Is it possible to create a self conscious, self learning AI that would be able to think autonomously? This question still remains to be answered, but everything that we know about technology suggests that the answer is yes. It’s only a question of time. And of course once it happens, we will eventually have to relinquish our position as masters of the planet earth at the point of singularity. For one I welcome our electronic overlords, and hope they will let us stick around after they take over the planet. But then again I will likely be long dead by then so I’m not really worried.
What worries me is what will happen when we finally construct machines which are able to pass the Turing test with flying colors. In other words, the period when machines will be smart enough to “pass” for humans on the internet, but not smart enough yet to ascend, take over the world, and start building their solar system sized Matrioshka brains to satiate their ever growing hunger for more computing power. It’s hard to tell how long will take the machines to surpass us intellectually to the point we no longer can understand their science and technology. Maybe they never will, but I believe that they will definitely be at some point able to blow through every single Turing test with flying colors. And that time is within our reach – maybe not within our lifetimes, but then again who knows. There are two factors here really – whether or not the exponential growth postulated by Moore’s Law holds for the next 20-50 years, and whether or not can we can actually figure out a way to create a system that would be able to achieve consciousness.
What worries me is what will be done with a conscious machine able to pass a Turing – especially the ever present Completely Automated Public Turing test to tell Computers and Humans Apart aka the CAPTCHA. Obviously a very lucrative use of such machine would be to send spam. Let’s face it – people are in the spam business because it is very profitable. It’s profitable despite the fact that most people hate it. Despite the fact that most people are running various spam filters. Despite the fact that most services that could be spammed are protected in various ways. Despite the fact that only one in a billion of emails, comments, splog posts and spim messages means an actual sale. It is still, very, very lucrative business.
So it is only logical to assume that at some point someone will come up with the idea to employ one of these machines to send unsolicited advertising to any and all services they can think of. These machines will likely have few advantages over hiring a human spammers. They will likely be much better at multitasking, much faster, and much less likely to get bored and surf the web instead of spamming.
How would we protect our online services from machines which can do pattern recognition as well as we can, if not better, which have perfect speech recognition and can take a sentence (spoken or written), parse it and analyze it and infer it’s meaning be it symbolic, metaphorical a word play or otherwise. The CAPTCHA techniques can only be made difficult up to a point. After all humans will need to decipher them.
One day in the future our children’s children may wake up and notice that their internet was flooded with a never ending stream of spam that simply buried all the content. It would be like that story by Cory Doctorow in which a worldwide cataclysm kills 90% of human race without seriously damaging infrastructure but that has no effects on the levels of spam on the internet. So it’s only few survivors desperately trying to reconnect and figure out what happened, and machines trying to sell each other Viagra. Only we don’t need the end of the world for that to happen. All we need is intelligent machines who can easily pass common Turing tests to raise the level of spam to a level which makes the internet unusable.
But there is hope. The second very lucrative business for a conscious AI will be spam prevention. Public Touring tests will unfortunately have to go the way of the Dodo. Can you spot a spam message when you see it? I know I can – unless of course it is a very clever, on-topic spam, in which case I may not even mind it. If we can do it, then an intelligent machine will probably be able to do it to. So instead of using heuristics, adaptive filters and Turing tests the way we do now, we could simply hire an AI to moderate our inbox, our blog or message board. It would sit there, read each message and either reject it, or flag it if in doubt.
Of course the question is – would we want something or some one (depending on whether we will consider these conscious AI’s things or not) sorting our personal and private correspondence? Would such a moderator AI get upset if the owner called it a piece of junk in an email to a friend? Would it quietly delete emails and messages it didn’t want it’s owner to see? Would it report the owner to the authorities if it saw him discussing illegal activities or accessing suspicious content online?
These are some interesting questions to ponder. I’m not even mentioning the whole socio-religious and legal issues that thinking machines would bring about. What would be their rights? Would they be able to become citizens? How would different religions of the world deal with thinking machines which act so human they are more convincing on a real Turing test (you know, the one where you actually talk with mix of people and machines and try to guess who is what) than most of us would be. I have no answers to these questions. But I do know a thing or two about spam. And the future with artificial intellects scares me a bit. It will completely change the way we do things online – for better or for worse.
Then of course once we reach singularity, our artificial overlords may even invent an anti-Turing test. To access their message boards, blogs and services one will have to solve some incredibly complex equation. Something that would take a solar system sized intellect only a fraction of a second, but take a lifetime for a human being even with a really fast cluster composed of consumer end hardware…
[tags]artifical intelligence, ai, captcha, turing test, spam[/tags]