Maybe the millions of years they had to perfect their brains for the tasks needed for their survival has something to do with it; modern computing has not even one century in the making.
Naive/silly question: are we getting anywhere near actually being able to exploit all these "natural computers" that exist in the wild? Are we getting close to being able to use the brain of a moth to compute things with higher quality than what current AI is able to produce? And if so, how close are we to actually taking advantage of all that compute power, for instance to mine bitcoin?
This is one of the most terrifying comments I've ever read. The prospect of harvesting or growing brains to use as computing devices isn't just some weird thing from cheap scifi anymore, but something someone out there is actually researching and trying to make work at this very moment.
Growing organs for transplants or meat for consumption is one thing, but using the brain of a living creature purely as a tool is a whole different kind of horror.
They're not really "natural computers". What brains do really well is calculating heuristics. For example, "that smells bad, I probably shouldn't eat it". That's very different from asking an animal to calculate the optimal eating strategy given a pile of food. Calculating hash values requires exact computation and it's a tractible problem, meaning it can be done efficiently by machine, so it's not something moths are going to be better at.
Curiously, the things brains tend to be really good at are often NP-hard problems if you require the optimal solution. But this is really just a curiosity because brains don't calcualte the optimal solution and what they do calculate is good enough. So the question for me is whether we should even be trying to do any of this stuff with computers in the first place. You might be on to something there. Imagine a self-driving car controlled by moths reacting stimuli...
"One of the curious features of the deep neural networks behind machine learning is that they are surprisingly different from the neural networks in biological systems."
What a silly thing to say. It's pretty depressing that there are obviously scientists, engineers, and investors out there with the arrogance and/or naivete to believe that software neural nets are something more than mere conceptual mimicry of layman-level theory about how actual brains work.
What's really sad is that however good brains are at what they do, it's also pretty silly to think that emulating them more closely is a good way to improve digital computer systems, which have entirely different strengths. Yes, take what we can learn from how brains process information efficiently, but don't try to force an incomplete and inaccurate model of brain function onto non-biological hardware and expect it to be anything more than a poor brain and a poor computer.
Wonder if the Octopamine acts as some kind of bucket brigade algorithm to reward a recorded sequences of stimuli/actions pairs right before a positive result.
kind of like smell this go left, smell this go right...you win...reward previous actions.
11 comments
[ 1650 ms ] story [ 2080 ms ] threadGrowing organs for transplants or meat for consumption is one thing, but using the brain of a living creature purely as a tool is a whole different kind of horror.
http://www.research.ufl.edu/publications/explore/v10n1/extra...
You'll probably be able to find earlier examples.
Curiously, the things brains tend to be really good at are often NP-hard problems if you require the optimal solution. But this is really just a curiosity because brains don't calcualte the optimal solution and what they do calculate is good enough. So the question for me is whether we should even be trying to do any of this stuff with computers in the first place. You might be on to something there. Imagine a self-driving car controlled by moths reacting stimuli...
https://news.ycombinator.com/item?id=16363479
What a silly thing to say. It's pretty depressing that there are obviously scientists, engineers, and investors out there with the arrogance and/or naivete to believe that software neural nets are something more than mere conceptual mimicry of layman-level theory about how actual brains work.
What's really sad is that however good brains are at what they do, it's also pretty silly to think that emulating them more closely is a good way to improve digital computer systems, which have entirely different strengths. Yes, take what we can learn from how brains process information efficiently, but don't try to force an incomplete and inaccurate model of brain function onto non-biological hardware and expect it to be anything more than a poor brain and a poor computer.
It just goes "octopamine ... something something, biological neural nets don't use backpropagation".
kind of like smell this go left, smell this go right...you win...reward previous actions.