Virtual Intelligence Is Dangerous

“AI will be either the best or the worst thing, ever to happen to humanity”.

Said Stephen Hawking when asked about his opinion on Artificial Intelligence.

A few months earlier, the greatest South Korean GO player, Lee Sedol was being challenged by Google’s artificial player Alpha GO. GO game is considered to be the toughest game in the world. We can play our first move by 20 different choices in the game chess, while in GO game first move can be played by 361 different ways. After the initial one or two moves, the game becomes more and more complicated.

Lee Sedol had got the status of the professional player in GO game at 12 years of age. He nearly won the 18th international world championship and had became a South Korean superstar at a young age.

So the game played in South Korea from 9th March to 14th March, 2016. 60 million users from China and 25 million users from Japan were watching it live. In South Korea, it was a festive atmosphere as people expected Lee to beat the bot.  However, South Korean hearts broke after the results came out.

The famous star of GO game, 33 Years old, Lee Sedol lost by 4-1 against the Alpha GO!

South Korea mourned but it also brought forward the fact that Human intelligence would slowly be overshadowed by AI.

However, the fact was that Alpha Go was only calculating way ahead of its counterpart. There was factually no creativity involved in the game of GO.

Largely, the past four decades of AI has focused on ever more sophisticated methods for solving ever more highly constrained problems (e.g. chess, Go, memorizing labeled data-sets like Imagenet, or constrained quiz tasks like Jeopardy).

The field has unfortunately entered a downward spiral where publications are often judged by how well a given method performs on a particular artificial dataset, compared to 20 past methods on the same dataset. This approach of relying on artificial datasets to measure progress can quickly stifle creativity, and I see rampant evidence of this decline at even the best ML/AI conferences, like NIPS or AAAI, where year and year, papers that are accepted are largely highly incremental advances on previous work.

Very novel ideas have little chance of success, because they are usually unable to “play the same game” of showing marginal improvement on MNIST, or Imagenet, or COCO, or one of the dozens of other artificial datasets. It is as if physicists judge their profession by seeing how fast a car they can build with the latest advances in quantum field theory.

Creativity is an ability closely tied to “imagination”. The emphasis in creativity and imagination is not problem-solving at the expert level, but rather “problem creation”, if you will. It is a way of stretching the boundaries of what is possible by being able to ask counterfactual questions. Einstein was a great believer in the power of imagination.

Imagination is what led him to develop the theory of relativity, because he could ask questions like “What would the world look like if I rode a beam of light?” Imagination, he said, “would get you anywhere”, whereas “logic will only get you from A to B”. It is hard to imagine how one can do world class physics these days without a healthy dose of imagination. It is highly likely that this year’s Nobel prize in physics will go to the leaders of the LIGO detectors, which detected Einstein’s gravitational waves, a 100 years after they were predicted. The latest report of detection comes from two black holes that collided 1.8 billion light years away, releasing more energy in this one event than the energy released from all the stars in the observable universe. How can one even begin to understand the power of such events, without using imagination, since it is so far removed from our everyday experience

There is strong evidence that imagination is unique to humans as it is strongly localized in the frontal lobe of the brain, a structure most developed in humans as compared to other animals. Humans with damage to the frontal lobe are largely normal, although they are strikingly “in the present”, and unable to imagine the future. If you ask such a person what their plans are for the next week, they will understand the question, but say that their mind is a complete blank when they try to think of the future. Imagination is largely tied to the processes that go in the frontal lobe, and it is probably also the “seat of creativity”.

Fundamental advances are needed to understand how imagination works, and it will take at least the better part of the next decade or two before we begin to develop effective methods. One of our favorite examples of creativity is art. Jean Michel Basquiat’s untitled painting of a human skull sold at a New York auction recently for over $100 million. It is a strikingly original piece of art, and the 27 year old Brooklyn born painter was originally a graffiti artist, whose paintings now command prices similar to Van Gogh, Picasso, and Monet.

Will AI ever be able to produce great art of this caliber?

Perhaps, that day we should be bothered about the future of AI.