Machines ‘to match man by 2029’

Machines will achieve human-level artificial intelligence by 2029, a leading US inventor has predicted”, says BBC.

And I say: bullshit.

I’ve worked on an AI project straight out of college. It was my first real job with computers. And I am not talking about “expert learning systems” or other such boring academic stuff, but a project about real AI. The goal was to create an intelligent “being”. I quickly mentioned the project in my Feb 2001 editorial at the now defunct BeNews.

I can tell you one thing: we are far, far away from real AI. Even OpenCYC at this point is simply a collection of data without any real intelligence behind it. In fact, when it comes to AI, people somehow think that it will be ready in “20-30 years time”. They were kept saying that in 1970 too. And each year, it would slip one more year. Just like in some movies and TV series where they depict some really high technology and space exploration that supposedly takes places only 70 years in the future (e.g. “Space: Above & Beyond”).

Personally, I don’t think that human-equivalent intelligence will be ready before 2100. And I am probably optimistic already.


JBQ wrote on February 16th, 2008 at 9:56 PM PST:

I agree, making such predictions at this point is bullshit.

If the difficulty is purely an issue of processing power, 20 years is essentially a factor 1000 in processing power (if we assume that Moore’s law continues to hold for the next 20 years). By combining parallelism and non-real-time processing, it’s possible to give it a shot right now.

The question becomes: can we set today a rack of computers (84 machines if you pack them tight) capable of working at 1/12th the speed of a human brain? Let’s take an example: can we feed a 2-hour movie into a rack of computers (that can be pre-loaded with arbitrary information except anything about the movie), let them process the movie for 24 hours, and once that’s done ask questions and get answers in less than a minute?

If we can’t currently do that, that means that there are still some fundamental issues that haven’t been resolved, i.e. there’s still some research to be done, i.e. nobody can have any idea how long anything will take.

Brendan wrote on February 17th, 2008 at 1:58 AM PST:

The likes of Jimi Hendrix, Van Gogh or Oscar Wilde from a pc?

Never. Stupid article.

Ivan wrote on February 17th, 2008 at 2:24 AM PST:

I think it will ‘be’ suddenly, without anyone expecting it, and only years later, when AI has become a household thing, scientists will pin point the very first AI.
It could happen tomorrow, it could happen in 20 years time. It hasn’t really to do with processing power, but with the process itself.

This is the admin speaking...
Eugenia wrote on February 17th, 2008 at 2:29 AM PST:

Ivan, AI requires lots of knowledge in order to be deemed “intelligent”. And lots of knowledge means lots of data. Moving so much data around in an instant, and process them, is not something that can be done in a garage. I learned that much from my first job.

irbis wrote on February 17th, 2008 at 4:01 AM PST:

Simulating human thinking in the sense of “strong AI” is really, really difficult. Human mind is not just a calculator combined with a chatterbox. There’s also senses and other human biology, feelings, relationships, culture, complicated languages, human history, independent thinking abilities, values and goals etc. – all those things that may make people talk about human soul too instead of just intelligence.

Weaker AI in the sense of expert systems etc. does work, but expert systems have nothing to do with real thinking. Those computers and programs are just tools that people use, a bit more complicated calculators combined with human made databases of information collected by people.

I seriously doubt man-made machines could ever reach the level of strong AI. Even if they came a bit closer – theoretically – they would probably still depend on their human programmers so much that they could be described only as tools that people use and that depend on people. Movies or games may simulate things in life rather well – but they are still just movies and games, made by people.

A classic (and funny) long two part article “Artificial Stpidity” by John Sundman still shows how far we are from simulating human mind with computers. Read it, be educated, and have a few good laughs too:

Artificial stupidity (part 1)
The saga of Hugh Loebner and his search for an intelligent bot has almost everything: Sex, lawsuits and feuding computer scientists. There’s only one thing missing: Smart machines.

Artificial stupidity, Part 2
Can chatterbots be as dumb as a box of hammers and still pass the Turing test? Go ask ALICE, she might know.

I just found another good article on the same subject too:

The Dream of A.I.: There is no substitute for man by Derbyshire, John

This is the admin speaking...
Eugenia wrote on February 17th, 2008 at 4:06 AM PST:

Everyone, please use HTML for links (around normal words), do NOT copy/paste just bare URLs on my blog. Thanks. Irbis, I had to edit your comment.

irbis wrote on February 17th, 2008 at 4:42 AM PST:

Ok, Eugenia. Thanks for the hint.

As to human intelligence and thinking, I wonder if there’s any kind of real (human) intelligence without the understanding of language too? And to understand language, in an intelligent way, one also has to understand human culture, values etc.

Of course, also animals like dogs, parrots or even cuttlefish have some kind of ability to solve simple problems, without the need for real language. Maybe that kind of “dog” intelligence could be a better goal for AI researchers than trying to make machines understand human thinking?

Cesar wrote on February 17th, 2008 at 7:47 AM PST:

Evolutionary computation, spiking neuron models plus the advances in developmental biology is the way.

Just some notes:

(1) Evolution not in the sense of optimization (as it’s usually applied), but adaptation (Darwin’s theory is a simple and elegant algorithm);

(2) Solving the full Hodgkin-Huxley set of equations demands extreme computational power, but that won’t be a problem in the future. For the time being we can stick to simpler models and we’re well served;

(3) Yet we don’t know how to wire the network, and our best guess is: let artificial evolution alone take care of it;

(4) Embodiment: intelligence (i.e., adaptive behavior) is an emergent process (following the view of the New AI) and can’t come to existence without a body and its interactions with the environment;

(5) We must begin to study simple artificial organism before going to the more advanced ones (bottom-up approach).

These are just some points I’m currently working on.

Thom Holwerda wrote on February 17th, 2008 at 8:37 AM PST:

Intelligence implies unexpected behaviour. How do you program unexpected behaviour? Even if you find out how to do that – if unexpected behaviour is programmed, is it still unexpected?

It won’t happen. We will destroy our current civilisation before we can ever get that far.

M wrote on February 17th, 2008 at 2:40 PM PST:

I think it can be done, perhaps not in 2029 but someday. Not by using ai software but trying to imitate the brain in hardware. I think some team has emulated a half rat brain in 1/10 of real time using 4096 processors. A human mind requires a few more processors but that is only a question of time, 20, 50 or 100 years. Unless we humans disappears before, which is perhaps more probably.

jeff wrote on February 17th, 2008 at 7:26 PM PST:

2029, the date he used which matches the terminator 2 plot, isn’t viable. The inventor forgot that in the movie they also had time travel, which probably allowed accelerated advancement of technology.

Jim wrote on February 17th, 2008 at 8:40 PM PST:

I think you folks are not familiar a German research program known as the Super Human Intelligence Technological Enterprise.
They have perfected the science for AI, currently they are testing their discoveries on a German Dairy with a program known as the Bovine Super Human Intelligence Technological Enterprise
commonly known as Bull S.H.I.T.E.
S.H.I.T.E. has actually been around for a long time, but leave it up to German Science to get there first. Go Figure!

Vassilis Perantzakis wrote on February 18th, 2008 at 6:10 AM PST:

I don’t think we are going to see a mr.Data soon… The complexity of AI is so great that no hardware available today can cope even if the software existed. 2100 is too soon… I am betting middle 23rd century… (see Star Trek: Noonien Sung)

Comments are closed as this blog post is now archived.

Lines, paragraphs break automatically. HTML allowed: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

The URI to TrackBack this blog entry is this. And here is the RSS 2.0 for comments on this post.