Section Two - Introduction in AI
A story about chess computers, cobots and computers playing videogames (17 minutes)
History of AI
Artificial Intelligence. AI. Neural Networks. Deep Learning. Machine Learning.
You hear people talking about AI all the time. AI is a buzzword. As a company you have to say that you use AI, even if it is not true. According to a survey in 2019, from London venture capital firm MMC, 40 percent of European startups that are classified as AI companies don’t actually use artificial intelligence in a way that is “material” to their businesses.
But despite the current hype, it's good to know that AI has been around a long time (especially when we calculate in computer years). That is why we start this section with the history of AI in a short, animated video (3 minutes).
So, artificial intelligence has been around since 1950. That is a long time. A calculator uses artificial intelligence. The chess computer that defeated Garry Kasparov in 1997 (Deep Blue) used artificial intelligence. However, most people do not consider their calculator to be artificially intelligent. That is probably because a calculator can’t 'learn.'
The same goes for Deep Blue, the chess computer. This computer was very good, and very fast, but could not learn from its mistakes. Calculators and older chess computers can be very powerful and can do very impressive things, but they basically employ a traditional algorithm.
But, what exactly is an algorithm? Let's watch a short one minute video with an explanation.
Winter is over
In the history of artificial intelligence, an AI winter is a period of reduced funding and interest in artificial intelligence research. The term was coined by analogy to the idea of a nuclear winter. The field of AI has experienced several hype cycles, followed by disappointment and criticism, followed by funding cuts,then followed by renewed interest years or decades later.
Today, most people will say that winter is over.
There are three major reasons for this. These are explained in the video below (2 minutes).
So, faster GPUs (1), big data (2) and (layered) algorithms (3) are the main drivers behind the rapid development of AI. These three make machine learning possible. Machine learning is a somewhat misleading, popular term for software that, in certain cases, is able to improve itself.
This video explains the concept of machine learning in 5 minutes:
Yes, we admit this is very simple explanation. And, yes, there was no need to fire the grandpa, And yes, machine learning, is of course way more complicated, but that is not really important for this crash course. The central idea is that computers can 'learn.' Learning is enclosed in quotation marks. After all, it is not really learning. There is no consciousness. It's just software that can improve itself. If you define a certain desired output and make sure there is enough input (training data), the computer starts to 'learn' from its mistakes and it gets better (in certain cases).
That is still very impressive.
To get an impression of how impressive this is, let’s take a closer look at the aforementioned example from the chess world. In 1997, Deep Blue, IBM’s chess computer, beat chess grandmaster Garry Kasparov. But that was the only thing Deep Blue could do well. Everyone could beat Deep Blue effortlessly in a game of tic-tac-toe. The year 1997 was followed by a short period in which artificial and human intelligence worked together and played chess against each other. These teams were called Centaurs. This is often used as an example of how robots and humans can work together. Sometimes the term cobots is used. However, what is often not told, is that Centaurs hardly exist anymore, because the chess computers are too good and human players, therefore, have become irrelevant.
So the chess computers play against each other and humans hardly play a role anymore (some will say: a bit like Formula 1).
A few years ago the best chess computer ever was called Stockfish 8. It was very powerful, but still classically programmed by humans. Just like Deep Blue a long time earlier. Stockfish 8 was mainly a computer that had a lot of computing power, and so it could calculate a lot of scenarios very quickly and choose the best chess move. But recently this computer played against a system that uses machine learning, named AlphaZero. This system, had never played chess before, but it knew the scope (a chessboard and some riules) and the object: winning. So AlphaZero trained by playing millions of chess games against itself and within 4 hours it managed to win 100 of the 128 games against Stockfish 8.
That is what we mean by impressive.
Because these machine learning systems know the scope and the goal, they can also get very creative in solving problems. Below a short video (1 minute) of an AI system playing breakout and finding a solution the programmers of the system had never heard of.
AI is very good at games mainly because they have clear rules and a clear goal (winning!). AI is even better at playing the board game GO than the best players in the world, a feat that, due to the complex nature of Go, until recently was considered impossible. At the moment AI is also beating the best poker players in the world. Maybe you care, maybe you do not, but it is advisable to restrain yourself from playing online poker.
However, remember, life is not a game, although computer programmers sometimes think differently. A good conversation has no clear rules and no clear goal. A computer like AlphaZero, despite conducting millions of conversations with itself, is not the best conversation partner ever.
Artificial Artificial Intelligence
The data with which artificial intelligence is trained is often created by microworkers. With their work: transcribing, writing and tagging photos, microworkers are feeding algorithms with data and contributing to a more powerful ‘artificial’ AI.These microworkers are often deployed by the Amazon Mechanical Turk platform (AMT).
The 'Mechanical Turk,' was a fake chess playing machine invented in the 18th century. It was a mechanical illusion, an automated machine, concealing the fact that a human being was hiding inside to operate it. The creators of Amazon Mechanical Turk got inspired by ‘The Turk’ of the 18th century. Further, the comparison between the chess master in the machine and the microworkers behind AMT is crucial in order to demystify artificial intelligence.
“AI is made by people and with human input,” says Lilly Irani (2016). Because there is little artificial behind artificial intelligence, a more suitable descriptor would be artificial ‘artificial intelligence’.
Weak & Strong AI
Computers get more powerful really fast. Humans learn how to build better machine learning software. Machine learning software itself is able to improve. There is acceleration everywhere. That is why a lot of people are hyped or worried by the rapid developments of AI. The hype mostly comes from weak AI. The worries from strong AI.
Weak AI which is really good in one thing. For example, driving a car or playing chess or recommending things on Netflix or recognizing your voice commands.A lot of people think weak AI will assist us everywhere. It will be in the cloud and connected to our devices. So our fridge can figure out for us what to cook for dinner. Our toothbrush can look for cavities. Our toilet can tell us if we need to go to the doctor, and so on. Of course, there are also worries with weak AI, but we will talk about them later.
Strong AI is AI that can do many things. It is the AI from the movies. For example, a personal assistant that trains your daughter's soccer team, reads your reports, talks to your wife, gives you mental support, arranges for a new mortgage, and so on.
Weak AI is around us everywhere, strong AI is nowhere to be found. Yes, there are some AI systems that can play a multitude of arcade games, but that is still a long way from the humanlike intelligent robots we know from the movies.
A lot of people believe that one day (soon) AI will become more intelligent than humans and then they will build their own AI, which will become so intelligent that we humans have no idea what is going on anymore. It would be like explaining this crash course to your dog. If strong AI emerges we are truly f*cked or blessed depending on what you believe. Other people have enough reasons to believe that another AI winter is coming, at least when it comes to strong AI.
If you are interested in strong AI and the debate on the future of AI we offer you some links in the additional materials section (six). For now we focus on the rapid development of weak AI and the problems that this entails.
Starting with the black box of AI in section three.
Take aways from section two:
- AI has been around since the 1950's;
- Powerful GPUs, big data and (layered) algorithms paved the way for machine learning;
- Machine learning is not really learning, but software that can improve itself;
- The results and possibilities are very impressive;
- Artificial Intelligence itself is often artificial;
- Weak AI has and will change our lives drastically;
- The emergence of strong AI is still very uncertain (fortunately).