EA & LW Forum Weekly Summary (30th Jan - 5th Feb 2023)

I don't have an account on the EA forum, so I will react to "EA, Sexual Harassment, and Abuse [EA · GW]" here. Many comments are about polyamory. Either from the angle "if you become famous for supporting unusual sexual behaviors, guess what, you get other unusual sexual behaviors, too", or "speaking negatively about polyamory should be considered just as inappropriate as speaking negatively about gays or trans people".

*

I am generally in favor of people doing whatever they want, if they do not hurt others (but the problem is, some people were hurt in the EA movement, so now it's just a problem of whom/what to blame), but my steelman for anti-polyamory would be the following:

There is an analogy between enforcing the norms of monogamy, and the criminal cases where crime bosses get in prison not because of their murders (there are either no witnesses alive, or they are too scared to talk), but because of tax evasion. It is difficult to convict a crime boss. But it is also difficult for the crime boss to keep his paperwork in order. So we put him in prison for failing to do the paperwork properly.

In a monogamous society, if you have a married man in a position of power... and he turns out to be a sexual predator... if you cannot fire him because of the rape (which you often can't prove; it's "he said, she said"), you can still fire him for violating the norms of monogamy. It is not the punishment he deserves, but at least it removes him from the position where he could have caused further harm.

(Of course this norm also has disadvantages; for example it discriminates against unmarried people.)

Normalizing polyamory removes this check. Now, men in positions of power can declare themselves polyamorous and start hitting on every woman in a vulnerable position... and there is nothing you can do about it, unless you can prove actual harm (which is often difficult to prove). It becomes difficult to argue clearly what exactly is wrong. And remember that accusing people in positions of power of doing something wrong can be expensive, so people in general tend to avoid that unless they can make their case very clearly.

(It's as if some country abolished taxes, which many people would celebrate... and after few years found out that is has too many crime bosses per capita, and is unable to convict them. And if you ask cops from other countries for advice, they just say "yes, it's almost impossible... that's why we usually get them for tax evasion".)

Basically, you have to consider this balance. Removing the norms of monogamy decreased the barrier for sexual predators. No, I am not saying that polyamory makes you a sexual predator. I am saying that if you already are a sexual predator, polyamory provides for you a very convenient role. ("No, I am not 'hitting on every woman in a vulnerable position'. I am simply polyamorous, and if you have a problem with that, I suggest that you talk it over with other high-status people in our community.") In order to fix this damage, you have to increase the barriers for sexual predators in some other way.

*

That said, polyamory is a red herring here. The question is, what norms of sexual behavior to adopt, in order to minimize harm. Remember that rape if often difficult to prove, so ideally you want to remove the entire class of situations of "maybe rape, maybe not", by banning them for some other reason.

And yes, there is a trade off. Banning ambiguous situations means also banning some perfectly consensual situations. Shut up and multiply!

The traditional solution is to ban sexual relations between people in positions of power. Sometimes it is difficult to define the power relations (your boss is clearly in a position of power over you, but how comfortable would you feel defending yourself from sexual advances of his best buddy?), so another solution is to ban all sexual relations in the group. It works okay for small groups or temporary groups.

So maybe the question is, what would be the optimal group size in case of EA. "All people who identify as effective altruists" is too large. A single EA organization, limited to its employees and interns, that's just saying "no sexual advances at workplace", which seems perfectly reasonable to me. (I do not know how much gray area there is between the EA organizations, e.g. whether there are people who depend on EA organizations in general for living, but cannot be clearly assigned to one specific organizations.)

This would address the problem, without worrying about polyamory. Poly or mono, simply do not make sexual advances at your workplace. And if you do, you get in trouble (and saying "hey, being polyamorous is like being gay" is not going to save you, if gay people are also not allowed to make sexual advances at their workplace).

tl;dr - the core of the problem is tolerating sexual advances at workplace; polyamory is merely a multiplier

They also separately believe that by the time an AI reaches superintelligence, it will in fact have oriented itself around a particular goal and have something like a goal slot in its cognition - but at that point, it won’t let us touch it, so the problem becomes we can't put our own objective into it.

My guess is this is a bit stronger than what Nate believes. The corresponding quote (emphasis mine) is

Separately and independently, I believe that by the time an AI has fully completed the transition to hard superintelligence, it will have ironed out a bunch of the wrinkles and will be oriented around a particular goal [...]

and I wouldn't be surprised myself if by the time an AI is superhuman at basically all tasks, it is still as incoherent as humans, especially if it uses more inference compute than a human brain.

Thanks for writing these summaries!

Unfortunately, the summary of my post "Inner Misalignment in "Simulator" LLMs" is inaccurate and makes the same mistake I wrote the post to address.

I have subsections on (what I claim are) four distinct alignment problems:

• Outer alignment for characters
• Inner alignment for characters
• Outer alignment for simulators
• Inner alignment for simulators

The summary here covers the first two, but not the third or fourth -- and the fourth one ("inner alignment for simulators") is what I'm most concerned about in this post (because I think Scott ignores it, and because I think it's hard to solve).

I can suggest an alternate summary when I find the time. If I don't get to it soon, I'd prefer that this post just link to my post without a summary.

Thanks again for making these posts, I think it's a useful service to the community.