with a large lump sum or in a way which scales with their expected future earnings, take you pick, and pay the proceeds to the people who put together the very cases which lead to the fines being levied, you can very effectively freeze AGI development. If you don't think you can organize international cooperation around this you can simply put such people on some equ…

In my world there are multiple countries who each have an incentive to win this race. I know of no world where you can penalize AI researchers across international boundaries nor to believe your scenario could ever play out. You're dreaming if you think you could actually get all the players to co-operate on this. It's like expecting the world to come together on climate change. It's not happening and it's not going…

>I know of no world where you can penalize AI researchers across international boundaries nor to believe your scenario could ever play out.

This world where wars and stuff happen? You can just use actual force, you know.

>You're dreaming if you think you could actually get all the players to co-operate on this.

Let me play it out more explicitly since you lack the eyes to see. Say China implements such a rule. Top researchers in the US start mysteriously turning up in China and getting turned in for the cash fine. They aren't allowed to leave China until they empty their bank accounts, because they have broken a very serious Chinese law.

Now the US has to seriously ask itself whether it's actually willing to unleash World War 3 for the sake of a couple thousand eggheads. Not gonna happen.

Then someone takes a poll and finds this actually has mass support _in_ the US, because the average person is terrified of where AI is going. If you can't beat em, join em. The US implements its own law similar to the one in China so that this madness can stop. All of the top researchers bitch and moan and move to Portugal.

Top researchers in Portugal start mysteriously turning up in the US... So on and so forth. Rinse and request until everyone is on board with the new Nash equilibrium.

>[I]t doesn't take a huge lab to do it. You can do it at home. It might take longer but there's an 1.4kg blob in everyone's head as proof of concept.

You are thinking so unbelievably small. So, so unbelievably small.

You decide you want to push the state of AI forward and create an enormous new model. Okay, you're going to need to buy or source a bunch of GPUs. You try to order a bunch of GPUs from Nvidia. Nvidia flags you as a high risk client, turns on the tracking firmware, gathers info on your training runs, hires some people in your local area to peek through your curtains to confirm, and turns you in themselves. Nvidia walks away getting roughly half, the other half went to its contractors. You are out of the game.

Alright, so maybe you need to buy a bunch of GPUs on the secondhand market to get started. And maybe you need to source them one or two at a time from several dozen, or hundreds of different players to get away with it. A few of these players take an interest in watching you, notice you're buying GPUs from several secondhand sources, realize this would be extremely weird if you weren't planning on doing something illegal with them, and put together a small posse to turn you in. Your life savings are emptied out and your condo repossessed. You are out of the game.

> It might take longer

Thousands of years longer by my reckoning, at minimum. It's very hard to design an AGI entirely inside your own head when the outside world is so inherently adversarial.

Re: Underwriting Superintelligence

#36
post #33

> As insurers accurately assess risk through technical testing If that’s not “the rest of the owl” I don’t know what is. Let’s swap out superintelligence for something more tangible. Say, a financial crash due to systemic instability. How would you insure against such a thing? I see a few problems, which are even more of an issue for AI. 1. The premium one should pay depends on the expected risk, which is damage from…

Not to detract from your argument but, expected risk is the expectation of [loss x probability of said loss].

Thank you, I noticed that too late unfortunately
Post reply on HN