>>15
The most obvious is this bizarre notion that "we could do it if we wanted, we just don't want to yet."
The problem is that few people have even tried it. There's early general AI models which are useful for some small inference problems, but training them is about as hard as training expert systems, if not harder, and some of those models can be too computationally expensive for them to be useful for many things besides theorem proving.
Then you have some newer models, some of which are promising, but few that have actually been tried for the task of achieving human-level intelligence. I'd say the real problem is that just not enough effort was put into verifying if those models would work in the real world (either due to lack of time, resources, funding, etc).
There are endless theories on how we might do it, of which that "numenta HTM" shit is one of the weaker.
See the previous point: people have yet to verify some of these theories. Some of them seem quite promising actually, but if people are not commited to implementing them in full, we will never know if they work (or not, and if not, what are their shortcomings). Since you claim that Hawkins' idea is not a good one, can you give a reason why not and what theories are better?
Just simulating a human brain has been proposed over and over, and just as often, it has been pointed out that there is no evidence that this approach would work, even if you simulated every neuron.
I didn't propose that. The problem with simulating a human brain is with the sheer size and the time it would take to simulate it, not to mention you would need to provide sensory input and have it match the motor commands (it would have to be fully consistent with the world, which is what we were evolved to live in.), if they didn't match, the brain would not be able to learn how the "world" works, or such learning would be greatly impeded. So human full brain simulation is not a true solution to general AI, however it may be useful in better understanding human intelligence or verifying some current theories. It may be of more interest to simulate neocortical columns to verify (or confirm, as there are some highly plausible theories about its work) or discover (in the event that current theories are incorrect) their functionality. As far as I know, there is at least one neocortical column simulation project (
http://bluebrain.epfl.ch/ ). What I proposed instead is to implement the current high-level models of the neocortical column - this is within our abilities, and if those models are good enough, you could create an AI which would exhibit functionality similar to that of the mamallian brain, and with enough tweaking/"scaling", of human-like one. Still, even if you implement such a model, it would have to be fed rich (realtime) sensory data (as that's what our brain was evolved to process) and it would have to have a way of expressing change in the world (motor commands) which can then be validated by the sensory data it gets. I could elaborate some more on why I think the latter is important, but I suggest you consult the literature on these models to better understand my reasoning.
So my question to you
>>15 is:
Why do you think these models are no good? We can't know how suitable they are until we try, and it's a bit annoying that few people are willing to invest the time and the money into implementing and testing them out.
I'm a bit biased to think that biologically inspired models have a higher chance of success since we already know that "we" are intelligent, as that's how we define intelligence in the first place.