Alan Turing, a well know mathematician and computer scientist, has always been obsessed in finding a solution for the question “Can machines think?”. I think this question hasn’t been answered yet, not because of the lack of technology, considering that machines have been made by human engineers, we perfectly understand how they work. What we don’t fully understand is how our brain work, that gray mass made by countless neurons connected by countless synapses conceal a learning mechanism still unknown even to modern scientists. However, the huge improvements made by computer’s technology during the last seventy years allow machines to behave like a human brain by simulating it thanks to the use of artificial neural networks, or at least, this is what they try to do.
How does it work
In order to prove that a machine can actually think Turing developed a test called Imitation Game. This is a game involving three players, A is a man, B is a woman and C is the interrogator who has to guess the identity of A by making questions to both them communicating only through written notes, possibly typewritten. Player A’s role is to trick C into making him take the wrong decision, while player B has to help the interrogator in making him choose the right one. Of course, player C can’t see neither player A nor player B and A doesn’t necessarily have to tell the truth.
If we replace player A with a digital machine the role of the interrogator is to determine which one is the computer and which one is the human, if player C can’t understand at all who is who we can say that the computer behaves just like a human. Nevertheless, this doesn’t mean that the machine can also think like a human since we don’t know what does “thinking” mean.
Can machines think?
The Theological Objection claims that “Thinking in only a function of man’s immortal soul. God has given an immortal soul to every man and woman, but not to any other animal or to machines. Hence no animal or machine can think”. However, this is a mere speculation, this theory would be more convincing if animals were classed with men having themselves a consciousness too and so the capability to think like we have. It’s different for machines, I support the fact that nowadays’ machines can’t think but I can’t exclude that in the future they won’t be able to do it.
Until now, no machines managed to successfully pass this test but big steps have been done since Turing’s years. As long as proposed questions have simple answers like “yes” or “no”, then is an easy task for modern calculators to answer like a human would do whereas old ones struggled. But what if the question would be a complex mathematical calculus that a human has no way to solve it correctly? Even if we don’t take care of the reply speed (processor computation speed is not comparable with human’s one) a machine will certainly answer with an exact and accurate result, while a human would probably miscalculate it. Even more complicated is if they are both asked not-deterministic questions like “Do you like this painting? Which emotions are you feeling now? Did you enjoy listening to this melody?”. We understand that machines are still far from managing such human-feeling questions but we can’t exclude that in a near future machines won’t be able to answer them.
How do machines learn?
Actually there are two different ways a machine can acquire intelligence, the first one requires the presence of a developer to write code, storing it into machine’s memory and then executing it, this is how a computer normally works, in this way the intelligence of the machine is bounded to the skill of the developer and, above all, the machine can’t independently learn. That’s why the future of artificial intelligence is oriented towards a different technology: neural networks and machine learning. Thanks to this new technique, computers won’t be constrained to follow a precise set of instructions to compute the output but can learn through experience how to calculate it. They only need a set of training examples formed by several inputs with the corresponding outputs and their goal is to learn how to compute the right input given an unknown output, this is process il also known as generalization. Artificial neural networks are formed by a certain number of nodes (neurons) that receive signals and send them to other neurons through their connections (synapses). Each connection has a weight representing its strength and the total input to a neuron is the weighed sum across all inputs from other neurons. A neuron is activated if his input is higher than an activation threshold and then its output is propagated to the other connected neurons. The strategy consists in adjusting the connection weights until obtaining the desired output. That’s why nowadays digital data is extremely important, we can say that these machines are hunger of data, the more data they have the more intelligent they become.
This way of learning is called supervised learning, but what if the training inputs don’t have any associated outputs? Is like when an infant observing the world for the first time, he doesn’t have any past memory, thus he attempts to create them by observing the surrounding environment. A machine, like the infant, tries to observe the significant recurrent patterns within the analyzed data and the algorithms are left to themselves to discover interesting structures in it.
The importance of the big data
Now the problem is how to get such a huge amount of data to feed these algorithms? Data is everywhere and is constantly being sent to a common place: the cloud. Every message we send, every picture we post, every information we put on the web it is stored into a server. Se we can say servers are the containers for storing the data needed for machine learning. If we collect all of these pieces of information and put them together we realize that almost all the knowledge acquired by mankind is available there, that’s why the web is the perfect source of data for intelligent machines and also the key to successfully pass the Imitation Game.
Toward the solution?
Applications exploiting big data are not science-fiction but are already among us, let’s think for example to automatic translators, economic stock prevision tools, advanced illnesses diagnosis and self-driving car prototypes. These machine learning algorithms are already being implemented but the candidates to win the Imitation Game are certainly the voice assistants and the chatbots. In particular the last ones result suitable for playing the role of player A of the Imitation Game, while the human user plays C and B is excluded, this particular one to one version of the Imitation Game is also called “Viva voce Turing test”. The goal remain the same, still the human has to guess whether he is chatting with a machine or with another human. Of course, chatbots are just a simulation of the game given that the human already knows he is talking with a machine. Although most modern chatbots are able to keep a quite good conversation with a human user, though is still easy to recognize they are just machines and lot of improvements have to be done before being able to trick a cunning human.