Short summary - Superintelligence - Nick Bostrom

Scandinavian literature summaries - 2023

Short summary - Superintelligence
Nick Bostrom

Machines have already surpassed man in many areas, but the general level of human intellectual development is incomparably higher. But the day is not far off when machines will become superintelligent.

The Fritz program plays chess much better than the leading grandmasters, but it is not superintelligent, because it surpasses a human in only one narrow area.

Nick Bostrom believes that the superintelligence is an intelligence that is many times superior to the most prominent people in mental development, scientific and technological activity, worldly wisdom and the development of social skills.

The book provides a variety of information about the history of artificial intelligence and the current state of affairs. The author speculates whether the superintelligence will be useful or destructive, friendly or hostile.

Reading this book is very difficult: scientific vocabulary, examples from physics, mathematics, economics and nanotechnology, along with numerous tables and diagrams, do not let the reader relax for a minute. The author himself admits that he tried to make the book easier to understand, but he did not succeed. He also warns that not all information is reliable and has scientific confirmation. At the same time, the book is of interest to those who care about the future of mankind and who are interested in modern technologies.

1. Past and present of artificial intelligence

After the advent of the first computers in the 1940s, scientists started talking about the imminent emergence of a superintelligence. The predictions did not come true, but modern futurists, like their predecessors, believe that superintelligent machines will be created in the near future.

1.1. History of artificial intelligence

1642-1940s - zero generation - the simplest mechanical computers.

1940s-1955 - first generation - vacuum tube computers.

1955-1965 - second generation - computers on transistors.

1965-1980 - third generation - computers on integrated circuits.

1980 - ... - fourth generation - computers on huge integrated circuits.

The fifth generation focused on distributed computing was never created; it was to become the basis for devices capable of imitating thinking.

In the summer of 1956, a group of scientists came to the first artificial intelligence symposium at Dartmouth College, USA. It was this event that became the starting point for research in the field of artificial intelligence, and many participants in that symposium gained worldwide fame. At first, they created only small systems, each of which, in the laboratory, could do something that was previously inaccessible to machine intelligence.

One early system, The Logic Theorist, excelled at proving theorems, with one of the proofs being more elegant than the original.

In the mid-1970s, the fashion for artificial intelligence passed. Sponsors refused to finance projects to create artificial intelligence, because they considered this direction unpromising.

The beginning of the 1980s can be considered the new golden age of artificial intelligence, when Japan launched a major project to create fifth-generation computers. The project was funded by the state together with commercial structures, and its goal was to create a machine with high performance that thinks like a person and can work with large databases. Many countries followed the example of Japan and resumed work on the creation of artificial intelligence. The developers tried to create so-called expert systems - programs that process significant databases, designed to partially replace a person in various fields of activity. Hundreds of such systems were created, the codes for which were manually written by thousands of programmers. However, after a few years it became obvious that it was difficult and expensive to develop, control and update expert systems. Thus, the project to create fifth-generation computers ceased to exist by the end of the 1980s.

Since the 1990s, methods of neural systems and genetic algorithms based on the simulation of the human brain began to develop.

1.2. In what areas is artificial intelligence superior to humans?

By now, machine intelligence has already outstripped humans in many activities. Previously, people naively believed that in order to become a strong chess player, it is necessary to have developed abstract thinking, be a strong strategist, be able to create ingenious flexible plans and try to "read" the thoughts of an opponent. It turned out that the strongest chess player can be defeated by a program with a special algorithm, connected to a powerful processor. However, such artificial intelligence is limited to playing chess.

Checkers. In 1994, the CHINOOK program was able to defeat the reigning world champion. This was the first time that the program actually became the world champion in the intellectual game.

Chess. In 1997, the Deep Blue program won a match against world champion Garry Kasparov, who admitted to seeing glimpses of real intelligence and superhuman ingenuity in the machine.

Scrabble In 2002, the program defeated the strongest players.

Today, these achievements no longer surprise anyone. As John McCartney (the American scientist who first coined the term "artificial intelligence") remarked: "Once artificial intelligence begins to function, people stop considering it to be artificial intelligence."