nzonbi

>There’s plenty of active research in higher-level “cognition” like decision theories and semantics, so I don’t think it’s crazy to believe good results will emerge there in the next few decades.

Interesting. Can someone elaborate about this?

Some thoughts:

Progress is not like a liquid, that increases continually with research. Progress is an unknown function, composed of huge number of discrete "discoveries". Each discovery is hidden behind research of variable level of difficulty. Each discovery, may or may not enable more discoveries. The reality may have either a finite or infinite amount of discoveries waiting to be made. We don't know for sure. This universe could have a finite amount of discoveries. But it is possible that we may find a way to travel to infinite other universes, and these may have infinite discoveries waiting to be made.

The shape of the technical progress function, depends on these unknown factors. So it is wrong to assume that it is an exponential. Although an exponential seems like a good approximation around the local point in which we currently are. The population have been growing, and the economy have been growing. These two factors have enabled an increased amount of resources dedicated to research. The more research being done, increases the probability of discovering the available discoveries that exist at the current technological level. The speed of progress depends on the amount of research, and the number of latent discoveries hidden in our reality.

We are approximating the point, in which we can build artificial general intelligence. This will be a machine similar to a human mind, but capable of dramatically faster reasoning. Its internal dialogue will be millions of times faster than a human mind. Because it will move at electrical speeds, instead of biological speeds. Also it will have practically perfect and unlimited memory (compared to human capabilities). And will have almost instantaneous capability to resolve mathematical calculations of reasonable level. With these improvements, it can be expected that it will be much more effective at making discoveries than a human. Additionally these machines will be industrially replicable. So it will be possible to put a large amount of them at work on problems. It is reasonable to expect that these machines will resolve the chain of discoveries available to be made faster than humans.

These artificial machines will maximize discoveries, from the chain of discoveries available. What this is going to mean, depends on the actual unknown amount discoveries available in this universe. If things like nanotechnology, molecular machines, biological machines, etc, are actually possible, these intelligent machines are well equipped to discover them dramatically faster than we humans. If there is new physic available to be discovered, these machines have much better chance than humans at discovering it.

Will machine superintelligence actually create a singularity? maybe, or maybe not. Depends on the amount of discoveries to be made, contained in this universe, and their level of difficulty. It could be the case that our universe is running out of hidden discoveries. So any prediction on the shape of the curve of progress, is pure speculation. For example, we could have already run out of exploitable significant discoveries in physics. Or we could be on the verge of discovering faster-than-light/instantaneous communications, and lots of other things.

In my opinion the invention of artificial general intelligence, and superintelligence is imminent. A matter of years. I base this on introspective observation of the thinking process of my mind. And in comparison of it with the operation of artificial neural networks. They show similarities. The thinking process of the mind is entirely reproducible with deep learning networks, assembled in the right structure. An interesting topic is, who is doing this research. Obviously, the big tech corporations are working on it. But are states organizations also working on it? Who is doing the biggest investment? Who has the best odds of inventing it? What will happen when someone gets it? Are they going to immediately announce it?

I am working on creating basic artificial general intelligence. My research is still early, and I don't want to give much details for now. For example, in my opinion it is wrong to feed a giant neural network with tons of text, and expect it to understand human language. That will never work. Language understanding requires a grasp of a world where the agent exist first. A general understanding of things. Then, on top of that, it can build basic communication. And then it can be hoped that this communication can become sophisticated enough, to reach human language level.

Another example: Feeding a neural network with millions of product reviews, and expect it to be capable to understand and write product reviews is hopelessly laughable. Not even with petabytes of data.

I started working on this, because I think that not enough focus is being put on AGI, or research is not creative enough. At least not that is being published. I am optimist with my work, and soon I will reveal more. But even if I don't reach my goals, I think that it is just a matter of persistence. At any moment, someone will solve the problem of AGI.

Interesting article, in a difficult topic. Speculating about the future of deep learning. The author deserves recognition for writing about this. In my personal opinion, within the next 10 years, there will be systems exhibiting basic general intelligence behavior. I am currently doing early hobbist research on it, and I see it as feasible. These system will not be very powerful initially. They will exist and work in simpler simulated environments. Eventually we will be able to make these systems powerful enough to handle the real world. Although that will probably not be easy.

I somewhat disagree with the author. I don't think that deep learning systems of the future are going to generate "programs", composed of programming primitives. In my speculative view, the key for general intelligence is not very far from our current knowledge. Deep learning, as currently we have, is a good enough basic tool. There are no magic improvements to the current deep learning algorithms, hidden around the corner. Rather what I think will enable general intelligence, is assembling systems of deep learning networks in the right setup. Some of the structure of these systems will be similar to traditional programs. But the models they generate will not resemble computer programs. They will be more like data graphs.

I expect within 10 years there will be computer agents capable of communicating in simplified, but functional languages. Full human language capability will come after that. And within 20 years I expect artificial general intelligence to exist. At least in a basic form. That is my personal view. I am currently working on this.

My horse carriage is very useful. The horse just feeds itself, if you leave him in a grass field. It rarely gets sick and last many years. The wheels are easily replaced from the store.

The other day I saw a crazy guy with a mechanical contraption that, according to the guy, moved by burning wood! The guy was demoing it at the town. He filled it with wood, and then had to start the fire. The crazy machine got so hot that nobody could get near it. The thing barely moved a few meters and then Kabooooom!! it exploded in a black cloud. What sane person may ever believe that one of these contraptions could be useful?

Am I right? Are you right? we don't know. Research is the only way to know.

Interesting presentation. One thing that we agree, is that the current state of programming is not final. There have to be better ways for humans to program. I wonder how much people thinks like that. A poll would be interesting.

In reality, programming has been stuck in the text approach for many years. Perhaps some people take that as evidence that there is no more progress to be made. Another factor that may be undermining research in programming, is the AI boom. Perhaps some think that AI will displace programmers in a not too distant time.

I agree with you that programs are data. And would benefit from a more specialized representation. I have been researching the aspect of program logic. For example, one thing that I think is terrible in programming, is deep nesting levels. That makes it complex to follow programs. I have been exploring ways to improve that. In my opinion graphical visualizations can provide ways to improve that.

Imagine how much energy the world would save if most people where working from home. Electricity, transport fuel. How much less pollution and traffic jams. How much less need to build cars and transport systems. How much extra time to enjoy life would people have by not commuting.

With so huge positives. I don't understand why not more research is put into work-from home.