5 ms·

Cold fusion is a hypothesized type of nuclear reaction. Solving AI is more like solving flight. Before flight was solved, leading experts believed it would be

Cold fusion is a hypothesized type of nuclear reaction. Solving AI is more like solving flight. Before flight was solved, leading experts believed it would be impossible. Same could be said for the moon landing. Hackernews crowd seems to be in denial about how true AI would make most of them redundant.

umvi 7y ago
Until we fully understand the human brain, we won't be anywhere near AGI or "true AI" as you call it. Current AGI efforts amount to cargo cult science where experimenters create increasingly vast neural networks and hope some kind of intelligence/consciousness magically emerges. So far "AI" seems to be a combination of if statements, neural networks, and linear algebra - which is great for specialized tasks! But light years away from the way humans can pick an area and self-teach. Believe it or not, not all of HN is atheist either, so that means many people believe the human mind is more than just a biological Turing machine/neural network and that there is some sort of "spirit" or "soul" if you will that accounts for our first person consciousness experience. Hence, without a spirit/soul, you'll never be able to create a conscious computer.
- rmah 7y ago
  First off, I agree with your general message that getting an AGI is very very very hard and that we are no where close. That said, however, we do already know quite a bit about how neural nets work, both biological and artificial. One of the things we know is that they are most certainly not a Turing machine -- they are associative memory machines. That is, memory and processing are one and the same. Most (all?) current artificial neural nets have fixed interconnects between neurons. One thing we do know is that biological brains change which neurons connect to which other neurons. And as far as I'm aware, we don't know how that's done or what drives it except in the most general terms. This lack of plasticity (not sure if that's the appropriate term) in artificial neural nets is, IMO, a major blockage in the road to an AGI.. IMO, "intelligence" is an emergent of a complex, malleable, neural network that has been exposed to the right environment. I suspect both are necessary. And finally, we don't need an AGI as "smart" as humans (we have plenty of humans, after all). But even one as "smart" as a bright dog would be very useful. I don't think most religious people think dogs have souls, so I don't see how souls are an issue.
  - umvi 7y ago
    > I don't think most religious people think dogs have souls, so I don't see how souls are an issue. Really? I was taught growing up that all life has a soul/spirit. It's what sets life apart from machines or other non-living matter. Sure, a dog's spirit might be a lesser one than a human spirit, but it still has one...
poilcn 7y ago
People talking about AI dramatically underestimate complexity of intelligence. The computation power it would require to solve such complex problems and human's ability to comprehend such complex things. We can understand how a neuron works and even how several of them would work together. We can emulate nervous system of the simplest organism. But our brains have no physical capacity to analyze this network as a whole. Requirements of resources used to improve AI is not linear to the output. Increasing processing power in 10 times won't scale to 10 times more effective AI.

umvi 7y ago

Until we fully understand the human brain, we won't be anywhere near AGI or "true AI" as you call it. Current AGI efforts amount to cargo cult science where experimenters create increasingly vast neural networks and hope some kind of intelligence/consciousness magically emerges. So far "AI" seems to be a combination of if statements, neural networks, and linear algebra - which is great for specialized tasks! But light years away from the way humans can pick an area and self-teach. Believe it or not, not all of HN is atheist either, so that means many people believe the human mind is more than just a biological Turing machine/neural network and that there is some sort of "spirit" or "soul" if you will that accounts for our first person consciousness experience. Hence, without a spirit/soul, you'll never be able to create a conscious computer.

rmah 7y ago
First off, I agree with your general message that getting an AGI is very very very hard and that we are no where close. That said, however, we do already know quite a bit about how neural nets work, both biological and artificial. One of the things we know is that they are most certainly not a Turing machine -- they are associative memory machines. That is, memory and processing are one and the same. Most (all?) current artificial neural nets have fixed interconnects between neurons. One thing we do know is that biological brains change which neurons connect to which other neurons. And as far as I'm aware, we don't know how that's done or what drives it except in the most general terms. This lack of plasticity (not sure if that's the appropriate term) in artificial neural nets is, IMO, a major blockage in the road to an AGI.. IMO, "intelligence" is an emergent of a complex, malleable, neural network that has been exposed to the right environment. I suspect both are necessary. And finally, we don't need an AGI as "smart" as humans (we have plenty of humans, after all). But even one as "smart" as a bright dog would be very useful. I don't think most religious people think dogs have souls, so I don't see how souls are an issue.
- umvi 7y ago
  > I don't think most religious people think dogs have souls, so I don't see how souls are an issue. Really? I was taught growing up that all life has a soul/spirit. It's what sets life apart from machines or other non-living matter. Sure, a dog's spirit might be a lesser one than a human spirit, but it still has one...

poilcn 7y ago

People talking about AI dramatically underestimate complexity of intelligence. The computation power it would require to solve such complex problems and human's ability to comprehend such complex things. We can understand how a neuron works and even how several of them would work together. We can emulate nervous system of the simplest organism. But our brains have no physical capacity to analyze this network as a whole. Requirements of resources used to improve AI is not linear to the output. Increasing processing power in 10 times won't scale to 10 times more effective AI.