(April 7, 2018 at 12:50 am)ignoramus Wrote: We have power issues to solve
We have general AI issues to solve
We have mobility/strength issues to solve.
What other issues/obstacles do we need to solve?
Sensing issues. A robot needs to sense its environment. Think about the kind of sensory input we have as humans. The amount of information each eye ball receives, how we can listen to the sound of a pin drop or a rock concert, or isolate a single voice in a crowded room. The same with taste and smell. The amount of tactile sensation we have all over our skin but also proprioception and knowing where our limbs are while they are still moving. This is being processed continually all the time and yet we take it for granted, until something goes wrong.
(April 7, 2018 at 12:50 am)ignoramus Wrote: Do you think nothing like this (singularity) can ever be accomplished unless we master quantum computing first?
The problem with quantum computing is that it requires a large infrastructure (at the moment) to keep everything cooled. This touches upon another challenge in that the processing has to take place locally. People assume that signals can be sent elsewhere, processed and the results sent back. But that induces lag which fundamentally limits what an agent can do in real time as it interacts with the world. For example, keeping its balance.
I actually think it would make more practical sense to grow and engineer real brains grown from natural or artificial cells. That would solve both problems of connectivity and power consumption.There has been a lot of work in this area to create prosthetic limbs. A classic experiment a few years ago used the brain of a lamphrey fish to control a robot.
The problem is that we do not have the ability to measure the workings of the brain down to a fine enough scale to see what's going on in situ. And if we could, it would still take a very long time to figure it out.
On the other hand, quantum computing would be extremely useful to simulating environments and evolving designs. And this brings up another challenge. Evolution.
No matter what system you create, it will require a set of parameters that need to be configured. In the real world we have benefited from many millions of years of evolution taking place using a large population acting within many different environments. A potential solution can't be tested just once to determine its fitness, it needs to be tested many times in many different environments. If nothing else to average out the lucky / unlucky ones. So even if we have sufficient processing power to simulate an entire human brain, you'd need many, many, many orders of magnitude more to simulate and test all the different possible variations in many different ways.
My AI is simple enough and fast enough to adapt in real time but it still takes weeks to evolve for example. It generally takes me about 3 months of work to figure out how it actually functions in practice though so I don't normally bother unless I have run out of ideas.
Another fundamental problem with the idea of a singularity is the idea that AI will be able to come up with better designs. This ignores that all designs still need to be tested in order to determine if they are in fact better. How then is this any different from using a genetic algorithm or artificial evolution which we have been using for the last 20-30 years? Our experience with GAs tells us that there is no such thing as a free lunch because you need to simulate the real world accurately. And remember that mechanical movement is many orders of magnitude slower than processing if you instead prefer to try out your new designs in practice.
(April 7, 2018 at 12:50 am)ignoramus Wrote: Do you believe there are technologies which haven't been invented yet, which will be needed?
Yes. At the very least some form of processing that is capable of connectivity many orders of magnitude greater than what can be achieved using silicon. Computing with cells would solve this. I am thinking that analog computing and the field of Cybernetics need to be revisited.
(April 7, 2018 at 12:50 am)ignoramus Wrote: Or do you believe if the world threw all its concentrated resources at it with current tech, we can do it in under 10 years?
Not possible in 10 years because there just aren't that many people in the world with the necessary expertise and experience. Add to that the time taken for a paper to be written up, submitted, peer reviewed, published at a conference half a year later, read by others who then go off to write their own experiments, most of which don't work because it is difficult replicating experiments, you're talking a cycle of about a couple of years. There would need to be a fundamental change in how people collaborated.
