Science and Technology Intelligence testing Who are you calling bird-brained?
科技 智力测试 叫谁是笨蛋呢？
An attempt is being made to devise a universal intelligence test
WHAT is the IQ of a chimpanzee? Or a worm? Or a game-show-winning computer program? Or even an alien from the planet Zaarg who can learn any human language in a day, can beat grandmasters ten at a time and can instantly factor the products of large prime numbers?
At the moment it is impossible to say.
IQ tests depend on language, and even Watson, a computer program that beat two human contestants in a special edition of "Jeopardy!" (an American quiz show) on February 16th, does not have a perfect command of English.
In any case there is, at the moment, no meaningful scale on which non-human intelligence can be compared with the human sort.
The most famous test for artificial intelligence is that devised by Alan Turing, a British computing pioneer.
To pass the Turing test, and thus be considered intelligent, a program must fool a human being into believing that it is another human being.
But the Turing test still requires the program to share a language with the tester and, because it is all or nothing, cannot be used to rank different forms of artificial intelligence against one another.
Jose Hernandez-Orallo of the Polytechnic University of Valencia, in Spain, and David Dowe of Monash University, in Australia, think they can do better than this. They believe not only that a universal scale of intelligence can be devised, but also that it can be assessed without reference to language.If they are right, an insult like "bird-brained" will, in the future, be finely calibrated.
来自西班牙瓦伦西亚理工大学的Jose Hernandez-Orallo和澳大利亚莫纳什大学的David Dowe认为他们能设计出比图灵更完善的测试。他们相信不仅能设计出可通用的智商标尺，而且测试可以与语言无关。如果真能如他们所说，那么将来就可以精确计算出"愚蠢"的程度了。
Dr Hernandez-Orallo and Dr Dowe, both computer scientists, propose to make their measurement by borrowing a concept called Kolmogorov complexity from information theory, a branch of computer science.
The Kolmogorov complexity of a computer's output is the shortest possible program (measured in the binary digits that lie at the bottom of all computer code) that could produce that output.
On this measure, an entity's intelligence would be measured by the Kolmogorov complexity of the most complex tests it can solve—a clear, numerical value.
In practice, calculating the true Kolmogorov complexity of a system is almost impossible.
But an approximation can be made. And that, the researchers reckon, will be good enough.
The actual tests would employ the well-honed methods of operant conditioning, developed initially on pigeons, in which the test subject has first to work out what is going on by trial and error.
As in operant conditioning, correct responses would be ewarded—by money, perhaps, for a human being; by bananas for a chimpanzee or by the numerical value itself for an appropriately programmed computer.
If the test were noughts and crosses, the test-taker (if it had never seen the game before) would first have to work out that the game is won by getting three in a row on a 3-by-3 grid, before actually going on to play.
A chimpanzee might not manage a test of this level of complexity, but could, maybe, work out the idea of three-in-a-row when only a single row was involved.
Chess, though, would surely be beyond it (and probably beyond most people, too, if they did not already know the rules).
Games like draughts and dominoes would lie somewhere in between.
In fact Dr Hernandez-Orallo and Dr Dowe do not plan to use existing games. Instead they are employing a computer to enerate novel games and patterns.
Their approach eliminates human bias.
It also allows them to generate tests with any level of complexity they like—even ones that are far beyond the ability of humans to complete.
When it comes to testing the tests, then, aliens from the planet Zaarg will be particularly welcome to apply.