Expert poll: when artificial intelligence will surpass human
Artificial intelligence is gradually taking over the world. So far, we are talking only about the weak form of AI, which, for example, can perfectly play go or recognize images, but which is incapable of much of what a person can do. Neural networks, cognitive systems and other software platforms already help a person drive a car, predict the weather, direct financial flows, assess the level of information security and diagnose patients of various kinds of medical institutions.
Can a computer ever surpass a person in everything that is now considered our prerogative? If so, when will this happen? According to a survey of experts, this will be necessary, but when - a question, the answer to which requires a thorough analysis and assessment of current trends. Employees of the Institute for the Study of the Future of Humanity of Oxford tried to make such a forecast.
Scientists carefully analyzed the work of world experts in artificial intelligence and took several interviews from leading experts in this field. A preliminary assessment of the answers showed interesting.
In total, more than one and a half thousand scientists were interviewed, of which 352 responded. Then the answers were studied and the average figure was derived. Experts' forecasts show that AI will become better than a person in translating languages by 2024, write school essays by 2026, drive trucks by 2027.
In other areas will have to train for a long time, experts say, artificial intelligence. So, it would be better for a person to sell AI goods no earlier than 2031, to write bestsellers - not earlier than 2049, and to conduct operations - not earlier than 2053.
Perhaps the experts are wrong. For example, it was previously believed that AI would win the best of people in go without handicap no earlier than 2027. AlphaGo system beatThe most famous champions for this game are already in this decade. Lee Sedoll was the last person to beat this kind of computer system. Perhaps the forecast is erroneous and ten years behind reality.
Today, scientists give a 50 percent chance that in 45 years, AI will become better than a person, if not in everything, then in the vast majority of areas and disciplines. On the other hand, experts may be overly optimistic, since there is no guarantee that AI will not encounter any difficult obstacles to overcome and then forecasts may not come true at all.
Interestingly, scientists from different countries gave different temporary assessments of improving AI capabilities. Experts from North America believe that a computer will surpass a person in everything in about 70 years, in Asia they hope that this will happen in 30 years. It is not clear why there is such a discrepancy in the temporal assessment. Maybe scientists from Asia know something that their North American colleagues do not know.
The most interesting answers to these questions were given by Robin Hanson, Martin Rothblatt, Raymond Kurzweil.
Hanson believes that after about a hundred years, the capabilities of computer systems will equal or surpass human capabilities. But this is only if a person can create emulators of the human brain that will allow machines to think in much the same way that a person does. If this does not happen, then the evolution of AI is delayed for a period of 2 to 4 centuries.
Hanson also says that if a quantum leap in the development of AI does occur, it will positively affect the future of a person - for example, the economy will develop at an unprecedented pace.
Martin Rothblatt claims that in a few decades it will be possible to create a digital copy of a person, as a result of which people will stop dying. The physical shell will go away, of course, but the digital copy will become eternal.
Raymond Kurzweil is confident that the computer will surpass man by 2029. At the same time, he agrees with colleagues that AI will be able to help us overcome the most dangerous diseases, improve food supply, develop the economy and improve the environment. Kurzweil claims that it would be nice to keep AI under control so that computers benefit a person without any side problems.
Full text of the study: arXiv: 1705.08807 [cs.AI]