These researchers are determined to create machines that go against us and they are getting closer and closer to achieving it.
Years ago, a stand-up comedy club whose name right now I am unable to remember defended that what gave the most pleasure in the world was not “neither food, nor money, nor having your back scratched … What gives the most pleasure in the world is to be right”. That it was part of the low-key allure of robotics: to have a cornucopia of pots that would listen to us and agree with us.
However, from the beginning, science fiction warned us that it was not going to be easy. Do not forget that already the first work of fiction that included robots, ‘R.U.R‘ of Karel Capek, already ends dozens of these artificial beings organizing a revolution and destroying humanity. Sorry for the spoiler, but it’s important: especially since there are teams of researchers working on that, on training artificial intelligences to oppose us.
Can an AI beat us in a debate?
The “good” news is that they still don’t do it well enough to win us over in the middle of the debate, but they don’t do it wrong. Noam Slonim’s research group just presented in the magazine ‘Nature’ an autonomous agent who can participate in a competitive debate with humans. And it is not a trivial matter.
Up to now language-related ‘imitation’ work has yielded mixed results. On the one hand, we have succeeded in making algorithms capable of understanding the emotionality of a sentence; on the other, things like summarizing texts or engaging in dialogues have turned out to be much more complex.
As explained by ‘Nature’, Project Debater (that’s what the algorithm set is called) scans an archive of 400 million newspaper articles and Wikipedia pages to craft arguments and counterarguments to use in a debate with humans. The team put it to the test in a debate on free preschool education and used the transcripts of the same for a jury to score the blind statements.
Project Debating got good scores, although he has not yet managed to win a debate. However, as is often the case, distances are gradually being shortened and these small steps bring us closer to a scenario where artificial intelligences can participate in complex human activities such as destroying the world or, who knows, winning an electoral debate.
Picture | Adam Lukomski