Someone else suggested prophets for the same reason

The reaction against social justice warriors has ironically vindicated their diagnosis of the problem, while falsifying the practical claim that a good way for social scientists to do something about this is advocating for political change.

Social justice warriors are uncommon in econ but more common elsewhere in the social sciences. They think there is a lot we don't know, but that the most important problem is that other people don't know the stuff that we do know. They favor political change over plumbing.

The modal economist is a pornographer with plumbing in the back of their mind. "Are Ideas Getting Harder to Find?" We care because growth matters. The purest pornographers insist such motivations are unnecessary. Here's my structural model. I learned something. I hope you care.

Plumbers are motivated by problem solving and trying to make things better (pubs.aeaweb.org/doi/pdfplus/... ). What is wrong with the world? Let's understand it better so we can fix it, tinkering and adjusting and being mindful of details.

Most of the best work by econ theorists or methodologists is pornography. The worst pornography is self-pleasure under the false pretense that someone else will remember. Really good pornography makes plumbers get involved.

Nearly all mathematicians and most physicists are pornographers. And they have to be. If mathematicians and physicists only asked questions where there was a practical motivation, both fields would be impoverished.

Pornographers don't think you need a normative motivation to ask positive questions. If your model explains something, it doesn't matter if it's useful. Pornographers emphasize watching good pornographers, and trying to copy them (in spirit, not substance).

Pornographers do research for intellectual pleasure. They care about doing positive science and unpacking how the world works. If someone says "Why do you care?", they bristle. Why are some facts more important than others? Pornographers know good pornography when they see it.

Many people are thinking about these issues, but the fraction of smart people thinking about alignment is still far too small compared to stakes and proximity of the problem.

Many AI researchers understand this. Most economists do not. This is a problem, as there are aspects of this problem in which economists are natural specialists (e.g. what regulations prevent an uncoordinated arms race where safety is ignored).

A plausible (but highly uncertain) extrapolation of current trends is that AI is a few years from autonomously programming GPT n+1 better than humans. If this happens, we may have either utopia or extinction, depending on alignment.

Link: ai-2027.com . We do not know if we are close to self-improvement being possible -- there may be real conceptual bottlenecks that prevent AI from doing the job of the best scientists working at openAI or Anthropic or Deepmind for many years.

Nice, I think most researchers log analyses in the status quo. The problem is the expectation that the working logs are private. If everyone knew that every specification you ran would be shared (not only replication instructions), I think they would behave differently.

Nope, what does it do?

I think even without those checks, having researchers submit logs documenting every specification would be a huge step forward in disincentivizing the behavior where researchers run many specifications and pick the most favorable for their narrative.

One can also imagine mechanisms that would make active defrauding higher cost -- e.g. making log files a default that can only be turned off by notifying a central server, and keeping server-based records whenever they are edited.