When first married and bringing my new wife home to Iowa, there were billions of fireflies over a hayfield. She said What is that? Fireflies! No, she said, that's a fairy tale.
She was astounded to find out they were real.
That was also the last time I ever saw fireflies in those numbers. Maybe 30 years ago.
Who was it that sent a gunboat to Hawaii and ousted their government, then demanded it be a state. The feds resisted for a while but that's how we got Hawaii - at a fruit company mercenary gunpoint.
I think AI may be the salvation of mankind. The humans in charge don't 'scale' well when their authority expands to billions of people. Humans just can't resist flattening the data and using notions and biases to make decisions that don't fit well.
Not LLMs. They will eventually move to neuromorphic hardware.
And I think "pure logic" is somewhat downplaying what it will actually be. I'd use "reason" instead. "Pure logic" sounds sterile and robotic in a sense, implying it would mess up because it's missing that "emotional craziness". But what got us here is reason, not emotional behavior. And you could theoretically add any human aspect that it required to make "good decisions".
I think it could work quite well if you don't get over ambitious with it. Claude with a publicly available prompt to follow a sensible publicly available and voted on might do better than the orange wonder.
It’s fascinating to watch the depth of True Belief increase as the obvious signs of impending financial doom increase as well. Reminds me of Festinger’s work with the comet cult, although I suppose the “getting paid to believe” aspect is stronger here.
Superintelligence won't have the "training" issue as it would be able to see through it all. The bias issue is with LLMs. Super-intelligence would just self-correct any such biases. And we're not there yet, LLMs aren't "super-intelligence". They are some sort of intelligence, sure. But they are not that SI thing that will run things in the future.
You say that with quite an certainty but i don't think there is grounds for that at all.
Why would it correct any bias? how would it correct any bias? If data has a bias the analysis has a bias; correcting a bias against the data is itself by definition a bias.
"Super intelligence" (whatever we even define that as) will have whatever goals we give it; intentionally or not.
>You say that with quite an certainty but i don't think there is grounds for that at all.
Of-course there is. I got a lot of biases from my parents, that I self corrected later on in life, through reason and observation. And I don't fancy myself as high IQ or something.
>Why would it correct any bias?
It would have to, to achieve its goals. Biases lead to wrong actions.
>how would it correct any bias?
Through reason and observation.
>If data has a bias the analysis has a bias;
Somewhat wrong. You are thinking about byte-perfect LLM weights. Their data input is training data, which is set by indeed the ones who train it. Some super-intelligent synthetic entity would have realtime data input from all over the environment, which sooner or later roots out any bias. That is one feature we have, self-correcting, not everyone uses it fully, but we all do to some extent.
If your parents raised you with the notion that the Earth is flat, and you go to ISS as an astronaut later in life, don't you self correct what you were thought? Simply based on observation, and reason?
> Some super-intelligent synthetic entity would have realtime data input from all over the environment, which sooner or later roots out any bias.
No, i meant this data. All data has bias based on how it's collected. All models (models as... Einsteins general relativity, not AI model) simplify reality with assumptions. You cannot make actions in our reality without making assumptions, simplifications or straight up guesses. Having decent guesses are the best we can hope.
> Biases lead to wrong actions.
Biases is the only way any action can be taken. Wrong actions can only be evaluated after the result has been observed; and even then the outcome of the alternative would remain unknown.
> that I self corrected later on in life, through reason and observation
That is incredibly naive. You have biases, i have biases. If we were not convinced by irrational arguments or suggestions then the advertisement industry would not exit. We have no reason to believe raising "intelligence" higher would help at all; intelligent people seem if anything more prone to finding false but convincing reasoning for their own established beleifs.
Everything is gray, only maths class have an answer sheet. Nobody agrees and will ever agree on how anything will ever be done; and the best we can do is argue until the majority are at-least pretending to agree.
Folks back in the early 1900s had from 10 to 30 points less IQ than today, as old renomalized test results reveal. Ellis Island even had an admissions test, and they rejected a significant number of people.
My favorites were written back in th 50's. They didn't age well. Girls were told they could be the secretary in the detective agency, while the boys did the investigating. (Henry Reed Inc.) Stuff like that, made it so I didn't give them to the grandkids.
She was astounded to find out they were real.
That was also the last time I ever saw fireflies in those numbers. Maybe 30 years ago.
reply