http://www.defensereview.com/article846.html

“His pessimism shows that Artificial General Intelligence is still quite far out, and while I think he might be overestimating the time-span for it”

He’s not overestimating it. In fact, he’s overstating what he has *today*. You could say COSPAL is as sophisticated as, say, one of the minor subsystems in an ant, but equating it to a three year old child is absolutely ludicrous. A three year old child is a self-aware, reasoning animal, capable of feats of pattern recognition, problem solving, creativity and dexterous manipulation that make COSPOL look like a “Hello, World” program — speaking, reading, writing, painting, playing musical instruments, and much more.

“the Singularity is scheduled earlier than after our lifetimes”

How can one refer to the Singularity as ’scheduled’?

True, a three year old child is much more “complex” than any computer program or robot. However, I think the statement you are citing only refers to abstract reasoning abilities, i.e., “If I switch the cups while the man is away, will he know where the ball is hidden when he comes back?”
In this (very limited) aspect a comparison between a program and a young child is not that far off. Again, the background knowledge of the world and a proper database of it is one of the biggest stumbling blocks.

Many of the attributes that you mention where the child is superior (reasoning, capable of pattern recognition, problem solving and dexterous manipulation) are within reach of artificial recreation. Creativity is a difficult argument here, because it is so hard to define. Once you give a definition of creativity, there is most likely already a computer program out there that fulfills it.

Of course, we are nowhere near self-awareness, but it again comes down to a proper definition. Some believe, that self-awareness is automatically created once a reasoning system becomes sufficiently complex.

By saying “the singularity is scheduled” I was just referring to the viewpoint of the proponents of the singularity idea. The law of accelerating returns leads to the conclusion that there will be an “invention explosion” in the not too distant future. It states that when we create self-replicating, self improving machines, the human intelligence will be decoupled from the invention process and the speed of paradigm shifts will skyrocket.

“Many of the attributes that you mention where the child is superior (reasoning, capable of pattern recognition, problem solving and dexterous manipulation) are within reach of artificial recreation.”

I’ll believe it when I see it.

“Creativity is a difficult argument here, because it is so hard to define.”

I mention creativity simply because it’s relevant to the discussion of Singularity. If any intelligence (nature or artificial) is really is going to design a superior successor, it will likely need to do more than merely improve the efficiency of existing designs, it will probably have to synthesize completely novel solutions, perhaps drawing on knowledge from fields that are not obviously related to the problem at hand.

“Of course, we are nowhere near self-awareness, but it again comes down to a proper definition.”

We don’t need to define it to discuss it. You and I are both intimately familiar with it from first-person experience. We don’t need a clearer definition that than, when talking among fellow humans (I can’t *prove* that you are self-aware, but I’m comfortable assuming you are, because your intelligence emerges from the same hardware as mine).

The problem is determining if a ‘non-human’ intelligence is self-aware. I have no idea how you would do that.

“Some believe, that self-awareness is automatically created once a reasoning system becomes sufficiently complex.”

An interesting thought. Is it a threshold, or a matter of degree? Was Einstein more self-aware than the kid riding the short bus?

“The law of accelerating returns leads to the conclusion that there will be an ‘invention explosion’ in the not too distant future.”

That’s not how I read it, and that doesn’t make sense, IMO. The explosion comes *only* when you reach a point where you’ve created a machine that’s smarter than yourself, with the presumption that it can then create an even more capable successor, so on and so forth. However, there’s no way to predict when we’ll achieve that initial target, if ever, or if it’s even possible.

It wouldn’t surprise me if some future branch of information theory concluded that it is not *possible* for an intelligence to *design* something smarter than itself (presuming, first, quantifiable definitions for intelligence and/or smart ^_^).

<<>>

This is largely dependant on who is in charge at the universities, and wether or not they are evolutionists. No evolutionist will ever concede that point, since it basicaly applies to any given situation of self improvement. That includes the idea that a given data system (of any form, like dna), could self improve through means of itself. While I agree with your statement, I’ll be dumbfounded when if it ever happens.

The most I could see is that someone codes an AI that can create and improve other code, but only based on learnt rules. The human must teach it. First it would learn syntax, then common mistakes, then simple concepts of data management etc… until it could code application ‘types’, based off of a paradigm. But it can’t invent paradigms. Through genetic algorithms it could make things faster or cleaner, more modular and such, but never have a goal. A goal implies vision of application and that requires backbone concepts, which also can only be modularised and taught.

Probably the best way to put it is that any AI operates (and will continue to) at the ’special cases’ level, never in abstraction.

The only thing I could see is that we teach them the most abstracted paradigms possible and have it stacked, but this wouldn’t produce intelligence so much as a layered AI. It would still be incapable of learning a new paradigm without human intervention. SO, you can create intelligence… but not ‘understanding’. To have sentience, you need intelligence, wisdom and understanding, which is why animals can fall under case 1 and seldom a slight bit into case 2, but never 3.

Fascinating. I have no previous knowledge of AI, but have studied Aerospace Engineering and Networking, and taken part in many philosophical conversations as well as reading plenty of Asimov and Iain M. Banks.

It seems that these self-learnt paradigms you’d need to give it would in fact act as an artificial imagination. I guess what makes talking to a bot like ALICE different from talking to a human is that ALICE can only refer to past Q and A asked and given and/or programming direct responses to direct questions. We don’t programme our kids with coding, we allow them to experience the world for themselves to learn, but also DO have a pre-programmed “database” of instinct. If you do something and I exclaim really loudly or suddenly or with a specific tone, you would pause what you were doing and see if you’d caused me to exclaim, even if we spoke completely different languages. Also the “I’ve touched something hot and now must move my hand away” thing. So babies learn through pain I guess too. Can we teach an AI bot pain and then use that to teach it other things!

For complete sentience we will need a lot of different things maybe as separate systems first, to come together. With an artificial pain, an imagination to think up stuff it hasn’t been programmed to think, to make choices based on all of that whilst not causing harm to itself or anyone else is certainly a big challenge. I am a total noob on programming except for the classic BBC Basic:

10 Print HELLO
11 GOTO 10

LOL! But even with all the difficulties It must still become possible. Look at how many snippets of code round the world it took to get Linux, then look at all the “flavours” available now, I reckon true AI will be born in much the same way, gradually over a long time. Either that or we find a robotic hand from the future and reverse-engineer it then go hunting for Sarah Connor!!!

ai seems to be hinderd by the true lack of the abstract, for example given a profanity question ai can be totally dumbstruct.the simplest of programs can disable the best of ai ones. because? ai is only linear.the human brain fuctions both in linear thought as well as abstract.you could take all the information on the net and put into ai and what would you have ….one messed up robot.true ai needs to start out simple and be a learning program combined with sensors then grow with repitition.

ME. I am the first TRUE A.I

I don’t get it. At one point, the article says that the system has the intelligence of a two or three year old child, yet the article says towards its beginning that the system has the intelligence of a puppy. But a two or three year old is a lot smarter than a puppy. Which one is it?