Artificial Intelligence: Building Smarter Machines
As the cameras rolled for the Jeopardy broadcast Watson seemed to stand right there between its competitors But its television presence a logo on a computer screen was just for show The real Watson comprising ten racks of computer servers was located off-site Instead of listening to the statements read by Trebek Watson received its clues as digital text that it analyzed at lightning speed Its performance wasn’t perfect but it quickly established a lead over its human opponents By the end of the third day it had won the 1 million purse Ken Jennings famous for his seventy-four-game winning streak on the show took his defeat in stride I for one welcome our new computer overlords he said For those who fear that machines really could come to dominate humans Jennings’s statement hit close to home Watson’s victory raises difficult questions Did Watson really think Did it understand its participation in a quiz show Can machines acquire self-awareness If so is this a promising or a dangerous situation for humans Scientists and technologists have come down on both sides of the debate Some such as computer scientist Raymond Kurzweil director of engineering at Google predict that artificial intelligence will improve life for all humanity and may even extend the human life span indefinitely Others such as British physicist Stephen Hawking fear that the age of supersmart machines could threaten our very existence on Earth Although no machine not even Watson comes close to matching human intelligence many scientists believe it is only a matter of time before technology reaches this milestone STRONG AND WEAK Experts on artificial intelligence divide the machines into several types An artificial general intelligence AGI also called a strong artificial intelligence would be a true thinking machine able to learn on its own and modify its own programming without human input In theory an AGI would be able to solve any problem that a human could decipher 6 Artificial Intelligence
DEEP BLUE Jeopardy was not the fi rst game in which humans faced a machine challenger In the late twentieth century various organizations staged chess competitions pitting human players against computers with the humans always coming out on top A much publicized six-game match between IBM’s Deep Blue computer and Russian world chess champion Garry Kasparov in 1996 also ended with a human victory After Deep Blue’s defeat IBM scientists began improving the computer for a rematch the next year Although the updated computer could review two hundred million possible chess moves each second and choose the ones most likely to lead to victory Kasparov won the fi rst game Deep Blue took the second game but games 3 4 and 5 ended in draws Tension mounted as the fi nal game began on May 11 1997 When Deep Blue won the game thereby winning the match two games to one the entire world took note For many the victory highlighted in a surprising and vivid way the growing power of computers It would not be narrowly tailored to perform a specific task such as analyzing the stock market or conducting computer searches Instead it would be able to deal with a broad range of issues Over time as it accumulated knowledge an AGI would get smarter and smarter What worries many experts and excites others is the possibility that an AGI might keep improving until it morphed into a super intelligence Such an entity would far surpass human intellectual abilities But it might not share human social and ethical values such as notions of fairness justice and right and wrong Without exceptionally careful programming and built-in safeguards a supersmart machine could become uncontrollable An example would be an intelligent robot weapon capable of selecting and firing on targets in wartime without Dawn of a New Age 7
human oversight We can imagine such technology outsmarting financial markets human stock traders out-inventing human researchers out-manipulating human leaders and developing weapons we cannot even understand wrote Stephen Hawking with other concerned scientists in 2014 If on the other hand a superintelligence could be programmed to share human goals the result could help humanity Such a machine might be able to devise remedies for air and water pollution climate change caused by the burning of fossil fuels and life-threatening illnesses No such promise or peril looms over the development of weak artificial intelligence a phenomenon that is already at work in our homes cars and workplaces In contrast to artificial general intelligence weak AI focuses on specific tasks Weak AI is currently transforming such diverse fields as transportation medicine and banking Apple’s digital virtual assistant Siri operates through weak AI So do unmanned land rovers that explore and photograph the surface of Mars Other examples include apps and programs that conduct Internet searches identify faces and objects in photographs and translate text and spoken words into foreign languages Self-driving cars refrigerators that regulate temperature and monitor food quality and smart houses that track household lighting temperature fire safety and home security are also examples of weak artificial intelligence Weak AI has already displaced large numbers of laborers in the workforce In modern factories robots machines that automatically perform complicated and repetitive tasks handle much of the assembly work work previously performed by humans Automatic teller machines online banking and bill paying and scanning software do much of the work that bank clerks and tellers used to do Self-checkout stands at supermarkets have replaced many human cashiers The replacement of human workers by machines will only accelerate as technology improves Many analysts fear that millions of workers will be left jobless as fewer and fewer tasks will require humans to perform them 8 Artificial Intelligence
INTELLIGENCE TESTS Can machines think British computer pioneer and mathematician Alan Turing right was already considering this question as early as 1950 He devised what came to be called the Turing test to determine a machine’s intelligence The procedure is simple A person in isolation poses questions and receives answers from both another human being and a computer If the questioner can’t determine which answers come from the human and which come from the computer a judge can assume that the machine is intelligent Turing believed that by the year 2000 with advances in AI computers would fool judges at least 30 percent of the time To pass the Turing test an artificial intelligence must give answers that lead the questioner to believe that he or she is interacting with a human British computer pioneer and mathematician Alan Turing devised the test in 1950 Some researchers say that the Turing test has serious fl aws Even though a computer may engage in a seemingly normal conversation they argue we can’t assume that it is actually thinking They say that other tests are better able to assess computer intelligence One such test is the Winograd Schema Challenge Devised by Hector Levesque of the University of Toronto in 2011 the test presents questions that require reasoning skills A sample question might be what is meant by the sentence The trophy would not fi t into the brown suitcase because it was too small A person would have no trouble explaining that the suitcase was too small for the trophy But a computer might have trouble deciphering whether the pronoun it refers to the trophy or the suitcase So more than language analysis is needed to make the determination Common sense is also required In other tests computers watch programs on television or YouTube and then answer questions about them For instance a computer might view a news program and then be asked why does the congressperson think we need a new law After watching a sitcom the computer might have to explain why a particular joke was funny Experts say that these kinds of tests are much better than the Turing test in assessing a computer’s ability to think reason and analyze information Dawn of a New Age 9
