Until recently, I was living in the land of bliss not knowing anything about Artificial Intelligence (AI). Once I started learning more about it all kinds of fears came to the surface, and I felt the need to blog about it, which I shared in Part 1 of this series. This week the conversation continues and I share some of the basics of AI in case you need an update and want to make some sense of it for yourself.
The Land of AI.
There was a time when I knew little about AI. Thankfully, these days there is a whole lot out there to fill the gaps in my ignorance. After watching videos and reading a few articles, I think I now have a general sense of things, but I’m by no means an expert.
What follows are the things I’ve learned and made sense of so far on my AI journey. If you want another good place to start your own journey I recommend reading this popular Tim Urban blog post on AI. It will also give you a good summary of things.
This post breaks from the tradition of blogging about my mental health to lay down the AI basics so that you can start to understand the basis for the fears I and others have around this technology.
Personally, I also think these things are worth learning about since Artificial Intelligence is something that will continue to shape our future and could have, if it hasn’t already, an important impact on our mental health.
The Future is Here.
Artificial Intelligence is here now, in the cloud, on your phone, and all around. It is not something that will happen in the future. We use it every day and may not even realize it, at least that’s what I discovered. One good example? Google. There’s a lot more behind that simple-looking search bar. Google is a big player in the AI game. It’s one of the biggest AI companies in the world.
What is Artificial Intelligence Anyway?
Honestly, it’s something I’m still making sense of as I write this, but if you’re thinking Siri, robots, or self-driving cars you’re on the right track. Though, as I recently learned, that’s not quite right. These programs and machines are not AI in the truest sense, which I will explain later.
In general, Artificial Intelligence is an umbrella term for any computer program that does something smart. Okay, so that’s a circular definition and a vague one.
To put it into context, consider some of the ‘smart’ things you and I do including our ability to solve problems, to learn new things, to plan, perceive, move and to manipulate objects. Systems with Artificial Intelligence can do the same, making them pretty smart, almost human. And some big name players in technology, like Elon Musk, are concerned about that. Something I shared in Part 1.
Coding that Codes Itself.
In case you need a refresher, computers run on codes or programs – the software – that bring the hardware to life. Until recently, these programs were solely designed and developed by computer scientists, engineers, and programmers. This has been changing rapidly over the last few years.
Programs (also called algorithms) that are considered to have Artificial Intelligence are programs that do most, if not all, of the programming. No human required. I mean except for developing the initial AI program of course, and in the future even that could change.
How is this Possible?
Computers do this through Machine Learning where algorithms, statistical models, and computer systems come together to form the brain of the whole operation; and in AI programs the thing that thinks is called an intelligent agent. This is the part of the program that interacts with the environment it was designed in, whether it’s in a video game or in the real world.
To gain its intelligence, agents are programmed to achieve a specific task or goal. The agent gains knowledge and learns to achieve its goal by making predictions from it’s environment and taking action based on those predictions.
In other words, it tests things out by guessing and when those guesses align with its goal it continues on its path. If those guesses are wrong it learns what not to do, and approaches that particular situation in a new way the next time it encounters it. If that sounds familiar there’s a good reason for that. Its kind of like the way you and I learn about our world and about life, through trial and error, through experience and rewards.
That’s not a coincidence. Artificially intelligent systems are designed this way. To mimic the human brain through what are known as artificial neural networks (that was the brain that I referred to earlier). This specific kind of Machine Learning is called Deep Learning.
Artificial Intelligence May Not Be What You Think.
In the intro I mentioned that Siri, robots, driver-less cars (even the algorithms of YouTube and Amazon), are technically not true AI.
Even though these systems do depend on AI, I recently learned that in these examples software engineers are really behind the wheel guiding the learning process, correcting the intelligent agent when it gets things wrong or when engineers want to optimize its behaviour. In other words, these are not completely autonomous programs.
For example, in driver-less cars if the AI predicts that the vehicle driving next to it is a truck, when it’s an ambulance, the engineer would need to reprogram the AI code so it recognizes ambulances and pulls over to the side of the rode as they pass by.
A Simple Example of True Artificial Intelligence.
To get a real sense of what true AI looks like I recommend you check out this talk by Danny Lange.
Danny helped develop AI in companies like Microsoft, Uber, IBM, Amazon, and now Unity, the world’s largest online video game developer. In his presentation he shows in basic and simple ways what AI is all about. In one example he uses a chicken to make the point.
The Chicken that Learned to Cross the Road, Real Fast.
In video games that remind you of Atari, Danny shows how an intelligent agent can do some incredible things, from scratch.
In one fun example, an intelligent agent in the shape of a chicken needs to learn to cross a series of roads with cars streaming by. The goal for the chicken is to collect packages along the way while not going splat. The packages act as a reward that help direct the chicken’s behaviour.
The program code is basic with instructions on where the chicken can move – forward, backward and side to side, and coding that reflects a reward function: a positive value for picking up the package and a negative value for getting hit by a car. That’s it. The rest is up to the intelligent agent to figure out. It has to learn how to stay alive by learning through experience and reward.
A Super-Human Chicken.
At the beginning, when you watch the chicken in action, you see how every other move it makes leads to road kill. It hasn’t yet learned the rules of the game and has to start all over again. As dumb as the chicken appears at this point that stupid is as stupid does, doesn’t last.
Thanks to it’s neural networks it remembers every move, and mistake, it makes. And after a few short hours the intelligent agent gets good at the game, like real good. So much so that the chicken learns to race through the streaming cars in what feels like warp speed, without getting killed. It is clear when you watch it that the machine has learned to do something that is super human, something a human player controlling the chicken could never do.
A Total Game Changer.
That’s been a common theme in the story of AI over the last couple of decades: computer programs beating us at our own games.
In case you haven’t already heard, the world champion in chess, Jeopardy, or Go, is no longer a human but a machine. Since 1996, when the AI program Deep Blue, beat the then world champion in chess, Garry Kasparov, new AI programs were developed that not only have crushed the human competition, but have also gone to beat themselves. Let me explain.
From AlphaGo to AlphaGo Zero to AlphaZero.
Those may all sound like the names of college fraternities or codes read out by a NASA engineer to an astronaut, but all these alphas are AI programs that were designed to play a board game called Go. It’s one of the oldest games in the world, coming from China over 3000 years ago.
The rules of Go are simple, but it isn’t a an easy game to play. The number of possible moves a player can make are astronomical because of the size of the board (its a board bigger than chess). To give you a sense of things, there are more positions in a game of Go than there are atoms in our Universe.
In Go, the name of the game is to beat your opponent by surrounding as much of the board with your pieces or stones. Up until March 2016, Lee Sedol was one of the best in the world in doing that. That month, Google’s AI program AlphaGo took on Lee in what is now considered a game-changing event.
The Match That Changed How Humans Look at Machines.
This match was a big deal. It showed that AI programs not only can win at sophisticated games, but can also do so intuitively.
Using unconventional moves that not only surprised everyone, but also taught it’s opponent new ways to play the game, AlphaGo beat Lee four games to one over the course of five days. Lee was able to win one game over the program thanks to lessons he learned from watching it play. After losing the first three games, Lee decided to use some of the unconventional tricks he learned from the computer against itself, which confused the system.
Want to know an even bigger deal? Unlike the program for AlphaGo, the AI successor, AlphaGo Zero, was designed to learn the rules of Go from scratch, without any human data or tips about game strategy. Much like the chicken that learned to cross the road, it learned to play Go all by itself starting only with the rules of the game as input – and by playing a series of games against itself.
The Greatest of All Time.
In the course of forty days the system learned enough to be able to play against the original AI program, AlphaGo, which was no match for the new kid in town. AlphaGo Zero stood by its name and went undefeated winning 100-0. It also defeated all the other AI programs designed to play Go.
Since this AI learned everything from scratch without human input, this was another game changer. Often, for AI technology to be successful, it relies on an enormous amount of human data and input for learning, and for developing its intelligence, whether that input is from millions of images, stock market prices, or people’s voices.
Today, more artificially intelligent systems are doing the learning all on their own. The case of AlphaGo Zero reinforced the idea that at the heart of AI technology is the algorithm not the human input.
All this talk about AI can make it seem like AI is just for games, but underneath the surface, and if you’re reading between the lines, it’s hard to ignore that AI is serious business. Considering its computing power and its ability to do what we’ve prided ourselves best in doing for centuries, thinking, it’s clear that AI is becoming a force to be reckoned with.
For now, and perhaps for a very long time, it looks like the reining champion of Go and other board games (Google’s latest AI program, AlphaZero, has mastered the games of Go, chess, and shogi or Japanese chess) will be a machine not a human.
This trend, as you might know, is happening in other areas too, areas where we were considered experts in all of history. Until now. Our domination as the smartest person in the room or species on the planet is changing fast as AI accelerates us into the future. I wonder what the future holds and what kinds of machines we are bringing to life.
Seeds of Fear.
If you’ve been following along and are able to make your own predictions of where things are going, you may be realizing that this is where the seeds of fear of Artificial Intelligence begin to grow, at least for me they do.
And this is where this post will end.
In part three of this series I’ll share some of the predictions experts are making and what sort of things people are worrying about, as I continue to share how I’m learning to make peace with my fears of AI.
Also, in a future post to follow, I will share how I think there’s another intelligence that’s emerging aside from what we’re seeing in AI. It is an idea that initially inspired me to write this series and is meant to offer a more hopeful view of things.
Thanks for taking the time to read this post and for your patience between this post and the previous one in this series.
Did this help orient you to the world of Artificial Intelligence? I’d love to know what you think. Please share your thoughts in the comments below.