I get asked this question a fair amount. Yeah, I’m a software engineer, and I work for a company that does AI. But still. My job also has me design databases, and people don’t ask me nearly as often about that. Or ever, really.
But this AI thing. Man, it really has people interested. And worried. And excited. And confused. Obviously people want to know how it works. And I’ve always been good at that, at explaining technology to laypeople. So if you’re trying to understand AI without having to do multiple years of full-time research, welcome. Hopefully you’ve come to the right place.
Why AI is So Hard to Explain
Now this is most likely not going to be just one post. AI is a giant topic. It’s like trying to write a blog post about the animal kingdom. How deep do we want to go? Do we explain all the nifty biology stuff? Are we going to mention the dinosaurs? Is it okay if we just do mammals, fish and bugs? Or do we need to include mushrooms? Are mushrooms even part of the animal kingdom? And what about single-cell organisms? And so on.
AI is a lot like that. For one thing, not even techie people can agree on what, exactly, AI encompasses. If you’re here because you’re a Computer Science student prepping for an exam and you think “What is AI?” is going to be a question, I have bad news for you: your professor most likely explained their own definition in class while you weren’t listening, and that’s the answer they expect. Good luck to you.
For another thing, AI is not a topic that belongs entirely in the realm of Computer Science. There is a genuine amount of overlap with Philosophy. Questions like “what is intelligence” just can’t seem to be entirely answered with mathematical formulae, although computer scientists have given it a good try.
And then of course, there’s the whole “how it works” part. It’s really not fair, when you think about it. This is cutting edge research. By comparison, nobody at the grocery store expects the people working on nuclear fusion to explain how their stuff works on a weekly basis.
Then again, nuclear fusion research isn’t impacting people’s lives on a daily basis. You can’t get access to a nuclear fusion reactor with a simple internet search. And given how trendy AI is right now, it’s definitely something I think I can use to generate traffic towards my blog. So here we go!
Who Do I Think I Am to Explain This
What with the current hype surroundig chatbots, generative images, and digital impersonations of Scarlett Johannson, a lot of people who know nothing about AI have popped up in the media and online to explain their idea of AI to people smart enough to realize that they know nothing about AI, usually in order to sell AI bridges and make doomsday predictions. I like to think I’m not one of them. For one thing, I did spend several years studying Computer Science, and that included several exams with “What is AI” questions. We also built AIs and had them compete against each other (study Computer Science, kids, it’s fun!). I even wrote my Master thesis on a very technical form of AI known as federated learning, which it is too deep to go into right now. I then went on to work as a software engineer for several software companies building AI, including my current job. Hopefully this convinces you that I am not making this all up as I go. If it doesn’t, that’s okay too. Your traffic to this page has already been registered.
But if you’re still willing to listen to me, let’s take a quick tour of the History books.
Uncle John: What’s in a Name
Let’s start with the name itself: AI, short for Artificial Intelligence. Now if you’re a fan of science fiction, you probably already have an image in your head of what AI is: killer robots, HAL, and Arnold Schwarzenegger traveling back in time to kill Sam Altman. I’m going to need you to put that image aside for now. The real AI is both a lot lamer and a lot cooler than that.
According to the internet, the term “Artificial Intelligence” was coined in 1956 by a math teacher nicknamed Uncle John by his students at Dartmouth. Of course later he was known as Professor John McCarthy, one of the founding fathers of AI, creator of the AI lab at Stanford, and god among computer scientists, but at this point it’s just John. Oh, and there were a bunch of other people there too, but you’re not going to remember them anyway, and some of them were apparently not very nice people, so let’s keep it simple with Uncle John.
So, in 1956, Uncle John gets together a bunch of mathematicians for a brainstorming session about what he calls “Artificial Intelligence”. He allegedly chose a very vague term because he wanted to avoid it getting bogged down in preexisting work and university politics. This will explain a lot later. Anyway, the proposal he and his friends submitted to request funding to pay for pizza expresses their goal so nicely that I’m just going to quote it right here.
The Original Definition
“An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves”.
So there you have it. That’s what AI is. The study of how to make machines talk, solve problems, and learn. Have a great day, everyone.
Some Philosophy
Except not really? A calculator can solve math problems better than most humans. Is a calculator AI? If I type “how to do my taxes” into an internet search bar, and it gives me a bunch of articles about how to do my taxes, it is a computer solving my problem, but is it AI? And at this point we’ve all probably talked to one of those robocallers on a customer support line. It’s definitely a talking machine, but I don’t think anyone would call it intelligent, artificial or other.
Hopefully you’re starting to see the problem here. Artificial intelligence is just a bit too open to interpretation for its own good. Because, at the end of the day, what does it mean for something to be artificial? Man-made? You mean like a baby? Is a baby AI? Are we all AI? See how we just crossed from Computer Science into Philosophy?
And what is Intelligence? We generally consider humans to be intelligent, except for politicians and of course the idiot who ran the red light right in front of you on your commute to work this morning. We typically consider people who are good at math to be especially intelligent. Except of course, you know who else is good at math? Pocket calculators, that’s who. Okay, so maybe we just think of people as intelligent when they’re good at thinking. Except what is thinking, but a bunch of electrical signals in your brain, and how is that different from the electrical signals in a computer again? The point is, intelligence is actually pretty hard to define, and now we’ve somehow ended up among the philosophers again.
If you’re confused and unsure at this point, good. Now you know how computer scientists feel when they have a few beers and start arguing whether grid search (that’s an algorithm which solves problems by trying every possibility and seeing which one works best) is AI. Nobody really agrees on what AI is, so really anything could be.
A Brief History
I know I started this article in 1956 with Uncle John, but it’s worth mentioning that the idea of thinking machines has been around for much longer than that. The ancient Greek philosophers (philosophy again!) got the ball rolling with a lot of arguing and writing about logic and the nature of intelligence, which became the foundation of our own research on artificial intelligence. In the Middle Ages (and before then in China), people build pretty smart robots called automatons, which could do things like walk around, play music and even tell time (also known as clocks). In the 19th century, Lady Ada King, Duchess of Lovelace, mathematician, compulsive gambler, and author of the first computer program (for a given definition of computer), theorized about a machine capable of writing poetry, an idea which sadly was brought to life in 2015 when some Finnish computer scientists taught a computer to write rap music. (Although it did lead to this amazing article).
And last but not least, in the 1940s, just a few years before Uncle John made his summer plans with his nerd friends, celebrated all-around genius and now-proclaimed father of AI, Alan Turing, had already written extensively about thinking machines, before being driven to suicide by the British courts’ attempts to make him be not gay. He wasn’t the only one thinking and writing about AI either, but he was definitely the most famous, and we’re going to talk about some of his work now.
The Turing Test
As you might be able to tell, this test was invented by Alan Turing. Of course he didn’t call it that. He referred to it as The Imitation Game, though, and it’s pretty well explained in the 2014 movie, “The Imitation Game”. It goes like this: imagine you’re alone in an empty room. You write a question on a piece of paper and slip it under a door. The question is answered by either a person or a computer, the answer is written on a piece of paper, and the paper passed back under the door. You now need to guess if the answer comes from a human or a computer. If you guess human, but it was from a computer, the computer “wins”, as in, it successfully displayed human level intelligence, and is therefore a true artificial intelligence. (Either that, or your question just wasn’t very good.)
Turing was a smart guy. He realized how hard it is to define things like “intelligence” and “thinking”, and then he solved that issue by making it irrelevant. He just started from the idea that humans are the benchmark for intelligence, whatever that is, and then found a way to measure computers against humans. Smart, right?
The Turing Test, as you can imagine, is not without flaw, however. In 2014, the Test was famously beaten by a chatbot posing as a 13 year old Ukrainian boy called Eugene Goostman. This pretense allowed the chatbot to fool 33% of the judges after a five-minute conversation, because the judges mistook its bad English and foolish answers for those of a foreign child. Many computer scientists immediately called this cheating.
Where Are We Today
Whether pretending to be a 13-year old is cheating or not, what’s for sure is that as of 2023, the Turing Test is officially broken. OpenAI’s ChatGTP chatbot pretty much took it out back and shot it. Try it yourself (you just need to create a free account). Would you be able to tell the difference between this and a human? Uncle John’s goal number one – a computer that can talk like a human – is done.
Now that we’ve done a high-level overview of AI History, you doubtlessly have more questions, like “Ok, but how does it actually work?”, and “Great, but what does AI actually do except write silly songs in German and give me hot chocolate recipes?”, and “Will I lose my job to AI?”
We’re gonna tackle those next.
Why are Artists Angry?
Remember the second thing Uncle John wanted AI to do? Solve problems reserved for humans? Yeah, that’s what AI can do. Now in an ideal world, we would want AI to work on the boring, menial tasks that humans don’t like, such as looking for mistakes in computer programs, sitting in traffic, and writing cover letters. In the real world, AI is of course pretty good at those things, but also pretty good at the more…fun…tasks. Like drawing pretty pictures, and writing movie scripts.
Now is the point where we go a little bit into the “how”, just to better explain the “what”. We will go more in-depth into the “what” later below, don’t worry.
The “how” is, at a very high level, as follows.
How AI Works
Imagine a human baby. This baby knows basically nothing, except for how to breathe, sleep, and cry when it wants something. The job of the adults caring for it is to show the baby things. Show them over and over and over, in different forms, repeating until the baby has them memorized. This is a sheep. This is also a sheep. This stuffy toy is a sheep. This picture is of a sheep. This drawing in a book is a sheep. This is not a sheep, this is a goat. This animal in the field is a real sheep. The animal next to it is a baby sheep. That black one is also a sheep. And so on. Until one day, little baby is a toddler who picks up a crayon and draws what it insists is a sheep, and not a cloud.
This is basically how AI gets created. Its “parents”, aka the computer scientists, show it what a sheep is (or a dog, or a cover letter, or a drawing of an anime character), over and over again, until it (it being a computer algorithm – more on that later), has memorized the idea of the sheep, and can recognize the sheep in different contexts, and can even make representations of the sheep on its own.
Whose Work it Needs
There’s just the tiny issue of where all those sheep came from, a question linked directly to the question of why authors and screenwriters and photographers and artists everywhere are so angry. AI is not as smart as a small child. A small child needs only a couple dozens examples of something to learn what it is. An AI currently needs thousands, or millions, or even billions. ChatpGPT version 3 was trained on 570 gigabytes of text data, about 300 billion words. Dalle-2, Openai’s fun chatbot that makes pictures for the text you type in, was trained on 650 million images. Take a wild guess whether OpenAI asked 650 million photographers for permission to use their images.
So now you know why writers, photographers and assorted artists everywhere are mad about AI. Not only is AI directly competing against them to create content – but to add insult to injury, AI learned how to do it from them. Sort of. There’s a lot of philosophical questions here again. For example, if I, a human, like anime-style drawings, and I learn to draw them by copying the images of my favorite online creators, and then I start making and selling my own stuff, am I committing the same infraction as Dall-E? Discuss.
So yeah. How does AI work? It works by looking at millions of examples of a thing, until it has identified a set of patterns – sometimes very complicated patterns, like entire screenplays. It can then use those patters to identify other examples of the thing, and create new versions of it. Patterns just so happen to be a fundamental aspect of a lot of fields of art, and now we have a problem.
AI Applications
Now OpenAI has gotten lots of attention for their tools, and rightfully so, because, love them or hate them, you have to agree they are impressive. But OpenAI is not the only AI company out there. In fact most AI really does do the boring work people don’t want to do – whether it’s predicting the weather in East Africa, finding weeds in fields or cracks in sewers, choosing ads for you, driving cars (sort of – they’re still working on this one), day trading, identifying the weaknesses of an opposing sports team, and even deciding how likely a criminal is to reoffend (this one turned out predictably problematic – see section on AI Ethics in following article). Anything you can think of which has 1. Patterns and 2. Large amounts of data available for training, whether it’s text, pictures, video, player stats, crime stats or your instagram data, absolutely anything with lots of data you can repeatedly shove at an algorithm, could become an AI. You just need lots of money to pay for expensive computer scientists (seriously kids, it’s a great job market right now) and expensive computers to train your AI on.
What’s Next (On This Blog)
So I’ve blathered on for about 6 pages now, in which time we covered the basic definition of AI, the philosophy behind AI, the History of AI and a high-level overview of how it works, and even touched on some of the ethical dilemmas involved. Class is dismissed for now, but tune in next week, when we pop the hood on a couple of the most popular AI out there and get our hands dirty on the inner workings of an AI algorithm. Possibly with a healthy side of feeling dirty while talking about the ethics of AI, with examples of cases gone horribly wrong. Until then, have a great week!
