AI is the next Industrial Revolution

New software from Google called AlphaGo just beat a human grandmaster, something that a lot of people thought wouldn’t happen for another decade. The pace of Artificial Intelligence (AI) improvement isn’t that surprising given the resources that companies like Google, Facebook, and Microsoft are pouring into research. AI powers many of the new features we see, such as image search, facial recognition, and language translation.

Why is AI development so important? Well look at this:

What happened around 1800? The steam engine and the industrial revolution is what happened. We saw the advent of technology that for the first time overcame essentially all limitations on our muscle power. I think we’re on the verge of a similar inflection point, and we just can’t quite see the slope yet. What’s happening is that AI is on the verge of helping us blow by many meaningful limitations on our mental power.

What got us here? A few things. First, computational power is increasing exponentially. Thanks to Moore’s law, by the year 2020 a computer that can match a human brain on calculations per second will cost around $1,000. The second is Big Data. AI is to big data what the entire petrochemical industry is to crude oil. All this data can be used to train AI. For example, if you show Google a cat, it can find other cats (actually it can do this now without even showing it a cat in the first place, but that’s a different story). And third, we have better and more diverse algorithms. Advances in machine learning and neural networks are basically reverse engineering the workings of the human brain.

That last one is what really led to AlphaGo’s success. Go is a ridiculously diverse game. There are 250 possible moves each turn (in chess there are around 35). Even our best supercomputers can’t brute force Go. There are more possible Go positions than there are atoms in the universe. Not to say that quantum computing or something else won’t eventually get us there, but for now it’s a bottleneck. Instead, AlphaGo learns. It plays itself over and over, generating new data and strategies, mimicking the secret sauce of human grandmasters, what we see as vision and intuition.

We’re still in the very early stages of understanding how to use AI to its full effect. It took about 30 years for companies to take advantage of the advent of electricity. At first, they just replaced the central steam engine in a factory with an electric generator. Everything was still crowded close together, since back in the day everything had to be physically close to the steam engine. It took decades for electrical wiring to become mainstream and for factories to be redesigned more efficiently.

But it’s coming. Computers aren’t just good at repetitive tasks anymore. Now they can drive cars, pack boxes, diagnose medical conditions, and print replacement parts for themselves. And thanks to the internet and smartphones, billions of new eyes now have access to essentially all of human knowledge, and will think of brand new things for AI to tackle.

Unfortunately, like a lot of early technology, the first winners will be the rich and the uber-rich. At the moment we’re on a trajectory that leads to the world having 10 trillionaires while the rest of us drive each other around and serve each other lattes, depending on which contract job you’re working that day. I’m a big fan of the gig economy and it will help address some of the imbalance, but unemployment and inequality are serious risks going forward. I don’t have any specific cure-alls, but I do see a lot of externalities that our government could help address with positive action:

• Spend Way more on education, higher teacher salaries, accountability, etc.
• Encourage re-education and re-employment tools (online courses, job matchmaking, etc.)
• Encourage startup ecosystems worldwide
• Spend more on basic science worldwide
• Upgrade our core infrastructure
• Fix immigration and let talent move where it needs and wants to
• Tax very rich people more
• Encourage democracy everywhere
• Create a guaranteed income/basic welfare floor

It’s possible that we’re going to have Artificial Super Intelligence (ASI — an AI that’s much smarter than the best of us across all disciplines) by 2050 or 2100. The scary thing is, once we hit the intermediate level of Artificial General Intelligence (AGI — as smart as a normal human across the board), recursive self-improvement could lead to ASI very quickly — some experts think it could be within months, if not hours or days.

So yeah, exciting times. And it would be nice if we had a few more bits of our house in order when all this happens.

For a super entertaining read on AI, check out this Wait But Why post.