A new report from the Massachusetts Institute of Technology (MIT), Stanford University and OpenA's Center is trying to answer two questions that cause the insomnia of politicians, economists and researchers in artificial intelligence: how much artificial intelligence is there? What is the speed of its progress?
This week, the AI Index (Artificial Intelligence Index) was published this week - who began by saying that we "shaved like an eyeliner" in our estimates of artificial intelligence. He concluded with two main points. First, the field of artificial intelligence is more active than ever before, with brains and money pouring at an astonishing rate. The second is that although artificial intelligence outweighs humans in performing a few very specific tasks, it is still very limited in the realm of "general intelligence".
To reach these results, the Artificial Intelligence Index has been examined in a number of measures of progress, including "activity volume" and "technical performance". The first factor examines how everything happens in the field, from attending conferences to joining classes to capital investment and startups. The short answer here is that everything happens a lot and at a rising pace.
The other factor, "technical performance", attempts to measure the capabilities of artificial intelligence to outperform humans with specific tasks, such as object recognition in images and speech decoding. Here the picture is more accurate.
There are certainly tasks in which artificial intelligence already does or surpasses human performance, including the recognition of objects common to images (in the test database the human error rate was 5% vs. 3% for the machine) and decoding (as of 2017 a number of artificial intelligence systems Decoding speech at the same rate as human error) Artificial intelligence also outperforms a number of games such as the "Quiz" and some Atari games such as "Pacman" and the famous game "Go".
But, according to the report, these measures give us only a partial view of artificial intelligence. For example, it is easy to train artificial intelligence in the field of video games, where he can always see everything that happens, but can not say the same thing about other tasks where we put artificial intelligence such as the management of transport infrastructure in the city, although researchers began to develop Video games reflect these challenges.
The same is said of functions such as "voice dictation". Artificial intelligence may be the precision of a human being when it comes to recording recorded dialogue, but he can not distinguish ridicule or identify jokes, or calculate another million pieces of cultural context that are crucial to understanding even the most spontaneous conversations. The artificial intelligence index recognizes this, and adds that the biggest problem here is that we do not have a good way to measure this kind of logical understanding. There is no intelligence test for computers, despite some claims.
However, this does not mean that we do not fear the impact of artificial intelligence on society. Although the most advanced artificial intelligence systems are considered stupider than the rat (according to Yan's Artificial Intelligence Chief), this will not prevent them from having a significant impact on our lives, Employment.