We know things in a distinctly human way. That’s why so many of us are intrigued when we run into other beings that think in non-human ways.
That curiosity compels us to explore the consciousness of a cephalopod, an ape, or an alien heptapod, and even how your dog might experience the world. This particular moment in history is noteworthy because, for the first time, we are coming into contact with an intelligence that isn’t just non-human, it’s non-animal.
Ironically, this new intelligence thinks very differently from us even though it is our thinking that creates it.
The Alpha and Omega of Machine Intelligence
At no time in recent years have we been more captivated with this new machine intelligence than the response the world had to one shockingly elegant move in a game of Go. It was the 37th move in the second game against a world champion Go master, Lee Sedol. Lee’s opponent was a deep learning system called AlphaGo, developed by an Alphabet (Google’s holding company) subsidiary called DeepMind.
The below video links directly to DeepMind CEO, Demis Hassabis’ detailed explanation of why this move was so unusual. In short, AlphaGo ignored conventions and played ‘the fifth line,’ which is to say, it moved into a region of the board that made no sense at that point in the game. This single move, more so than any other in the match, symbolized AlphaGo turning centuries of received wisdom on its head.
In fact, the move so puzzled Lee that he stood up and left the match for fifteen minutes to collect himself. He had already lost one game to AlphaGo at that point and was clearly thrown by this seemingly amateur move. It wasn’t until many moves later that the move’s brilliance became apparent to the many Go experts watching the historic match.
For one of those experts, however, the move seemed to match a pattern. Six months earlier, another Go master named Fan Hui had faced AlphaGo himself and lost five games to zero. Since then, he had advised the AlphaGo team in training their algorithms. Through that experience, Fan had gained some sense for the unique way that the system played. When he saw move 37, he first thought it odd, but then came to see its beauty:
“It’s not a human move. I’ve never seen a human play this move,” he says. “So beautiful.” It’s a word he keeps repeating. Beautiful. Beautiful. Beautiful.
The Beauty of Earthly Intelligence
When we see displays of non-human intelligence in nature, the instinctive response is awe. These glimpses of the beauty and diversity of Earthly intelligence help us know in a most visceral way that intelligence is much bigger than humanity.
I believe that one of the most important lessons we will learn through machine intelligence is that intelligence is larger than any human categories of artificial or natural. As a species, we are about to broaden our understanding of intelligence through the novelty, surprise, and wonder that systems like AlphaGo generate from our seeds of thought.
Learning from Diversity of Intelligence
Lee Sedol did eventually lose to AlphaGo: four games to one. In true heroic fashion, Lee’s response was not to hang his head in shame. Instead, it reinvigorated his love for the game by helping him see a much broader plane of possibilities for play. It’s even changed the way he competes.
“My thoughts have become more flexible after the game with AlphaGo.” — Lee Sedol
In fact, since the Lee-AlphaGo match, many other top Go players are also changing the way that they play. Moves like “playing the fifth line,” which once seemed unthinkable, are now viable new Go strategies. It’s as if contact with this alien machine intelligence has shifted the world of possibilities inherent in the game.
Biomimicry teaches us how to build better systems by drawing from the designs of Nature. As we expand our understanding of Nature to include ‘artificial’ systems, we will have an opportunity to learn from their emergent and unexpected intelligence as well.
There is one important difference with most of these artificial systems though: the feedstock for most machine intelligence today is human data. DeepMind trained AlphaGo by exposing it to some 30 million moves by expert Go players. Similarly, Google trains the machine learning algorithms for its search service with billions upon billions of interactions with end users. The same is true for Facebook and its feed algorithms.
In this sense, the way we learn from our intelligent machines is like having a meta-conversation with them. We feed the machines our data. They then digest it and feed it back to us, just as AlphaGo did for Lee Sedol by playing a game with him. And just as Lee is now updating his Go strategies based on this experience, so too will humans update our behaviors based on what we learn from these systems. This ongoing ‘conversation’ between humans and machines will be the most powerful stimulus to intelligence that our species has ever known.
Doctor Doolittle Too
We all know intuitively that there will be a powerful feedback loop between machine learning and human learning, but what will happen when we expand these conversations to include other species on the planet? What might an intelligent system teach us through its interactions with the mind of a chimpanzee? Or with the collective neural activity of a colony of bees? Or with the biochemical reactions within a bamboo grove?
Human intelligence occupies a microscopically small niche within the vast universe of cognition. Though it may not look like the way Doctor Doolittle did it, one of the greatest gifts we will receive from machine intelligence is the ability to eventually converse with other forms of life on the planet. Through that process, we will finally see just how limited our perspective on intelligence truly is.