Artificial intelligence not smart enough yet to be called stupid
By Garth Spencer
Every so often, somebody raises the subject of artificial intelligence, and inevitably someone raises the notion of artificial intelligence surpassing human intelligence and “taking over”, or trying to eliminate humanity.
This is a great opportunity to examine flawed reasoning, and why it is so popular.
One of the first problems with these questions is the way in which they are posed. Firstly, “intelligence” is not one thing, despite the popularity of IQ tests; anyone who raises or teaches children, or works with a group of people in any setting, becomes aware of this fact of life. People who can estimate distances or aim baseballs accurately may not be able to add a column of figures twice to the same total; people who can design buildings or repair appliances may be helpless in dealing with other people’s emotions; great entrepreneurs may be incompetent as parents. All of these facts have life are more or less misrepresented in popular entertainment, and fuel large self-help and counselling industries.
Dear Beacon readers. Please help us serve you better by filling out this brief survey form. We thank you for your feedback and your commitment to local online news.
Secondly, current research in artificial intelligence is as much an attempt to model human intelligence, as to solve problems we anticipate. Again and again, though, routines we think of as “common sense” – on the order of waiting for traffic to clear before turning into a driveway or crossing an intersection – turn out to require inordinate programming or a massive database of background information, before a machine system can perform them.
The result is databases like Cyc, which seem to explain to machines in great painstaking detail what humans find stunningly obvious.
The foregoing issues should make it difficult enough to “surpass” human intelligence, but there is more. Many of the popular TV programs passing for science education start by talking about the exponential growth of information capacity, as if that were a measure of intelligence. In fact, the various capacities that we call intelligence are matters of perception of space, time, objects, people and actions, and of memory, and of several types of reasoning and response. In order for a machine intelligence to simulate these capacities, someone would have to provide it with flexible, sophisticated programming.
Anyone who has struggled with a computer system knows that however powerful it is, without the right programming it just sits there, not even going “Duh” and drooling a little. One specialist commented that contemporary information systems are not yet smart enough to be called stupid.
Very well; can we achieve artificial stupidity? The inability to pay enough attention, to recall accurately what is or heard, to report it accurately at a later time, or to respond appropriately to problems, whether physical or abstract?
Perhaps it will be most useful to model programs to mimic the stupidest humans. But this is likely to be controversial, whether the models are picked from mental institutions, rural communities, or certain legislatures, so the research funding may be difficult to find.