‘Artificial Intelligence’ has to be better than the real thing
Machines will achieve human-level artificial intelligence by 2029, a leading US inventor has predicted.
Thus begins a BBC News report about the thoughts of engineer Ray Kurzweil.
Kurzweil has been brought together with Larry Page, Dr. J. Craig Venter, and 15 others to identify the great technological challenges facing humanity in the 21st century. So he’s obviously pretty smart. So smart, in fact, that he may have overlooked the obvious: in seeking to replicate human intelligence, AI might be setting its sights too low.
Humans are brilliant, creative, vibrant creatures. We are also ignoramuses. In fact, it appears we are absolutely ignorant about how ignorant we are.
Following my piece about the Exploratorium last week, Brian Hayes proffered a couple of tasty posts for dessert. The first, from PhysOrg.com, describes a study that shows our propensity to fill in the blanks when the complete picture isn’t available to us:
Eighteen observers were asked to concentrate on the centre of a black computer screen. Every time a buzzer sounded they pressed one of two buttons to record whether or not they had just seen a small, dim, grey ?target? rectangle in the middle of the screen. It did not appear every time, but when it did appear it was displayed for just 80 milliseconds (80 one thousandths of a second).
?People saw the target much more often if it appeared in the middle of a vertical line of similar looking, grey rectangles, compared to when it appeared in the middle of a pattern of bright, white rectangles. They even registered ?seeing? the target when it wasn?t actually there,? said Professor Zhaoping, lead author of the paper. ?This is because people are mentally better prepared to see something vague when the surrounding context is also vague. It made sense for them to see it ? so that?s what happened. When the target didn?t match the expectations set by the surrounding context, they saw it much less often.
In other words, we see what we expect to see, as dictated by the surrounding context.
It isn’t often I’m required to identify the brief appearance of a target rectangle, but this phenomenon does have more real-world applications. Just yesterday I was in a cafe with a girlfriend when I spotted someone walking towards the door.
It wasn’t even close to our friend Lynne?but the face, which had been the first thing I had seen, was. I saw the face and thought, “There’s Lynne.” Then I thought, “She’s not wearing the right clothes. Her hair is wrong. The body shape is different.” But those things were contradictory to my initial conclusion, and I discarded them immediately.
“There’s Lynne,” I said to the woman I was with.
“That isn’t Lynne,” she said.
“Of course it isn’t,” I replied, stunned that I could have perceived all of the many non-Lynne characteristics displayed by the stranger and yet still insist that it was she.
The other piece Brian sent in was about ’split brain’ people, and the fact that our use of logic is usually employed after the fact as a means of retroactively understanding our behavior. This behavior is more easily seen in people whose left and right hemispheres are disconnected (a ‘callosal disconnection’), although the authors make it clear that we all do it:
…the right hemisphere, upon seeing an image with strong emotional connotations, generates the appropriate response. However, due to the callosal disconnection, it cannot transmit the associated sensory data to the left hemisphere and its language centers. The left hemisphere perceives a change in the body’s state, but does not know why - and so it “fills in” the missing details, fabricating a logical reason for the emotional reaction. This happens at a subconscious level, so that the person genuinely believes the verbal explanation they provide. In the language of psychology, this filling-in process of unconscious invention is called confabulation.
All of these ideas are utterly bizarre. On the one hand, we only ever perceive a microfraction of the potential data in our environment. On the other, our brains go to the trouble of adding in stuff that’s not there. The hemispheres of our brains are like Steven Wright’s humidifier and dehumidifier, left in a room to fight it out.
So are attempts at Artificial Intelligence aiming to replicate phenomena such as these? Will a smart machine be able to justify its actions? Or perhaps this will be the stumbling block limiting the development of AI: that we don’t even understand our own intelligence beyond the flimsy rationales we invent after the fact, and that we’re reverse engineering our behavior based on false assumptions.
Have you ever caught yourself making up stories to explain situations that otherwise don’t make sense? Or do you think your perceptions are absolutely genuine at all times?
