"The question of whether a computer can think is no more interesting than the question of whether a submarine can swim." - Edsger Dijkstra
One can imagine a long, heated and intricate debate among fish philosophers (in schools of philosophy), on the question of whether it could ever be possible for a submarine to truly swim. While some animals (for example humans and jellyfish) can achieve a certain amount of propulsion through water, fish-philosophers have long believed that only a fish can move with the power, speed and agility underwater that constitutes true swimming. However, recent engineering advances in Artificial Swimming (AS) have produced devices that have extraordinary swimming power by some measures. Some of them can even outpace fish over longer distances. Most fish-philosophers, however, are dismissive of AS. They say:
"Why, these 'submarines' aren't swimming! They don't even have proper beating tails! They are just hunks of metal with tiny revolving tails and minuscule fins at the back! They don't swim like we swim."
Swimming competitions are held. The submarines win. The fish-philosophers simply laugh and say:
"Ok, so you can beat us at swimming in a straight line, but what about swimming together in a shoal while being chased by a predator?"
As advances in AS encroach further into the definition of swimming that fish-philosophers hold; new, tougher, tests are introduced, such as the Tuna test, in which a submarine has to convince a fish, while swimming alongside it, that it is actually another fish. After a half century of work in AS, simple, cheap, readily available submarines can defeat a fish in almost every test of swimming that was traditionally used in swimming competitions between fish; yet still, fish philosophers refuse to accept that a submarine really has that ephemeral, fishy-swimming quality that makes the way fish swim distinct from how machines swim. Some fish philosophers even maintain that it would be possible for an organism to exist that was physically identical to a fish, behaved just like a fish, moved through the water exactly like a fish, but was still incapable of true swimming. This idea had previously only been suggested in low-budget fish horror movies.
One can imagine endless debates among birds about featherism. Elephants using artificial trunks. Arachnid science-fiction writers concerned that spider-engineers will construct web-spinning machines so advanced that they turn against their creators and catch all their flies. When a skill is considered to be the defining characteristic of a species, attempts to artificially engineer a machine to emulate this skill will be considered an affront to the very existence of the species. So it is with humans and thought.
This hubris must end. Machines think. Machines learn. Machines are intelligent. There does not exist a computer that thinks exactly as I do, but in terms of the ability to "think", as that word has been used historically, modern computer systems think better than modern humans. The objective of this blog will be to discuss current research areas in artificial intelligence in which computers outperform humans.
5 comments:
Great post, and can also be applied to our fellow primates and dolphins.
Sure, they can converse with us, make works of art, show signs of emotion, solve problems we set before them and display memory and self-awareness, but they don't have that special quality that God only gave to Us, because we are special and Unique, say the human philosophers.
but the machines could never feel like a fish
'think' is too generic, but you make a good point.
there a number of things: calculus, sorting; that a human can do as a by product of their 'thinking.'
your interpretation of djikstra's quote is much differetn than my own.
the part that is uninteresting is whether or not you call it 'thinking' rather than something else.
a submarine moves through water, a computer carries out instructions which can include solving the types of problems that require humans to think to solve.
Swimming and "thinking" both have very narrow definitions. I accept that a chess computer can think, and can learn, and can behave intelligently. But that's only for chess.
Do you think AI research will benefit from these narrow-goal projects, or from a more general approach?
Hello there, I'm currently writing a book for Penguin that covers some of these subjects and was wondering if I could quote some of this blog post - it's a lovely way to frame the question. Do e-mail me at mark@optimistontour.com. At the moment I'd have to attribute it to 'blogger Scott'!
Do get in touch - and tell me who you are!
Post a Comment