3 ms·
Yes I should. Here is one of several important bits: Eugenics fundamentally won't work because you're replacing the environment as the learning target with th
by api 7y ago
Yes I should.
Here is one of several important bits:
Eugenics fundamentally won't work because you're replacing the environment as the learning target with the system's own model of fitness. You are saying "it is fit because the fit say it's fit." The result of eugenics over a long period of time would be random drift into oblivion as the goal function chases itself as the objective. Had Hitler won and the Nazis implemented their Aryan breeding program the result after 5-10 generations would resemble the antagonists in the film "The Hills Have Eyes."
Eugenics might seem to work great for 1-2 generations as you eliminate obviously unfit genotypes like severe metabolic disorders, then you'd top out and start spiraling down into a sea of ubermenschen drool.
This is also a problem by the way for runaway self-improving AI nonsense of the sort that played a role in the LessWrong mass psychosis. How will Roko's Basilisk get so super-smart without a goal function rooted in the physical universe it must model? Evaluation of that goal function requires embodiment, so you must have many many many generations of little basilisks and that takes quite a long time.
There are other problems with runaway self-improvement too, like P probably not equaling NP and the implications for the rate at which any learning system can go "where there are no roads." An AI could copy the human brain to achieve human-level AI, but beyond that there are no roads so our hypothetical super-mind is confronted by bignum^bignum combinatorial search problems. The search domain is convex so it's not as hard as cryptography, but that's like saying you must only examine every sand grain for fifty miles of the beach instead of the entire coast of Florida.
There is a reason nature took almost five billion years to "invent" conceptual thinking and such, and it's not because "evolution is slow." Biology is the largest massively parallel computer in the known universe. Biological information systems outperform our own computers by multiple orders of magnitude. Look at what the brain does on ~40W of power vs what our supercomputers can do on megawatts. That "machine" has been running for billions of years consuming many times more power than all of human civilization to generate present-day lifeforms including ourselves. Some hacker is not going to assemble some code on a Von Neumann machine that runs away and outperforms that in a matter of months or years. That's sci-fantasy nonsense on the order of Star Wars dog fighters making pew-pew noises in space. Evolution is incredibly efficient at finding and gobbling up any free lunches like that, so if such shortcuts existed they were probably found in the seas of the Cambrian epoch.
(I don't dismiss all potential AI or cyborg type scenarios, just the obviously impossible and/or silly ones.)
That's like 5% of my notes.
Problem is you're arguing with a cult. The "pill" ideologies are not based on reason but on rationalism driven by emotional appeals to personal wounds (triggers) like male insecurity and compensatory narcissism. I'm not convinced my post wouldn't be an exercise in pissing into a hurricane.
- tptacek 7y agoAll I can say is that I have some of the same frustrations dealing with otherwise productive and intelligent people who are being sucked into and spaghettified by the memetic black hole you're talking about. There aren't a lot of good counterprogramming resources and coming up with that material in the moment is a hopeless drag.
- cyril_figgis 7y ago>An AI could copy the human brain to achieve human-level AI, but beyond that there are no roads so our hypothetical super-mind is confronted by bignum^bignum combinatorial search problems. If you can copy human minds into a machine-world, then you can copy the most intelligent human minds. The most intelligent humans are pretty darn smart, and with no obligations, a surfeit of time, and effective immortality, they may make short work of your claims that intelligence can't be effectively bootstrapped. It's difficult to imagine the inner lives of people much smart than we are. So neither of us can imagine a world occupied by persons belonging to a 300 IQ race of supermen, but there are presently 200 IQ persons, if only as genetic anomalies, and technology already exists sufficient to replicate them an almost arbitrary number of times. The future's so bright I gotta wear shades.