5 ms·
A dog brain can't design an artificial dog brain, and a human brain probably can't design a human-level AGI. It will likely be machine-evolved on cheap, massive
by bobsil1 7y ago
A dog brain can't design an artificial dog brain, and a human brain probably can't design a human-level AGI. It will likely be machine-evolved on cheap, massively parallel hardware, with the key problem being speeding up evolutionary search.
- solipsism 7y agoA dog brain can't design an artificial dog brain, and a human brain probably can't design a human-level AGI. That's ridiculous. A dog can't draw a dog, therefore a human can't draw a human?
- wpietri 7y agoTechnically, a human can't draw a human. A human can only draw something that looks like a human to another human. The human viewer is doing most of the work by imagining that the drawn human is real (or by recalling the real human suggested by the drawing). As an example, consider all of these ASCII stick figures: http://www.ascii-art.de/ascii/s/stickman.txt http://www.ascii-art.de/ascii/s/stickman.txt Clearly, it isn't necessary to deeply comprehend the essential nature of humanity in order to draw one. Knowing how a mind works well enough to build one is an immense task. A better analogy would be building a human body from scratch, which is also something humans can't do.
- solipsism 7y agoThe philosophical musings about what it means to "draw a human" are pedantic and really not relevant. My point is that "dog cannot do dog-related task" does not imply or even suggest that "human cannot do human-related task". It's pseudo-logic of the sort that is unfortunately often very convincing to people.
- wpietri 7y agoThey are relevant because you used "draw a human" to prove something wrong. If mine was irrelevant, so was yours. The reason "dog brain can't design an artificial dog brain" is a useful contribution is that it gives people an intuitive understanding of a complex truth: things can't fully model themselves. Dogs can't fully understand dogs. People can't fully understand people. It's plausible to me that humans can evolve something akin to AGI. It's also plausible to me that a vast number of humans working together will manage to stumble into creating AGI. But I see no reason to think that humans have the intellectual capacity to understand a human-level mind well enough to build one intentionally.
- sterlind 7y agoIt is sorta interesting to wonder about how far we can bootstrap ourselves. We went from trees to moon landings and neural networks, maybe we and our tools can rise to the top of the Kardashev scale (barring any cataclysms.) If so, it implies there's a sort of intelligence-completeness, where species can accomplish any physically possible objective. Or maybe we will peter out before achieving practical fusion, quantum gravity or AGI. Either way, "X can't create X" is a silly argument. Humans create humans, there's whole websites dedicated to that.
- bobsil1 7y agoWe've never hand-designed anything as complex and novel as what bio evolution discovers. We can't even ship error-free word processors. The more bug-free space shuttle code doesn't do much relative to AGI. "Humans create humans" via what bio evolution found.
- bobsil1 7y agohttps://lmgtfy.com/?q=analogy https://lmgtfy.com/?q=analogy