Pay to get AI safety info from behind NDA wall?
OpenAI/DeepMind whistleblowers have blown the whistle about how they can’t blow the whistle because of NDAs.
Would it make sense to put up the money so they can?
(Both in this specific case, and in general.)
As far as I can tell, breaching an NDA is usually unlikely to result in jail time—but it will cost a lot (legal fees trying to defend it, financial penalties, adverse consequences for whistleblowers’ careers). So the info behind the NDA wall can possibly be ‘purchased’ by putting up money to meet these costs, both in this case and in others.
I’m interested in this question not just for AI safety cases, but generally for Big Tech transparency/societal transparency. Would also be curious to know any precedents.
My prior on this is that the opportunity cost of what the money could’ve been spent on is excessively high and there are much better uses it could go towards.
Obviously the more novel and actionable the information is, the lower that trade-off becomes. However, I expect that most of the time the information they can provide would be valued too high due to personal intrigue as opposed to it actually being anything that meaningfully moves the dial. Equally, if the information they have was especially ground-breaking then I’d hope the person under the NDA would sacrifice personal wealth to expose that, then retrospectively they may get support with any legal costs etc. A reactive system as opposed proactive would also help prevent weird incentives being created where people would try to hold out on blowing the whistle until they had funds confirmed.
If the info is useless, then for sure the opportunity cost is too high. There’s an info problem there, which is establishing the usefulness of the info prior to paying for it. Might have to involve some ‘weak’ breach of NDA, e.g. share the info or some meta-info with trusted experts (ironically, maybe under yet another NDA for protection!).
In this case, for instance, I think it would be useful if Bengio, Hinton, Russell (endorsers of the letter) were informally told about the info, at least in broad strokes, and could then make a pitch to someone to put up funds for full disclosure. Or, they could say ‘Na, it’s not worth it’ - but either way, there’s a system in place.
I’d also hope a person would sacrifice personal wealth, but don’t think hope is a good strategy and can see how more disclosure would happen with more funds to protect whistleblowers prior to making the decision to blow the whistle. I’d think banking on financial support after the fact is too scary for most people.
Agree re: the incentive misalignment, and that’s a problem. Wonder if some kind of contract between whistleblower and funder can help there.