Machines and models versus humans: Not a no-contest
Summary
- Downplaying the human mind’s value could doom us while using neuroscience for AI will help.
In 2016, the days from 6 March to 15 March were marked on the calendar for a much anticipated match-up between machine intelligence and human intelligence. AlphaGo, an artificial intelligence (AI) powered machine developed by DeepMind, took on Lee Sedol, the undisputed human champion of Go in a best-of-five-games contest.
AlphaGo easily won the first game. As the series progressed, as soon as AlphaGo made Move No. 37 in the second tie, watchers froze in bewilderment. Expert commentators shook their heads in disbelief. The machine had made an awful move. Even Go beginners knew that once you make that move, one can never win a game from there on. So, how could the machine make such a ‘stupid’ move? But, to everyone’s surprise, AlphaGo won the second game too. Only after the game was over, while retracing each move made by the machine, did the game’s experts decipher the brilliance behind what had seemed like a silly move.
Mechanophiliacs went berserk. One wrote, “It is now a mistaken belief that the only way to develop machines that perform a task, at the level of a human being, is to copy the way that human beings perform that task." Another said, “Machines are no longer riding on the coat-tails of human intelligence." In the third game of the match as well, AlphaGo comprehensively defeated its human opponent. With this, many proclaimed that the superiority of machine intelligence over human intelligence had been established once and for all.
In the fourth game, as AlphaGo made Move No. 77, commentators who had given it a 70% chance of winning did not seem to stir. Although Lee Sedol did not think very hard before making the 78th move, many saw it as a casual move made by someone who had given up on the match, having lost three games in a row. But as soon as the move was completed, those watching the match froze once again in bewilderment. Lee Sedol had made a move no human had ever made in a game of Go. Many called it the “God move." It was beyond the ability of AlphaGo’s algorithms to decipher. From a clear winning position, the machine proceeded to lose the match to its human rival.
This forms the context of a critical irony prevalent in today’s narrative about the world of machines, with AI on one hand ranged against human intelligence on the other. While the significance of AlphaGo’s 37th move in the second game has been widely discussed, very little has been written about the importance of Lee Sedol’s 78th move, the one that proved humans have it in them to generate their own transcendent moments. This prejudicial treatment meted out to human intelligence is not new.
Marvin Minsky and others led the wave of AI in the 1960s. They believed that logic and logical rules will help create an AI revolution. But this wave of AI did not get very far.
That is when professionals like Geoffrey Hinton and others stepped in with a new approach to AI. This was based on machines learning from large amounts of data. This data-based foundation of the ongoing AI revolution has been much discussed, of course. But there is an obvious omission in the AI narrative that appears to have gained prevalence globally: The fact that Hinton and others had taken much inspiration from the construct of the human brain while developing their new theories. This tendency to downplay anything that is human is an even older trend.
Almost all of us have heard of the 1776 book, The Wealth of Nations, by Adam Smith. Most classical economics and many mathematical models have clearly been influenced by this book. But how many of us have heard of—let alone read—Adam Smith’s other book, The Theory of Moral Sentiments? This was the work that he saw as one of his best works. Why did the world of economics come to ignore the role of emotions in economic decisions although Adam Smith himself laid such a profound emphasis on it?
There has been an evident historical inclination towards—and outright preference for—the assumed perfectness of mathematical models and the technological prowess of machines over the qualitative and subjective nature of human behaviour. The qualitative elements of human behaviour, like emotions, do not get adequate importance in the modern world of machines and mathematics. Downplaying the contents of Adam Smith’s The Theory of Moral Sentiments or the significance of Lee Sedol’s 78th move against AlphaGo in the fourth game is a form of intellectual dishonesty that is widely prevalent today.
The world of finance realized its folly of depending solely on the perfectness of mathematical models when it reached the brink of disaster in 2008. That was when the qualitative principles of Behavioural Economics finally entered mainstream economic discussions. The AI industry should not wait for a Great Recession-type crisis to hit before it recognizes what it stands to gain by holding human hands instead of trying to replace them.
The human brain is the most sophisticated and energy-efficient machine we know of. So a deeper understanding of its functioning can act as a validation— or better still, as an inspiration—for AI professionals. No wonder Demis Hassabis, founder of DeepMind, once said, “The more we understand the inscrutable algorithms between our ears, the better will be the technology we create for our machines." A combination of artificial intelligence and Neuroscience will brighten the future for us all.