Alipr’s new picture based CAPTCHA system has pronounced me “probably human”. Here is a link to try out their new system. It looks very promising. There is an interesting article about it here. I had written about the recent compromise of existing popular CAPTCHA systems here.
It would be ironic if the war between the service providers and the spammers provides us some of the biggest advanced in AI in years.
It wouldn’t surprise me though. When I was doing AI research in the 80s a lot of what was being done had a definite “solutions looking for a problem” feel to it. What we are seeing now is very different. There is a very specific problem that is trying to be solved and there are very clear financial rewards. The spammers are trying to impersonate real people. The service providers know that spam bots in their system would clog it up and chase away the real, advertising revenue generating, people that want more of.
War is often a technology driver.
Not only are we not making significant progress on the Turing Test, we are losing ground on the Reverse Turning Test.
The ramifications of this are not pleasant to consider.
In 1950 Alan Turing described what became later known as the Turing Test. The Turing Test, which is a test of one aspect of Artificial Intelligence, involved a person sitting in front of a terminal exchanging in a typed conversation with another party. If the person could not determine if the other party was a person or a bit of software, then that software would have passed the Turing Test.
Of course almost everyone one the web has by now participated in a Reverse Turing Test where a computer tries to distinguish between a human and a computer (the T in CAPTCHA stands for Turing). But AI researchers over the years have failed to create software that can pass the standard Turning Test.
Who knew that the whole approach has been wrong? Instead of grad students and professors toiling away in major research institutions, the problem might instead be solved by Russian hackers looking to rip people off? If this doesn’t describe a piece of software that passes the Turing test, it’s closer than anything else I have heard of so far. Graft is apparent for more effective than research grants.
There are three things makes this really interesting. First, the environment (chat rooms) exactly matches the Teletype mechanism Turing first proposed. Second, the “test subjects” don’t actually know they are participating in a test. Ironically the software developers don’t either. Third, it’s illegal. So if the developers have actually cracked the Turning Test, they will never receive any recognition for it.
(Mirrored from TalkBMC)