How to build a robot’s brain

If you haven’t heard about machine learning lately, you haven’t been reading the papers…

Robot's Brain

If you haven’t heard about machine learning lately, you haven’t been reading the papers…

Here’s some machine intelligence news from just the last week: a giant hedge fund, the Man Group, ordered an Oxford University department it funds to switch its focus to machine learning… Facebook unveiled its machine learning platform… And the FT ran a special week-long series called the Rise of the Robots which was largely about machine learning.

It all sounds very futuristic and important. But what does machine learning actually mean? What does it do? Can you invest in it, or is it just for computer scientists and hedge fund managers?

The answer to the last question first: yes, you can invest in it. Machine learning is spreading fast. In fact, three tiny companies in The Penny Share Letter portfolio are based on it. So machine learning isn’t just a passing fancy for me – I need to understand it!

I recently read a book about it called The Master Algorithm by Pedro Domingo. I first mentioned it a few months ago when I was talking about Google’s aggressive move into AI.

Today I want to share the stuff learned in the book. It’ll help you make sense of the headlines, and the business models of some fast-growing young companies.

The holy grail of computing

Pedro Domingo is a respected computer scientist who’s been working on machine learning since the 1970s.

His big idea is this: there is an as-yet-undiscovered software code which can teach itself how to do lots of different tasks, in the same way that our brains can teach themselves how to do lots of different tasks. That’s what the book’s title is about –The Master Algorithm.

Our brains can make sense of lots of different kinds of stimuli. If we see text, we can learn how to read it. If we hear music, we can learn how to play it and how to compose it. We can learn how do hard sums, play twister, vote in an election. The “master algorithm” in our brain teaches itself how to do all that stuff, provided it’s been given enough stimuli to work with.

In computer science circles it’s what’s called a “general learner”: a piece of learning software flexible enough to handle loads of different tasks.

The computer scientists haven’t gotten that far yet. A general learner algorithm, if it’s even possible, would be the holy grail of computer science. But they have made made big strides in the last five years or so.

[cfsp key=”footer-ps”]

The five schools

They’ve come up with five different types of machine learning algorithm. Each of them might be the route to the holy grail, the general learner. But then, none of them might be. Each type of algorithm is good at learning some things and bad at learning other things.

The first school of thought in machine learning is the evolutionist school. This type of algorithm mimics evolution in nature. As Pedro Domingo explained in a recent podcast,

“The idea there is that, well, evolution, that is the greatest learning algorithm on earth. It created you and I, not just the brain but every animal and plant, every living creature that exists. So, why don’t we try to evolve programs in the same way that nature evolves creatures?

And we know roughly how evolution works: evolution is very much an algorithm. It’s a search algorithm. It’s something that’s very similar to computer scientists; and indeed these days biologists do tend to view evolution that way.

It starts out with a population; each one of them performs the task; the ones of them that do best, they are the fittest; and then they get to mate with each other and they produce offspring. And then the next generation will be better at the task. And it turns out we can do amazing things that way. So, that’s the evolutionary approach.”

The second type of algorithm is called “connectionist. Connectionist algorithms are inspired by the human brain. The brain is made up of connections between neurons. It learns by strengthening or wearing the connections. Connectionist model this process on a computer.

The third school of thought in machine learning is the analogist school. This one is easier to get your head around because it’s sort of similar to how people actually think. It’s about reasoning by analogy – finding the similarities between two things, and then making guesses about what else they have in common. Your “what to try next” recommendations on Amazon or Netflix come from analogist algorithms.

The fourth school of thought is the symbolist school. This is sort of like learning by logic. The algorithm knows for sure that Lionel Messi plays for Barcelona, and that all Barcelona players are rich. So it can use logic to figure out that Lionel Messi is rich. Here’s Domingo again:

And this is very, very powerful, because you can introduce different rules from different things, and then you can chain them together in new ways to answer completely different questions from the ones that you originally saw.

The last school of thought is the Bayesians. Maybe you’ve heard of Bayes’ Theorem in statistics – this school comes from that theorem. Here’s how Domingo explains them:

The way they look at it is like this: I have a range of hypotheses that I could use to explain my data… I will always be uncertain about which is the right hypothesis, because induction is always uncertain. But what I am going to do is I am going to quantify the uncertainty with probability. And then, as I see more evidence, I update the probability with which I believe each hypothesis. So, the hypotheses that are consistent with the data will tend to become more probable, and the other ones will become less probable.

Domingo is optimistic about Bayesian algorithms because they’re very flexible. In theory, you can apply Bayesian learning algorithm to any type of problem. And that’s key to building a general learner.

Whether or not we ever get to the general learner, there’s plenty happening in machine learning right now using just those five types of algorithm. Big companies and small are putting them to work and making money from them.

Tomorrow I’m going to talk about how to put machine learning to work in a business… and where investors fit into all of this.

What are your thoughts on machine learning – do you stick to Warren Buffet’s dictum never to invest in something you don’t understand? Or do you need to learn in order to keep up with the changing market? Let me know –

[cfsp key=”footer-ps”]

You may like

In the news
Load More