AI: Explain It Like I Suck At Math (ELISAM)

You hear it all the time. AI Is exciting, AI will change our lives, but also Terminators are coming.

AI is also full of expert and insider lingo. If you’re not really into it, it’s hard to find a proper and simple read. Most articles will get you into formulas, programming, and abbreviations like GANs, SVMs, and whatnot.

Rightly so, AI is discussed by scientists, engineers, programmers, and data scientists. But this doesn’t mean you shouldn’t understand it. It influences your daily life and will amazingly influence your work.

in reality, if you’re not an AI professional who is tackling new era algorithms, things aren’t that complicated. In real life, as a user, AI is mostly visible to you in form of simplification of daily tasks.

Before we go further, just another clarification. A term AI is mostly used for self thinking robots (in possible future). Today, at our stage of development AD 2020, we talk about Machine Learning.

Machine Learning is a subset of Artificial Intelligence

Three major Machine Learning (ML) fields

There are 3 major ML fields, Supervised Learning (SL), Unsupervised Learning (UL), and a weird child that is called Reinforcement Learning (RL).

Supervised Learning is used when you have to classify things that are already known - for example when a machine has to recognize whether there’s a cat or a dog on a picture.

Unsupervised Learning is used, when you have huge data lakes (so a bunch of data of who knows what) and you want to find new patterns in this data - for example, say a bank is looking for unusual and forgery transactions.

Reinforcement Learning well, read below.

Three basic subsets of Machine Learning

You might also heard about Deep Learning (DL), which uses something called Neural Networks. These networks imitate the human brain (somehow) and can process more complex calculations. DL can be used to optimize SL, UL, or RL.

Simple representation how Neural Networks work to recognize image

Another subset worth mentioning is GAN (Generative Adversarial Networks). Behind this weird name is a cool technology that you already saw. GAN for example is used to produce completely new pictures out of nothing, create deepfakes, composes new music, etc.

Check Barack Obama fake video created by GAN.

This is all you need to know before your head explodes. Of course, you’re more than welcome to dig deeper.

Reinforcement learning

Let me tell you, why RL is really fun.

You remember as being a child, cleaning up your room got you a delicious cookie.

But when you fought with your friend, mom yelled at you (if you were lucky). That made you learn what behavior is good for you and what is not. And upon these rewards and punishments, you shaped your behavior in everyday’s life (well, also true some people never learn :).

What RL does is pretty much the same. It takes a little you (robot or agent) and puts you in some environment: say you are a stockbroker. So you can buy, sell, or hold stocks. And every time you earn profit you get a cookie and when you, well, loose, you get some angry mother’s roar.

Learning robot to walk with RL

Now with RL, you can do this on a huge scale (because it’s a computer, right, and it doesn’t have physical limitations). So this guy can trade stocks with light speed and does billions and billions of repetitions while you’re having lunch. And because it gets cookies and roars, slowly this guy gets better and better, even to the extent it becomes Gordon Gecko of trading.

How good? For example in 2016, where this technology won World Champion in Go, the most complex board game, by a huge margin. These days computer beats world champions in complex computer games.

But it’s not about the games, this tech can be applied to many different cases, like ordering material, optimizing self-driving cars, and so many other things.

Why is RL a weird child? Because it has somehow quite a lot of human touch. Unlike SL and UL, who are based on advanced statistics, RL includes a lot of psychology (and related stuff).

Raising this robot is like raising a child. If you fail to reward it properly, or you don’t use proper rewards, well, the robot will behave badly and do stupid things. It’s also a bit like a game, you can play many scenarios and enjoy the outcomes of this child.

In the Go game, the robot actually did all the learnings by itself (without human input), and was really impressive: it found new strategies until then unknown to mankind.

So you can imagine this technology has a lot of potentials to assist us in the future.

Good or bad?

It’s hard to avoid good old AI discussion, will this make our lives better.

I will always follow the premise, that tech is neutral: we make it either good or bad. But that’s a common standpoint, I know.

Point is, that technology will be what we make it. There is quite a common consensus that at the moment we are not on the right path. But it doesn’t have to be like this.

I’d like to do here is to invite you to read the short novel Manna written by Marshall Brain. It shows both outcomes, positive and negative.

And the book says, well, we decide where we want to go.

Marshall Brain - Manna (the book)

You can also read a simple explanation of Decentralized Finance or DeFi.