In the virtual world of the Internet, identities often get blurred. Senior citizens easily pass themselves for teenagers, and timid geeks, normally hiding behind their glasses, can take on the roles of seductive young men and women. If you are not careful enough when chatting with strangers, you might soon find out that the charming young person you have shared so many good talks with is, in fact, none other than your annoying neighbor.
Robots simulating people
The development of the Internet quickly revealed another problem concerning identification, one that can almost be considered a philosophical issue. Not only can people virtually take on other identities on the Web than they possess in real life, computers themselves can impersonate people. Today, this most certainly is not a thing of science fiction as these identity swaps occur on a regular basis.
Robots or, to be precise, programs portraying other people online, sign in to all kinds of Web forums where they post advertisements for their masters or creators. The smarter machines even know how to create their own free e-mail accounts which they use to flood other accounts with advertisements every day.
Robots roaming the Internet and impersonating people first became a serious problem around the year 2000. Administrators of larger websites realized they needed to invent an effective method of distinguishing real people from robots as soon as possible. The task seemed simple enough at first, but it proved to be a rather tough nut to crack when programmers took it on . Should websites have employed people to exchange a few words with every single visitor wishing to sign in? This struck the problem-solvers as an inefficient course of action. A more effective solution was needed.
The job description of a person hired to catch Web-surfing robots would not resemble that of officer Rick Deckard (Harrison Ford) in Blade Runner, the legendary maverick who chases runaway cybernetic “replicants” who look and act just like people. Yet a comparison is valid up to a point. The movie implies that officer Deckard himself is a replicant, a computer designed to hunt down other computers. Programmers followed the same principle in designing software which could differentiate between “replicant” and human Internet users. They also sought a method of exterminating these troublesome artificial creatures wreaking havoc on the Web.
The Turing test
Soon after the Second World War, Alan Turing, an English mathematician whose work was integral in the early development of computers, contemplated the possibility of computer thought. He quickly found out this was much too abstract a question, so he shifted to a simpler and more tangible line of inquiry. He started to develop a practical method of differentiating between artificial and human intelligence. How could one determine whether a machine can think? One of the most obvious solutions is to talk to the machine. If it is impossible to tell the difference between its responses and those given by a human being, then there simply is no more difference between artificial and human intelligence.
The Turing test, as it was later named, is, simply put, a method of telling a man apart from a machine. In separate rooms, a robotic and a human subject are posed questions they are obliged answer. Their interrogator, meanwhile, has no notion about which test subject is in which room. If an interview ends with the questioner unable to determine which room contains the living person, the other entity has successfully passed the test and, according to the Turing criterion, cannot be set apart from a real human being.
The Turing test thus offers a way of finding out, through a series of questions and answers with two subjects via a keyboard, if we are conversing with a man or a machine. The machine passes the test when the assessor cannot tell whether he is talking to a man or a machine. The problem for programmers was to design a robotic intelligence which could put the test to other robots.
What robots can’t do
Disturbing attacks by malicious robots forced the administrators of larger websites to seek help from experts specialized in artificial intelligence. They quickly concluded that the best solution for “exterminating” such robots would be to set a series of brief tasks that real people could solve in a matter of moments, but which would take computers or robots much longer to complete, if they could do so at all.
Setting mathematical questions was not an option, since computers perform even simple calculations much faster than humans Computers do however have much greater difficulty in recognizing images and symbols. Here human users have a clear advantage: for in the course of several million years of evolution we have become extremely adept at visual recognition, even when the form we recognize is distorted from its usual shape or colored differently.
This is not to say that modern computers are hopeless at spotting an image or symbol they have “seen” before. They can in fact identify letters of the alphabet by a rudimentary scan. This is how they are able to convert the copy of a printed text back into a file which can then be edited and changed anew. However, such recognition is only successful when the image of the text is sufficiently clear and legible. If letters are smudged, text-recognition programs are usually lost. They can only see indistinct blotches where people can still read damaged or smudged writing.
The Automated Turing test
This was the disadvantage engineers sought to exploit in creating a robot which could differentiate between artificial and human intelligence. Humans might wince a little when presented with scrappy handwriting; but computers are brought to a standstill by it. Researchers at Carnegie Melon University therefore developed an automated Turing test requiring no human guidance. They named it CAPTCHA: the Completely Automated Public Turing Test to Tell Computers and Humans Apart.
In simple terms, CAPTCHA is a program that can compose and evaluate a test which it could not actually solve for itself. It is a program for exterminating programs. It gives users a simple visual task that usually requires them to spell out a series of smudged letters and numbers shown on the screen. On the basis of the answer given, the program can tell whether visitors to a website are human or electronic.
Every Internet user has probably encountered such a program at some point. We have to pass its test in order to register on an Internet forum, to create a new e-mail account or to file an income tax return. With time, of course, artificial Web-surfers have become much better at reading unclear print, and programs for exterminating programs must be constantly improved. A delicate balance must be struck here: CAPTCHA must not make the letters in its tests so hard to make out that they baffle humans and computers alike.
Which is the best school for computer science?
In November 1999, a good while before the first online robotic robot-catchers were developed, the website slashdot.com conducted a survey to find out which US university offered the best course in computer science. This was obviously quite a delicate question, especially when posted on the Internet.
To make voting as fair and free as possible, each computer was given one vote. The program that registered the votes also registered the Internet addresses (IP addresses) of individual computers that had already cast their vote, allowing each individual to vote only once. However, students from Carnegie Mellon University found a way to bypass this safeguard and designed a program which could cast thousands of votes for their university. By the next day, MIT students had written their own voting program; and so began a battle between the two institutions to generate the greatest possible number of non-existent and fraudulent votes. When the poll closed, MIT snatched victory by 21.156 votes and Carnegie Mellon took second place with 21.032; while all of the other candidates languished far behind with less than 1000 votes each.
Certainly the students of the winning school and the runner-up made a strong case for the quality of their respective faculties. When it comes to deciding on the best school for computer science it might even be thought appropriate to grant computers an equal right to vote; although only, perhaps, if they could pass a Turing Test first. In the case of other surveys, however, one would probably be more comfortable knowing that each registered vote represents a human being.