19. THE AI EFFECT: MOVING THE GOALPOSTS OF INTELLIGENCE

24/11/2020

When it comes to measuring intelligence, Stephen Hawking's opinion was that "people who boast about their I.Q. are losers" - yet you'd be hard-pressed to find an interview which doesn't identify him as a genius or other similar label. In recent decades, conceptions of intelligence have come to encompass diverse but connected elements such as emotional knowledge, creativity and practical problem-solving. If a single definition for humans is so difficult, what hope do we have for machines?

Man vs machine

Debatably, Artificial Intelligence had its biggest moment in the spotlight when IBM's chess-playing computer Deep Blue defeated Garry Kasparov in a six-game match, becoming the first machine to best a world champion. Kasparov began with an unusual opening in an attempt to get Deep Blue out of its comfort zone. Deep Blue, with its ability to evaluate 200 million positions per second, was not phased.

Deep Blue is notable not simply for drawing the public's attention to machine intelligence but also as a clear example of 'the AI effect': the tendency for humans to move the goalposts by determining that a machine's behaviour is 'just' computing, even if it had been deemed real intelligence prior to achievement. Until 1997, chess was considered an excellent measure of AI given its 69 trillion possible games within the first five moves. Deep Blue, however, was quickly decried for having won on 'brute force' alone and accusations were even made that IBM had secretly pitted Kasparov against another human.

J'ai mangé un chien chaud

The AI effect may be explained in part by the fact that computational programmes which realise a practical goal are quickly co-opted into whichever domains can benefit from them. In the process, they become 'normal' and cease to be considered intelligent. Take translation as an example. 1997 also saw the release of the first online tool, Babel Fish, which could translate small blocks of text on a largely literal, word-to-word basis. I recall using it (or a similar tool) to translate my French homework in secondary school. I also recall my teacher's particular irritation at what it did to the sentence 'I ate a hot dog'.

I'm sure that most people who used Babel Fish would have been blown away by the idea of machines effectively translating thousands of words in everything from Albanian to Zulu, often (but certainly not always) capturing subtext and metaphor. We rarely think twice about the fact that such tools now exist. The current state-of-the-art even allows for real-time speech translation with 85% accuracy and has been included on all Android phones since 2019 without causing much of a fuss. In other words, "AI has become more important as it has become less conspicuous", as MIT's Patrick Winston put it. In every domain which uses complex computer systems - healthcare, manufacturing and energy, to name a few - digital breakthroughs have almost always been the direct result of work in the field of AI.

Humans on top

If the public at large knew more about this, would AI be considered truly intelligent? A slightly more cynical interpretation of the AI effect suggests that we would still reserve intelligence as a uniquely human trait because of a fear of being displaced at the top of the food chain. As long ago as 1863, a New Zealand newspaper published an article asserting that "the time will come when the machines will hold the real supremacy over the world and its inhabitants [...] Our opinion is that war to the death should be instantly proclaimed against them." A violent uprising of the inventions of that time - including typewriters, sewing machines and doorbells - has yet to materialise.

Nonetheless, one could point to how industrial revolutions have tended to increase unemployment rather than reduce it, as AI-based automation may do on a large scale in coming decades. Personally, I hope that the AI effect points to a middle ground in which breakthroughs are absorbed into existing processes, removing the more monotonous aspects while creating new avenues for humans. After all, chess computers overcame fears that they would destroy the game through a 'draw death' and are now used to prepare and analyse a much higher level of complexity in human play. It remains to be seen whether it's too optimistic (or naïve?) to imagine such unexpected outcomes in industry.

- Josh