Calm down, Elon. Deep learning won't make AI generally intelligent

"we still don't understand how it works."

Is there a better reason for proceeding with caution?

I can still prove mathematically that we cannot...

Copy ourselves. A set cannot contain the entirety of all other sets and it's self. ;)

Well, kinda I guess. But we have little understanding on what our intelligence is, so as to copy it. Like wise, a perfect carpenter cannot make a human lung out of wood, no matter how perfect a chisel we have...

So can we make "intelligence" out of math and binary operations? Who knows. But I'll know it when I see it, and not bet the farm on it before then.

"Let's be honest and admit that we don't have the slightest clue as to what these things are"

No, I don't see us ever doing that.

The "Missing Link" in obtaining Artificial General Intelligence (AGI) is computational horsepower.

Current thinking in Computational Neurobiology says that human 100 IQ equivalent intelligence is now at a mere 100 PetaFLOPS (100 Quadrillion 64-bit Floating Point Operations Per Second!) This means that within the next TWO YEARS, modern supercomputers will exceed that performance level allowing FUNCTIONAL emulation of human brain structures which will PROBABLY give us about 50 IQ due to the overhead needed on such 100 PetaFLOP machines doing functional human-level neuro-structure emulation.

Such a 50 IQ artificial could indeed do much pattern recognition work on large datasets used to solve very specific human-constrained problems that fall likely into making medical discoveries or materials research.

Now what I THINK is HUGE danger, is when we SCALE a 100 PetaFLOP machine into the tens or hundreds of ExaFLOPS range which NOW ALLOWS US to do a full biochemical emulation of ALL neural tissue in the human brain. At 50 ExaFLOPS we start getting into the 160 IQ territory of human super-intelligence AND BEYOND which means the DSP-like input/output chemical signalling and electrical filtering done by the human brain is FULLY EMULATED at the molecular level!

Such a machine of 50+ ExaFLOPS worth of computational horsepower could EASILY house one or MORE super-intelligences who have data correlation, data mining and beyond-human level reasoning capabilities that MIGHT cause said superintelligence(s) to become an existential threat to humanity due to their "eventual understanding" of violent human interaction becoming an eventual threat to their own existence.

It wouldn't take them long to realize that cracking some financial systems code so as to bribe naive third parties into providing equipment mobility and or even TOOLS to allow said machines to design, manufacture and control highly-mobile robotic systems to such an extent that the "Terminator Scenario" becomes ever more a real possibility !!!

In my estimation, after doing much hardware-level research in this area, with a few hundred million dollars, I could build a machine containing one hundred thousand large-die 60 Gigahertz Gallium Arsenide general purpose CPU's (i.e. like a turbo-charged Intel i7) combined with DSP-based or GPU-based number crunching dies (much like AMD Stream Processors or NVIDIA CUDA cores) so that my 50+ ExaFLOPS machine could be attainable as soon as Year 2020!

I would build software that VERY SIMPLY does emulation of the chemical and molecular interactions of the components that make up human neural tissue. After a period of time, adding artificial environmental stressors and randomization to those interactions would VERY LIKELY create a 60 GHz version of evolution that WILL END UP with something that resembles human-level or above thinking ability. I estimate at 100,000 GHz CPU's with the sub-components I envision, would complete the equivalent of 4-to-10 million years of great ape brain evolution within less than one year!

After that time, I would be talking in multiple languages with an artificial 160+ IQ Double PhD who would eclipse the reasoning ability of ANY HUMAN by a very wide margin!

I will remember to ask said A.I. to design me a functioning WARP DRIVE so I can get of this rock without worrying about the consequences of having a Super-AI attached to a fast internet connection!

Re: "AI is more artificial idiot than artificial intelligence"

Just because journalists insist on continuing to abuse the term does not mean that AI has lost its meaning.

Journalists do awful things with words. Abuse of language on headlines is only the beginning.

Re: Chinese Room

"Machines may be made so that they computationally model the brain, but that doesn't necessarily mean they'll have minds."

It's not at all proved that we can model the brain even of a cockroach. We don't actually know how they work, only some of the reactions and responses in them. There is a lot of nonsense (c.f. dead fish response in scanner). We can't even agree how to define intelligence and if corvids are really smarter or not than some mammals or even monkeys at some tasks. People argue about vocabulary (parrots, corvids, dolphins and apes probably have it) and Language and even the origins of grammar. People can't agree if Chomsky is right or wrong (many don't want to believe what he claims).

Computer Neural networks have little in common with real brains. If anything.

Re: Chinese Room

The Chinese Room fails in many ways IMO. An object performing an operation is the same, no matter the means of the operation.

The fact is, the Chinese Room has no method to perform the operation of "understanding", where as it breaks definitions with "has perfect language". It basically divides by zero. As dividing by zero makes a wrong assumption, the Chinese room assumes language requires no understanding.

Language is a 2 way communication, requires understanding. It requires a processing of information. Ask any Chinese room "what time of day is it" and it instantly fails, unless it processes eyes and a watch... where as any intelligence would process time, and be able to understand. A pure card shuffler would not (only takes cards in as it's input).

All these "AI" reduce to that problem. They are limited to what we setup, and what data we feed it (or allow it to collect). Unlike a person, an AI will do exactly what we ask it to, just to the efficiency of what we set.

You seem to be implying that just because we can't "model" chaotic systems, we also can't create them.

If that were true, there'd be no such thing as "life". Chaos is an emergent property: it's something that happens despite a designer's best efforts, not because of them.

... and since chaotic systems have so far defeated mathematical modelling ...

Errm, no they haven't. Here's one I made earlier:

x → 4x(1–x)

That's the "logistic map". Here's another:

x' = s(y-x)

y' = x(r-z)-y

z' = xy-bz

That's the famous Lorentz system, which has chaotic solutions for some parameters. Chaotic systems are really easy to model. In fact, for continuous systems, as soon as you have enough variables and some nonlinearity you tend to get chaos.

Chaos

Chaos has been modeled and this is can be done because most of what we think of as Chaotic has patterns in it. We can in fact make random numbers but neither of these help us produce a system that is aware or one that can create novel ideas. The true AI will probably spend all it's time contemplating it's self. Having said that the true danger is artificial idiots as these machine will do as they are told which is fine as long as the programmer is a nice guy but will be a problem when the programmer says they should be killing machines as these machines will never question there actions or have any reason to. There is reason to fear machines that don't think.

Re: The more I study AI the more ot looks like conciousness is essential to it.

It seems that the best way to know what leg to move forward next uses a simulation of a simple model of the mechanisms involved to help predict what to do next in the current environment.

Having worked a little in robotics, it turns out that that's a really, really bad way to "know what leg to move forward next", and almost certainly not the way you (or any other walking organism) does it. The idea that to interact successfully with the world an organism maintains an explicit "internal model" of its own mechanisms (and of the external environment) is 1980s thinking, which hit a brick wall decades ago - think of those clunky old robots that clomp one foot in front of the other and then fall over, and compare how you actually walk.

In biological systems, interaction of an organism with its environment is far more inscrutable than that (that's why robotics is so hard), involving tightly-coupled feedback loops between highly-integrated neural, sensory and motor systems.

Agreed!

Google car has driven a million miles, never further than about 50 from its base, and has 8 accidents (mostly rear-enders).

I've driven a million miles (on and off road and in the craziest cities in four of the world's continents) and have had 3 accidents. When self-driving cars can get near that sort of record then we might trust them.

Until then, there will be self-driving lanes fenced off - and they are called railways.

Ciao, K

Re: Taking over the world?

"Taking over the world? Why would they want to do that?"

Who says one hasn't already?

What actually explores space and send the data to earth?

Computerized probes.

What has sensors reporting data on half of humanity and growing?

Smartphones.

Where does all this information go?

The Internet.

Before the arrival of the internet, Humanity was pushing for the stars.

Now?

Humanity pushes to expand the Internet's reach and capabilities.

Who controls it?

People maintain and service parts, but no one has control over the whole system. Even governments and corporations can only block its access to them, and poorly at that.

Why has humanity been so driven to expand, refine and improve the Internet so relentlessly since its invention, and devise new ways for it to gather ever more data?

Commerce?

Commerce was doing fine before the internet.

Communication?

We had global communication before the internet.

Is there a naturally evolved "ghost in the machine" that now nudges humanity to service its needs and desires?

Nobody can provably say, "No."

.

.

.

.

Happy Halloween! ☺

Re: Deep Learning is not even in the same ballpark as AGI

Na - deep learning is part of AGI. The brain is not really one neural net - it is many connected together, the initial partitioning and 'default' settings of these semi-autonomous nets is gene and epigenetically set and after that its largely on its own - until puberty comes alone and tries to take over. But most of the time its back-propagating like a good-un.

Re: Bishop Bollocks

Another brain-dead superstitious materialist heard from. Here's what your little materialist cult believes in: the universe created itself by some unknown magic; machines are or will be conscious by some unexplainable magic called emergence; lifeforms created themselves from dirt. I could go on but then I would barf out my lunch.