How much intelligence and how much artificial?
Traditionally intelligence is understood as the ability to think and understand. As the development of artificial intelligence arises, multiple questions are inevitable around what is intelligence and what it means to be intelligent.
Someone who can handle a lot of information, remember a lot of data, is considered smart.
It was quickly considered a feature that distinguished computers from people. The machines are very good handling a lot of information and doing operations. Another question is whether they know what to do with that information or whether they require people.
Artificial Intelligence manages to perform tasks for which people need intelligence
Good chess players are smart. Machines play chess by brute force. They are able to calculate a tree of possible moves and act accordingly.
When in 1996 Gary Kasparov, the best chess player of the moment, faced Deep Blue from IBM was clear he was going to win. “I will beat the machine, whatever happens. It’s just a machine. Machines are stupid.” He knew what talked about, he trained daily against machines. In fact, in 1996 Kasparov won, but lost the first match. And in that game there was a movement of the machine rejecting a pawn seemed more than brute force. “I could feel -I could smell- a new kind of intelligence.” The following year Deep Blue won Kasparov.
The ability to communicate is a sign of intelligence. If we add humor, double meanings or subtlety to that communication, it becomes a human characteristic.
In 2011, Watson from IBM participated in a special edition of the Jeopardy! against two of the best contestants in the history of the program. It is a contest with a dynamic of questions and answers that require a lot of contextual understanding. The philosopher John Searle then wrote that “literally speaking, there is no such thing as computer understanding, there is only simulation”
Intuition emerges from intelligence and experience. Activities that require intuition can only be done by people.
After Deep Blue’s victory over Kasparov, there was only one game that machines could not defeat the best masters. The go, an ancestral game, with a very simple dynamics, but with an extremely complex strategy, for which intuition is required. Google’s AlphaGo managed to defeat a professional player for the first time in October 2015. A milestone. A machine that played “like a human” recalling the Japanese patient style. Even so, he seemed far from being able to defeat Lee Sedol, the best go player, against whom he played in March 2016. Before the game, Lee Sedol ventured a 5–0 or 4–1 result to his favor, although he expected artificial intelligence to improve in the next 2 years. AlphaGo won Lee Sedol 4 to 1.
Creativity is an inherently human capacity. Painting, writing or composing music are activities that can only be done by people.
A new Rembrandt was introduced in April 2016. More precise, a new painting with the style of Rembrandt. A system trained with tens of thousands of fragments of paintings and drawings by Rembrandt created this painting. A few months later Sony, within its project Flow Machines, presented “Daddy’s Car”, a song composed by a machine “in the style of the Beatles”. In September of the same year was presented “Sunspring”, a science fiction short whose script had been written by an Artificial Intelligence with a neural network trained with the scripts of dozens of horror and science fiction films. In short, several examples of machines creating new content better than the vast majority of people. In this respect, I find very relevant the work of Blaise Agüera and Arcas — whom I had the luck to know at Microsoft — about creativity and its linkage with perception.
Artificial Intelligence systems are trained for a specific task, but they lack the ability to learn in a generic way.
Google again, in the hand of DeepMind, published in the Nature journal of February 2015 the results of a general learning system. A system, DQN, that only with the information that is seen on the screen and with the aim of making more points, was able to play, learn and improve any person in the 49 games of the Atari 2600. They are very diverse games ( Cars, fights, platform …) and until June failed to overcome all.
People will always be needed for decision-making in complex situations.
A group of University College London presented in October 2016 a trained system with more than 500 cases of the Human Rights Court of the European Union. The system coincided with the court in its verdict in 79% of cases. Of particular interest is the 21% discrepancy. On the one hand, it can raise the doubt of a lack of human subjectivity, or simply an error. On the other hand, with this system as an assistant, which judge or doctor will dare to make a decision contrary to the machine?
I think it is important to note that most of these milestones have been achieved in the last 18 months, so it shows an accelerated pace of progress of Artificial Intelligence.
If all these cases really can be considered intelligence or if it is only a simulation, as John Searle commented when Watson won the Jeopardy!, it is a very interesting debate from a philosophical point of view. But for practical purposes it is irrelevant. In fact, I think it is likely that we will redefine the meaning of intelligence so that it becomes something exclusive to people.
The reality is that the amount of activity that machines will be able to perform better than anyone is overwhelming. In short, artificial intelligence manages to perform tasks for which people need intelligence.
(en español aquí)
