Technological heart and brain of CIMON, the astronaut robot, Watson is IBM’s artificial intelligence system. Became famous in the 90s for having beaten the world champion of chess, today he works alongside scientists in the fight against cancer and in the design of tomorrow’s robots. But he also enjoys cooking.
To carry out all the delicate tasks assigned to him on the ISS, CIMON needs a big electronic brain. Much larger, even from the point of view of size, than what ISS could host inside it.
In fact, the technological heart of the robot is not on the ISS, but on Earth, and is called Watson: it is an artificial intelligence system developed by IBM that can recognize natural language, answer questions and learn automatically.
These three capabilities, combined with a great speed of calculation and data analysis, allow Watson to process information, formulate answers and provide solutions to the various problems that are submitted to it.
Unlike traditional computers, which mechanically perform a series of logical and mathematical operations and make choices based on rigid decision trees (IF a certain condition occurs, THEN perform this instruction), Watson behaves more like a human being , providing the most correct, or statistically probable, answer based on the available data.
From a practical point of view, what Watson does is to predict the right answer to a question based on the statistical analysis of the information provided to him.
The question can be of any kind: “Who is the human that I am facing?” “Which fans should I turn on and how long to make CIMON float in the void?”, “How do I solve a Rubik’s cube?” “What can I cook with these 5 ingredients I have in the pantry? “
In order to respond to these, but also to many other questions, Watson needs to be trained: that is, he has to receive a great deal of information concerning the topic on which he will be called to pronounce himself.
For example, in order to identify the “cat” animal and distinguish it from a dog, a fish or a panther it needs to train to recognize the most different cats, of various colors and races, in different positions and seen from every angle . Only after having deepened the knowledge of the “cat” object will be able to say, with a good margin of certainty, if that animal with pointed ears and tail is actually a cat. The same learning – processing – analysis scheme allows Watson to answer any questions. Provided obviously to have been properly prepared.
The strength of Watson is to be able to use, in its learning process, both structured data and unstructured data, for example newspaper articles, web pages, videos and photos.
To prepare Watson on the subject “cat” could for example be enough to feed him an encyclopedia on felines, a collection of images of cats found on the web, or tweets with hashtag #cat published online.
Being able to recognize natural language, the system is able to extrapolate the meaning of a sentence based on grammatical rules, structures and cultural filters.
Compared to a human being, Watson is able to analyze and process a number of different orders of magnitude higher and in an infinitely shorter time.
In 2010 he beat the human samples of the Jeopardy TV quiz, systematically anticipating them in the booking of the answer. To prepare for the test, the IBM researchers had provided him with the equivalent of over 200 million pages of texts and all the content of Wikipedia.
A fundamental point in Watson’s preparation is therefore the creation of the data set that will be used by the artificial intelligence system as a basis for making decisions.
The creation of this entity is largely entrusted to human beings: it is a delicate operation, because if the system is fed with wrong, inconsistent or outdated information, it will provide untrustworthy answers.
During the real learning phase Watson is joined by one or more experts of the subject on which he is preparing, which helps him to structure this large amount of data in the form of questions and answers.
This operation does not serve to provide the system with pre-packaged answers to the most trivial questions, but it is essential to allow Watson to learn the linguistic patterns and terminology of that specific domain.
This process continues unabated, and with each interaction the system becomes increasingly good and always knows the subject better. Watson’s knowledge is however periodically updated and verified by human experts, so that the reliability of his answers is always at the highest levels.
At the end of what is therefore a real path of study Watson will be able to answer complex questions on the subject, providing solutions and formulating suggestions supported by certain data.
But he will also be able to critically analyze the information at his disposal to identify recurring patterns.
To answer the questions, Watson uses a rational process similar to that used by scientists: he extracts the fundamental elements from the question and based on the data available to formulate the possible answers, each supported by specific evidences. Each possible solution is then assigned a reliability value based on the data used.
Compared to any group of humans doing the same type of process, Watson is much faster, can analyze a much larger amount of data and provides neutral answers, not influenced by any kind of cultural or psychological bias.
But how is Watson made? From the hardware point of view, Watson consists of 90 IBM Power 750 servers. Each of them uses a Power7 octacore processor. In total, the system has 16 terabytes of RAM and 2,880 threads of Power7 processors available.
Watson can process 500 gigabytes of data per second, equivalent to the content of 1 million books.
Awesome, is not it?
Images sources (all cc0):