It's Saturday, June 7, and the keyboards are ablaze in the Royal Society of London room, with volunteers exchanging text messages with unknown users. With their questions, they seem to want to get to know their interlocutors – for example, their name, age, musical tastes or hobbies.
In reality, however, they are trying to understand whether the person answering is a human or a chatterbot, that is, an “interactive” artificial intelligence (AI) program. As part of the experiment organized by the University of Reading, five chatterbots were subjected to the “Turing test”, a test invented by the British mathematician Alan Turing in 1950, in order to determine whether a machine is as intelligent as a human.
As the test predicts, this would apply to any program that managed to "fool" a sufficient number of volunteers in the five-minute dialogues it had with them. The Russian software "Eugene Goostman" managed to do this in 33% of cases, impersonating a 13-year-old boy from Ukraine. "Eugene Goostman"'s performance is a record and for the first time exceeds the 30% limit set by the mathematician, making the University of Reading declare it the first software to pass the "Turing test". The news went around the world as a step closer to the era in which machines will be "endowed" with intelligence like ours. However, some wondered whether this era would prove disastrous for humans, reminiscent of the HAL 9000 computer from "A Space Odyssey".
From the Matrix to Hawking
Many media outlets recalled "A Space Odyssey" a few days ago, along with "The Matrix," "The Terminator," and "I, Robot." The common denominator of the films is that they describe a future era where "thinking" machines turn against their creator. A scenario that came to the fore following Stephen Hawking's statements that "the full development of artificial intelligence could spell the end of the human species.".
The statements were made to the BBC in early December, on the occasion of new synthetic speech software developed by Intel for him, which "learns" how the famous physicist expresses himself, to indicate which words he might want to use and, if he gets the "green light" from him, to pronounce them. Hawking does not deny that such "primitive" applications of artificial intelligence have proven useful, but argues that man will be in danger in the future if he develops a "thinking" machine. "It could improve itself with geometric progression, while our species evolves very slowly," he said.
Hawking agrees with Elon Musk, the man behind the first electric car company Tesla and the space vehicle company SpaceX, who has warned of “perhaps the greatest threat to our existence.” “It’s like making a deal with the devil,” he said last October at an MIT conference, while eight months earlier he had invested $40 million in the artificial intelligence company Vicarious FPC, along with Mark Zuckerberg and Ashton Kutcher, “to closely monitor developments,” as he told CNBC at the time.
In 30, 40 or 60 years?
Both Musk and Hawking are talking about what will happen once scientists develop the first machine with "strong" artificial intelligence, meaning one that is as intelligent as a human. This milestone has been called the "technological singularity" since the 1950s, and one of the best descriptions of it has been given by Ray Kurzweil, an analyst, inventor, and director of engineering at Google.
As he writes, for example, in his book “The Singularity Is Near” (2005), from that moment on the machine will begin to improve without any intervention, with the result that its intelligence will surpass that of all of humanity. However, although Kurzweil expresses reservations about whether algorithms can be developed so that such machines also have a moral code, he is not worried that they will want to enslave our species.
But when will artificial intelligence reach human intelligence? The Google CEO predicts that it will happen in 2045, while other scientists predict that it will happen in 60 years from now at the latest. In his book “Superintelligence: Paths, Dangers, Strategies,” Nick Bostrom, director of the Institute for the Future of Humanity at Oxford University, lists the results of four polls in which 170 scientists were asked when it is likely that machines will become so intelligent that they will take over almost all human jobs. The predictions reached 50% for 2040 and jumped to 90% for 2075.
"They will lack self-awareness"
There is no shortage of experts who have opposing views – even for “Eugene Goostman” scientists argued that the digital persona distorted the rules of the test. The same applies to predictions about the “technological singularity”, which, for example, Professor Tony Cohn from the University of Leeds, speaking to the BBC, postponed for a century later.
On the other hand, other scientists argue that it will never happen. "What will always be missing from machines is consciousness, that is, the ability that humans have to understand themselves. Besides, in order to encode human intelligence in algorithms, we would have to know exactly how it works, which I think is impossible to do," Dr. Dimitrios Kalles, president of the Hellenic Society of Artificial Intelligence and assistant professor at the Open University, tells "K".
Dr. Ioannis Vlachavas, professor of Computer Science at the Aristotle University of Thessaloniki, agrees that consciousness will always be what will distinguish humans from their creations. "Even if a software evolves to answer the most complex questions, it will never acquire self-awareness, that is, it will never 'know how it knows', like a human. Also, initiative will always remain a human characteristic, in the sense that a machine will never manage to do something different from what has been programmed," he tells "K".
Educating… programs
All researchers agree that artificial intelligence has been experiencing an “explosion” in recent years. From a technological perspective, an important reason is new approaches that make programming more flexible. “Essentially, these are “machine learning” techniques, that is, algorithms that show a software how to infer the patterns that are repeated in the huge amount of data that we feed it,” explains Mr. Vlachavas. Thus, for example, in 2002, Google “trained” a program to recognize… cats, after it had “seen” 10 million snapshots of the lovable four-legged friends on YouTube. In the same way, IBM’s Watson supercomputer managed to beat the two best players on the Jeopardy quiz show a year earlier.
To the new approaches, one should add the huge funds that technology giants are investing in artificial intelligence – “the “cache” of an expert has become about the same as that of a rising rugby star,” wrote Business Week in 2013, commenting on the “transfers” of scientists from Silicon Valley companies. Companies that already have driverless cars in the works, and even applications that “borrow” Watson’s… IQ, helping doctors with diagnoses.
At the same time, however, a recent UN session issued a resolution warning of the dangers of the proliferation of “smart” weapons systems that can identify targets and destroy them, without human intervention. “In addition to intelligence, all the applications that come will also reflect the intentions of those who developed them, which in some cases will be at least controversial. In this sense, I believe that the conversation that Hawking opened is essentially just beginning,” concludes Mr. Kalles.
Milestones in artificial intelligence
1950: In his article "Computing Machines and Intelligence," English mathematician Alan Turning formulates a thought experiment that a machine would have to "pass" to prove that it is truly intelligent. The "Turning test," as it was called, is still used for this purpose today.
1956: The first conference on artificial intelligence is held in Dartmouth, which also establishes the term.
1963: Entitled "Computers and Thought" and edited by Edward Feigenbaum and Julian Feldman, the first collection of articles on artificial intelligence is published.
1968: A chess software, MacHack, is being developed at MIT, the first to ever take part in a tournament.
1979: As part of his PhD at Stanford University, Hans Moravec developed the Stanford Cart into the first autonomous vehicle. The vehicle was able to navigate a room full of chairs, avoiding obstacles without external intervention. It took 5 hours.
1997: IBM's Deep Blue supercomputer defeats world chess champion Garry Kasparov.
2005: An autonomous vehicle developed by Stanford University wins the second Grand Challenge of the American Darpa, traveling 211 kilometers without a driver in the Mojave Desert.
2011: IBM's Watson computer beats the top two players on the American quiz game "Jeopardy.".
2011: Apple releases the iPhone 4S with Siri, the first "intelligent assistant" on a mobile phone.
2014: The "Eugene Goostman" software is declared by the University of Reading to be the first software in history to pass the "Turing test".
Source: kathimerini.gr
