My piece yesterday was called How to Build a Robot’s Brain.
It was about computer scientists’ quest to build a “general learner” – a computer algorithm smart and flexible enough to learn anything you put in front of it. In other words, software that can mimic the human brain.
I wrote about the five types of learning algorithm they’ve come up with so far: connectivist, evolutionary, symbolist, bayesian, and analogist. Needless to say, it’s highly complicated. I won’t pretend I fully understand it all. But I have learned a couple of interesting ideas from reading up on this stuff.
Today I want to share some of those ideas with you, and show how investors are putting them to work.
Fusty old academics
Machine learning is an old part of computer science. But until recently, it hasn’t been very sexy.
That’s basically because until recently, machine learning didn’t really work! It didn’t solve practical problems.
The old-school machine learning experts had their own ideas about how to solve machine learn problems (for example evolutionists, symbolists, and analogist algorithms). But they weren’t making great progress.
The problem with a lot of old-school approaches is that they relied heavily on programming. They tried to come up with a very smart set of instructions which would show the computer how to interpret the world, and learn from it.
Machine learning has come very far in the last few years because it’s broken away from that approach. The big breakthrough is called “deep learning”. “Deep learning” algorithms don’t learn by following rigid rules. Instead they interact with the world and figure it out by trial and error.
Why AlphaGo mattered
Did you hear about the AlphaGo robot which beat a world champion human player at the game of “go” earlier on this year? That’s a good example of deep learning.
Everyone was very excited about that because go is a different sort of game to chess (which computers cracked twenty years ago). Go has many many more possible moves than chess, which means it can’t be cracked by brute computing power.
To win at go, you need intuition. And it was thought that computers can’t have intuition. AlphaGo proved that was possible. The computer studied millions of games of online games of go. By studying those games and simulating them, it was able to figure out from them which types of moves were more likely to lead to a win in a given situation. Since every game of go is unique it couldn’t just copy the previous winning moves. But it could intuit the difference between a good move and a bad one.
Deep learning algorithms like AlphaGo are the reason why everyone’s so excited about machine learning these days. When you set them loose on big data sets, they’re capable of amazing things. If you give them data about how drivers use the roads, they can learn to drive. Give them data about how people speak and they can learn language. A company in the penny share letter portfolio uses them to train bad habits out of teenage drivers. Another uses them to tell employers who to hire and who to fire. Deep learning algorithms are everywhere, once you start looking.
By the way – if you’d like to see how to actually participate in this thing, and back it with your own money, you can take a trial subscription to The Penny Share Letter here.
The great robot boom of 2016
One big application, which has investors all in a tizzy, is robots. Robots which can learn. Robots which can work alongside humans, and learn from them, and mimic them, and help them work.
As the robot investor Dmitri Grishin says, “once you ship the device, you can apply more and more intelligence and machine learning. First put them in consumers’ hands, then learn from their behaviour.”
For now, they envision something called “cobots”. These are robots which learn from humans and complement their skills. James Settler of Barclays Capital estimates the cobot market could grow from just over $100m last year to $3bn by 2020. I’ll leave it to you to guess what comes after cobots…
Disclaimer: I’m deliberately putting the politics aside here. That’s a subject for another columnist…
There’s huge money pouring into robotics at the moment. Here’re a few big numbers: the market will be worth $135bn by 2019, at a 17% annual growth rate; robotics patents have tripled in the last decade; venture capital investments in robotics doubled in just the last year to $587m.
At the moment the world leaders in robotics are Germany and Japan. But the rise of deep learning tilts the advantage back towards the USA. Germany and Japan’s robotics are an offshoot of their advanced manufacturing industries. America’s advantage is in software and machine learning.
That’s the sweet spot for investors – advanced robotics married with cutting edge machine learning. As Steve Jurvetson, another robot expert, puts it, “everything gets better over time. This is happening in almost every hardware product: they are becoming minimal vessels for software.”
p.s. Thanks to all who wrote in yesterday. A few among you are way ahead of me on this topic. Here’s John S:
“Be careful of the hype – AI has been hyped several times and each time money spent has been wasted (the famous fifth generation project , Alvey and Esprit projects . The IKBS – Knowledge base tool era (circa 80s). Neural system in the late 40 were revived in the 80s) .
The key to AI is real-time environmentally orientated machines with the ability to fully sample the wonderfully consistent and rich environment (as we do). This is were AI started (with people like Nobert Wiener, Hopfield and Hebb et al) long before the computing industry appropriated the subject. Computers are not the best starting point.
… The most useful product of this research will be an understanding of our own relationship with our fantastic consistent detailed environment which can be said to determine our actions even now when we think that we are its master.”
To yesterday’s question – very, very few of us could be said to truly understand “deep learning”. But deep learning is obviously an amazing tool which is starting to totally remake society. What’s an investor to do?! Do like Warren, and sit on the sidelines? Or believe in the product?
I take the view that I may not understand the product fully… but I do understand “that it works”, and that it’s valuable. What about you? [email protected]