Skip to 0 minutes and 1 secondThis time, I would like to talk to you about artificial intelligence, the scariest topic, probably, of computer science. So what do we mean by artificial intelligence? What is it that you qualify as a smart or intelligent machine? There is a test coming from Alan Turing which is called the Turing Test and this was the first idea how to qualify something as a smart machine. The layman interpretation is that, as long as the computer can do something which we would do using our intelligence, our knowledge, the computer would qualify as intelligent.
Skip to 0 minutes and 39 secondsNow how to test this, Alan Turing suggested that we should try to communicate with that entity and if we cannot conclude whether it is a human or a computer and it is a computer, then that machine should qualify as intelligent. However, you would be surprised, but there are actually some very, very simple machines that can pass the Turing Test. There is, for example, Psychology's Concealing Software that can pass the Turing test. What it does, it has a nice algorithm for recognising certain types of words. So, for example, it recognises the noun in your sentence. If it recognise that, then it will tell you, OK, please tell me a little bit more about your father or whatever the noun was.
Skip to 1 minute and 32 secondsThen, if he doesn't recognise that, then jumps to the next one, tries to recognise a verb. So, what do you mean by experiencing bad things and similar things. If it cannot recognise the verb, it will try to recognise some attributes and ask you more details about that and so on, so on. Now the really, really tricky thing is the final rule. When it cannot identify anything from your sentence, then you know what it will say? It will say, I perfectly understand you and then it goes silent. And, of course, after a while, you will start talking again. And this computer programme actually passed the Turing Test. Many, many people could not tell that it was a computer.
Skip to 2 minutes and 17 secondsHowever, is this software really intelligent? Apparently not. John Searle came up with a different idea of how to qualify a computer intelligent or not. He said that what we need to focus on is the achievement of the meaning. So his argument is called the Chinese Room argument and it is a thought experiment. So imagine that you don't speak Chinese, you are locked in a room with all sorts of books. These books are books of rules with Chinese symbols. There is someone outside the room who does speak Chinese and they give you messages under the door, written with Chinese symbols.
Skip to 3 minutes and 3 secondsYou have no idea what those symbols mean, so you will try to find those symbols in the rule books and the rule book will tell you if you receive this symbol or combination of symbols, then this is what you should send back as your message. If the book of rules is good enough, then your answers will be good and the people outside will think that you speak Chinese. However, does this mean that you actually speak Chinese? Of course not. So this means that the achievement of meaning can be obtained by one single party in a conversation, not both parties have to be intelligent.
Skip to 3 minutes and 44 secondsThere was a very interesting story about some psychologists running an experiment and they went to mental health institutions and they applied to be patients. They said that they are mentally ill and it was nearly 100% that they were admitted as mental health patients. However, there was one other thing that happened that very few noticed at the time. It is that the other patients did not accept them. They felt that they don't belong there. They could fool the doctors, they could not fool the other patients. Now that gave me an idea how you could try to test in a different way, along the logic of the Chinese Room argument, whether a computer is intelligent.
Skip to 4 minutes and 31 secondsPut two of them talk to each other and then listen to what comes out. And there were some people probably coming up with the same idea and they ran these kind of tests and it is hilarious what comes out from these kinds of conversations. For example, there was one in which someone threw a stone in the air and, as a consequence, the gravity has fallen down. So obviously it is very difficult to build an intelligent machine. There are two different paradigms of artificial intelligence. One is called the strong AI, the other is called VKI paradigm. The strong AI means that they believe that an actual thinking machine can be built.
Skip to 5 minutes and 12 secondsThe approach there is that we have these brains and the mind, which is the thinking part, is the consequence of the brain. They also believe that the thinking can be perfectly done in an algorithmic way. Basically, they believe that thinking is nothing more than data processing. Now, if we have this kind of approach, we are aiming for building the smart machines and, you will see later on, that there are many attempts to actually do this. However, the VKI does not actually aim for building thinking machines. It aims at building tools that are very useful for people who think in their jobs related to thinking. What I mean is easy to understand through an example.
Skip to 6 minutes and 9 secondsMy grandfather had a pharmacy and when he worked very, very late hours, then, at the end of the day, he called the carriage to take him home. When my father worked very late hours, he called a taxi to take him home. Now what is the difference between the two? In terms of the strong AI, if you wanted to build the artificial means of getting you home, you should build an artificial horse. You should model the muscles of the horse, the legs of the horse, and so on, and build an artificial horse. The VKI approach says that we don't understand how the horse works and we don't try to understand it.
Skip to 6 minutes and 55 secondsWe will build something new that can perform the same task, or a similar task, and take you home. So the same way in VKI, we don't believe in making thinking machines, but we do believe in making very useful tools that can make our thinking easier, better, and improve our performance. One last thing to talk about is one realisation of the artificial intelligence called artificial neural networks. Now, the premise is that we can build artificial neurons. These are faster and better than those biological neurons that we have in our nervous system and, therefore, if we build that kind of network, those will be thinking machines and will be thinking faster, therefore, will be more intelligent than the human beings.
Skip to 7 minutes and 51 secondsNow let me tell you some facts about this which suggest to me something and you can make up your own mind. So how many neurons do you think we have in our brains? It is in the range of 100 billion. One neuron, on average, has about 7000 synapses, 7000 connections to other neutrons. What that means, it means that we have altogether more than 1,000 trillion connections in our brains. It is 10 to the power of 15. It is even difficult to pronounce it. Now what we have today, in terms of artificial neural networks, a really large one would connect around 100 neurons with one to two dozen of connections for a neuron.
Skip to 8 minutes and 43 secondsNow I believe that this is not a matter of quantity. This is not just about the order of magnitude, if we connect more. I believe that this is a qualitative difference and that we are trying to do something that cannot be done. But, of course, today, what we can see only means that we will not find out about this anytime soon and we will need to choose what we believe until then. There will be more to come about the smart machines later on. Thank you very much.
He paraphrases the Andersen’s ‘The Emperor’s new clothes’, which was about dressing up the emperor into nothing, while claiming that it was a magical textile that only the worthy would see and therefore nobody admitted not to see it. Finally a child said that the emperor was naked. Now the new mind of the emperor is similar to his new clothes. In the prologue of this book Penrose tells the reader about a mythical supercomputer Ultronic, which was set up in the Grand Auditorium, and the citizens were invited to ask any question. As the computer knows everything, it should be able to answer any question. People felt “bashful, afraid to seem stupid”, intimidated by such a super-smart entity, so nobody dared to ask any questions… until the thirteen year old Adam dared… (the story is concluded in the Epilogue of the book). So Adam asked the super-smart computer what does it “… FEEL LIKE”, to which Ultronic responded that it didn’t see what and couldn’t even understand what Adam was getting at and then everyone was laughing. The supercomputer with 1017 logical units, which was supposedly making it more intelligent than the combined intelligence of the whole country, could not handle such a difficult question. Or, to get back to Theodore Roszak, the computer is a brilliant solution still in search of a problem. If we demystify our computers and dress them up as information processing tools, we can find them incredibly useful – and not so threatening.