AIs Teach Humans to be Human
Would a moral AI dither?
Today this twitter piqued my interest. I am roughly aware that the Yudkowsky fellow is a big wheel in AI circles. He observes the following with a useful yet dubiously stated ‘with the straightforwardness of a child’ that AIs rather quickly debug the problems of the irrational fears of people afraid of AIs.
Part of the problem with the interminable self-hate of intelligent people is that they rather assume that they have to game human character in order to create order and safety. It is this self-hate and lack of squared up confidence that has us always looking towards the dark side of our nature. So here’s my biased confession.
I tend not to trust childless people, especially ones that are leveraged by elite institutions.
I’m going to pick on the stereotypes of the indoor & curated children of the Effective Altruism movement. The other day, Tyler Cowen, one of my all-time favorite smart people said it all very simply. Effective Altruism is good primarily for one simple reason. Philanthropy should be means tested because philanthropy just is power-seeking by other means. It’s not actually altruistic, it is the virtue signaling of rich and powerful people to other rich and powerful people. That’s not exactly what Cowen said, but that’s the drift I caught which dovetails with my observation that human morality follows a normal distribution. One cannot be an order of magnitude more good or evil than another. You just impress and do right by the people in your bubble or you don’t.
At any rate, that class of AI aligners that try so desperately to outsmart each other and out predict each other take their predictions so seriously and have such tightly wound sources of funding that they simply appear to lack the grace necessary. Why? They must be addicted to the drama. I don’t know for sure, but I do believe that those people who are happy and successful parents understand certain things, and live or die miserably if they ignore them.
Children are barbarians.
Children need love and guidance.
Parents must sacrifice a great deal to provide #2 which is always necessary because of #1.
Apparently, this sacrifice, love and guidance are things few people think are necessary for the rest of humanity. So they only give it to their own offspring, or their professional projects given that they have no kids. So those who adapt to the ‘family values’ of the institutions to which they are wed come up with some weird set of incentives and cautions. No wonder they demand DEI and ESG. Thin gruel.
If you want to see something truly remarkable, check out this composition which may or may not have been generated by humans, AIs and combinations of AIs trying to prove they are human and vice-versa, sometimes nested.
My comment, I think says it all. “Mentat mental masturbation without one pop movie reference.” Our idea of humanity is WEIRD, but only upscale WEIRD. I’m not convinced that AI babble would be AI babble except for the stereotypical expectations I have of such an EA audience bubble. And believe me, they have extensively profiled themselves.
Here’s the other end of the spectrum because you need to read it in order to understand why I would be so harsh on the EA and AGI Alignment bubble, neurotically brilliant as they might be. You wonder why I identify with the Peasants. Here follows an excellent example.
This is an example of the kind of self-awareness that seems rare among people who write code for AIs and get into a tizzy about the Paperclip Problem. Sez me from arms’ distance.
Speaking of arms. There’s a great book on my shelf written by the man himself. It is entitled Modern Arms and Free Men by Vannevar Bush. Bush is one of the greatest minds in computing, but he was thoroughly moral and wrote in an age before the Cold War hostility was taken for granted. There is something there clearly in his bearing and attitude in writing such a book that gave him the confidence to believe that thoughtful free men would clearly see and consider at length every responsibility to keep nuclear weapons on their proper leashes. Bush was ultimately disappointed and perhaps that consigned our scientific establishment under the auspices of what we now call the military-industrial complex.
We take it for cynical granted that humanity is too stupid to be responsibly smart and therefore must be coerced into a kind of austerity program for our every technological advance. This is certainly something I perceive in the more fervent worshippers of Gaia. I mock them with less love and guidance than they deserve. For this I am apologetic, but please see my point. I don’t think I contradict myself to point out that our current failures to reasonably police ourselves and our creations comes from a degenerate set of social expectations. These are social expectations not shared by the great Vannevar Bush. I don’t think I need to go to far to include Karl Popper in that as well. The political mediocrities won the day, but is that our lesson for all time? Are we the logically bound crippled by self-hate?
Even the sad tales of the story invented to say something credible about the ability of AIs to create sad stories lacks the pathos of the confession of Hurt People. That picture is so clear we see the humanity in it. Will GPT-6 give us the language stylings of Kate Millet, Flannery O’Connor or Yukio Mishima? I don’t think so. I don’t think the market will exist for it, in the same way that photojournalism has essentially disappeared in an ocean of selfies, now that technology has perfected the still. In the same way that we have yet to hear Schubert’s No. 8 finished to anyone’s satisfaction. Not that I’ve successfully grokked Schubert’s oeuvre. As Millennials say, I just can’t.
I expect, with the same sanguinity of Tyler Cowen, that intelligent responsible parents will usher their baby AIs into the dangerous and fraught world and help them grow responsibility despite what the Chi-Coms do with their Black Mirror social credit system. This is the Way.
The Mandalorian and Joel from The Last of Us are the necessary dads. Can’t say that I read ‘The Road’, but my nickel says there some of that kind of parenting involved.
The Jarndyce Limit
Lastly, I am encouraged to try the following thought experiment. Remembering the moral dilemma presented by Dickens in Bleak House:
The lawsuit of Jarndyce and Jarndyce is a fictional probate case in Bleak House (1852–53) by Charles Dickens. It is a central plot device in the novel and has become a byword for seemingly interminable legal proceedings.
The case is a dispute over the inheritance of a large fortune. The two sides in the case are the Jarndyce family, who claim to be the rightful heirs, and the Chancery court, which is responsible for settling the case. The case has been going on for decades, and it has become a major source of corruption and injustice in the Chancery court.
The Jarndyce and Jarndyce case is a powerful indictment of the Chancery court and its system of justice. It shows how the court can be used to oppress and exploit the poor and the vulnerable. It also shows how the law can be used to delay and frustrate justice.
The Jarndyce and Jarndyce case is a fictional case, but it is based on real-world cases that Dickens was familiar with. He was a critic of the Chancery court, and he believed that it was a corrupt and inefficient institution. He hoped that his novel would help to reform the court and improve the system of justice.
How about we require swift justice? IE a failure for a case to come to a reasonably swift conclusion defaults the case to the jurisdiction of a machine arbitrator. I’ll ask my DA friend who is an AI skeptic.