First Demonstration of Artificial Intelligence On a Quantum Computer

KentuckyFC writes: Machine learning algorithms use a training dataset to learn how to recognize features in images and use this 'knowledge' to spot the same features in new images. The computational complexity of this task is such that the time required to solve it increases in polynomial time with the number of images in the training set and the complexity of the "learned" feature. So it's no surprise that quantum computers ought to be able to rapidly speed up this process. Indeed, a group of theoretical physicists last year designed a quantum algorithm that solves this problem in logarithmic time rather than polynomial, a significant improvement.

Now, a Chinese team has successfully implemented this artificial intelligence algorithm on a working quantum computer, for the first time. The information processor is a standard nuclear magnetic resonance quantum computer capable of handling 4 qubits. The team trained it to recognize the difference between the characters '6' and '9' and then asked it to classify a set of handwritten 6s and 9s accordingly, which it did successfully. The team says this is the first time that this kind of artificial intelligence has ever been demonstrated on a quantum computer and opens the way to the more rapid processing of other big data sets — provided, of course, that physicists can build more powerful quantum computers.

Comments

Captchas!

[Lanforod]

Crap. now what are we going to do instead of using a captcha?!

Re:Captchas!

[Jane Q. Public]

The phrase "artificial intelligence" does seem to get thrown around just a bit too freely these days. I don't see anything "artificial intelligence" about this at all. It's just an image recognition algorithm.

There is no "working AI" at this time

[gweihir]

And also no working quantum computer, except for very small toys. Pattern recognition is not AI.

Re:There is no "working AI" at this time

[QilessQi]

You're right that the wording is overblown, but AI is a big field, and pattern recognition is a big part of it -- vision, voice recognition, decision making, and other facets of human intelligence all rely on automated categorization of inputs to some degree.

Getting a tiny piece of the puzzle to work in a test tube is a necessary first step to bigger and better things. No one is going to put together a working brain in one shot (if ever).

Re:There is no "working AI" at this time

[Impy the Impiuos Imp]

Pattern recognition, decision theory, game theory, and partitioning are AI subjects. AI isn't just the mysterious general-purpose thinking machine always on the horizon.

Some pattern recognition uses neural networks for training.

Re:There is no "working AI" at this time

[CaptainDork]

And it's a fucking misnomer.

Artificial implies not real and intelligence implies thinking.

Artificial intelligence is so unattainable that the original definition has evolved to something meaningless.

True artificial intelligence is when a computer becomes depressed because it lost its connection to Facebook.

Re:There is no "working AI" at this time

[captjc]

Of course it is AI, not cutting edge AI, but it is still AI. Just because it is now a mature solved problem, doesn't make it any less valid.

It is like saying that someone doing the old "calculate the landing position of a cannonball fired at X velocity at Y angle" problem isn't doing physics because modern physics now involves super-tiny particles and / or traveling at speeds near the speed of light.

Re:There is no "working AI" at this time

[i kan reed]

The way ya'll keep equating humanness to intelligence, even as a joke, is a really stupid thing.

Our tests, we humans use on each other to determine intelligence like IQ or GI tests? They aren't testing our humanity, our empathy, our emotionality, our drive, our neuroticism. They're testing, get this, our pattern recognition.

The exact thing the OP was whining about being called AI.

Re:There is no "working AI" at this time

[Immerman]

>I have no idea of the formal definition of AI

You seem to be thinking of "Strong AI" - which is an actual thinking machine and the potentially immensely dangerous holy grail of AI research. All the various components - pattern recognition, descision-tree analysis, etc constitute Weak AI - basically everything that we can do on "autopilot" without conscious intervention.

And incidentally there's a growing body of evidence that our own brains may be composed of a large number of complexly interacting "weak intelligence" modules. For example theres a small area that appears to be dedicated just to face recognition - damage it and cognition is apparently unaffected but you can no longer recognize faces. Stimulate it and strangers faces seem to shift and look like someone you almost know.

Re:There is no "working AI" at this time

[Immerman]

>Artificial implies not real and intelligence implies thinking.

No. Lets consult the dictionary shall we:
   

Artificial: made or produced by human beings rather than occurring naturally, typically as a copy of something natural.

Comes from the same root as "artifact" - a made thing.
So AI literally means a human-made intelligence, as opposed to a naturally grown one.

Now intelligence is a much more slippery term that has never been well-defined, and no it doesn't necessarily imply thinking in any sort of conscious sense. Let's see what wikipedia has to say:

Intelligence has been defined in many different ways such as in terms of one's capacity for logic, abstract thought, understanding, self-awareness, communication, learning, emotional knowledge, memory, planning, creativity and problem solving. It can also be more generally described as the ability to perceive and/or retain knowledge or information and apply it to itself or other instances of knowledge or information creating referable understanding models of any size, density, or complexity

I'd say that large umbrella covers an awful lot of the various domain-specific weak-AI research, wouldn't you?

Re:There is no "working AI" at this time

[Half-pint HAL]

But identifying the difference between a '6' and a '9'? I agree that this is 'AI' as much as me heating something in the microwave makes me a chef.

This isn't 'AI' as far as I'm concerned. It's neat, it's cool. But it aint AI.

You're forgetting one important factor: they did it in a quantum computer. Do you know how difficult those things are to build? Do you appreciate that this makes them expensive? And can you see how this would mean that all the quantum computers in existence are very very small in terms of component numbers compared to computers that work within the bounds of Newtonian physics?

The machine they used has 4 quantum bits. 4 quantum bits! That really is very little computing power. And with that they did a non-negligible task.

But the important thing isn't that this was a breakthrough in AI research, it was that quantum computing reduced the task from polynomial time to logarithmic. I think the summary calling this "a significant improvement" is a bit of an understatement.

Re:There is no "working AI" at this time

[Half-pint HAL]

Listen, asshole: I started this digital shit back when Moby Dick was a minnow and I've watched the wilting of the definition of AI over the years.

I didn't write the definition I cited, right?

True, but you also didn't highlight the first definition, the definition the dictionary compilers thought was more important: "a branch of computer science dealing with the simulation of intelligent behavior in computers."

Visual pattern recognition is intelligent behaviour. Unless your definition of intelligence is predicated exclusively on higher-order reasoning and free will.

A small vat of organic liquid?

[QilessQi]

I read this:

Their quantum computing machine consists of a small vat of the organic liquid carbon-13-iodotrifluroethylene, a molecule consisting of two carbon atoms attached to three fluorine atoms and one iodine atom. Crucially, one of the carbon atoms is a carbon-13 isotope.

And immediately thought of this:

The principle of generating small amounts of finite improbability by simply hooking the logic circuits of a Bambleweeny 57 Sub-Meson Brain to an atomic vector plotter suspended in a strong Brownian Motion producer (say a nice hot cup of tea) were of course well understood ...

God, I love how weird the future is.

Read the Paper, article is exagerating "Quahtum"

[Grantbridge]

What they actually did if you read the paper is:

1) Encode a 6 or 9 image into 2 numbers, based on the number of excess pixels in the left vs right, and top vs bottom quadrants. From the article: After these preprocessing, the two printed image with standard font can be represented by ~x1= (0:9872;0:1595) for character "6" and ~x2= (0:3544;0:9351) for character "9"

2) Use a training algorithm to find the appropriate pulse sequence to give a up result from the molecule's NMR C13 spectra from a 6, and a down signal from a 9.

3) Run the NMR spectrum, feed in pulses based on the parameters produced from pixels encoded in a vector form like 1), get the result of "up" for a 6 and "down" for a 9.

It's certainly neat experimental NMR work, but I don't really see how it's quantum computing. But then maybe that's the NMR spectroscopist in me talking....

The most important computing result of our time

[Scotland]

Ho.

Ly.

Shit.

15 or 20 years ago, I was saying that because quantum computers perform multiple calculations on similar inputs simultaneously, they'll be perfect for the sorts of pattern recognition tasks needed for (artificial) intelligence. And now these smart people have figured out how to do it for the first time, albeit with a miniscule 4 qubit quantum computer.

But since quantum computing capabilities scale according to 2^n, where n is the number of qubits, a 24 qubit computer (i.e. 6 times the size of what they just built, requiring a molecule with 24 atoms) would be 2^20 = 1 million times as powerful as this 4 qubit computer just demonstrated. A 64 qubit computer would be 10^18 = 1 million million million times as powerful as this 4 qubit computer. Good-bye conventional computer encryption. And hello general-purpose pattern-recognition (i.e. the basis for strong artificial intelligence).

My first thought was that a vat of "carbon-13-iodotrifluroethylene" isn't exactly a general purpose computing device -- except that because their control inputs are a stream of radio waves pulses controlled by a conventional computer, it actually is a general purpose computer. And though I'm no quantum physicist / quantum computer scientist, it seems like it would scale reasonably easily: you just need to find larger organic molecules with similarly discrete nuclear magnetic resonance 'channels' (i.e. independently manipulable/separable by frequency).

I am beginning to sense the coming Kurzweil Singularity...