Tag: AI

Even An AI Supercomputer Found This College Entrance Exam Tough

If you are getting stressed about upcoming exams then youre not alone, so is this artificially intelligent (AI) machine.

Last week, a top AI system was pitted against nearly 10 million students to face the maths paper for a much-feared Chinese university entrance exam, known as gaokao. Unfortunately for robotkind, its results were pretty mediocre.

The computer a humming tower of eleven servers with no Internet connection called AI-MATHS scored 105 points out of 150 points. On another version of the test, it scored 100. Although that beats the passing score of 90, humanities students had previously scored an average of 109 last year.

That said, the machine finished the exam in 10 minutes when humans are given two hours to complete the exam.

Scientists recently saidartificial intelligence will be able to beat humans at everything by 2060, whether that’squizzes, exams, chess, or the game Go. In response to the study, Elon Musk then tweeted that he believes AI-superiority will actually be earlier, around 2030 or 2040.

That doesnt mean this AI is slow off the mark, however. The computer itself would be able to deal with raw numbers with no problem. Instead, the purpose of this task was to understand the examination in terms of language, something that computers are not so sharp with at the moment.

“This is not a make-or-break test for a robot. The aim is to train artificial intelligence to learn the way humans reason and deal with numbers,” said Lin Hui, CEO of Chengdu Zhunxingyunxue Technology, who developed the AI, according to Chinese news agencyXinhua.

For example, the robot had a hard time understanding the words ‘students’ and ‘teachers’ on the test and failed to understand the question, so it scored zero for that question.

Gaokao isinfamously rigorous and renowned for being overwhelming stressful for the young people that take it. Made up of four three-hour papers in Chinese, English, mathematics, and a choice of either sciences or humanities, the series of tests rely on an extensive range of knowledge, problem-solving skills, and obscure creative thinking. The mathematics exam itself is said to be about as tough as the same level college exam in the West.

Nevertheless, the researchers continue to work with China’s Ministry of Science and Technology and remain optimistic their AI will improve in the exams in no time at all.

I hope next year the machine can improve its performance on logical reasoning and computer algorithms and score over 130,” Lin added.

Read more: http://www.iflscience.com/technology/even-an-ai-supercomputer-found-this-college-entrance-exam-tough/

Technorati Tags: , , ,

Google’s AlphaGo Beats Go Champion 4-1 In Landmark Victory For AI

 

AlphaGo, Googles artificial intelligence (AI), has trounced the current world champion Lee Se-dol at the game of Go, four matches to one. Despite having very simple rules, this ancientgame has trillions upon trillions of possible moves and permutations, so AlphaGos victory is seen as a landmark in the evolution of AI.

The tournament which took place in Seoul, South Korea showcased the AIs formidable calculating power, which is part of Googles overarching DeepMind program. Back in October of last year, AlphaGo defeated the reigning European Go champion five games to zero, and it has now risen to the very top of the field.

Mr Lee said after that, although the defeat had “challenged” his ideas about the game Go, he did not necessarily think AlphaGo was superior to humans, reported the BBC.After the comprehensive win,Google said it would donate the $1 million prize to various organizations, including UNICEF, STEM (science, technology, engineering and mathematics) charities and a few Go groups.

Go is a 2,500-year-old Chinese game that is widely considered one of the most complex games every developed. Its rules are relatively simple; each player has to use their beads, black or white, to surround their opponents beads on a 19 x 19 grid. Interconnected beads demarcate individual players territory, so at heart, it is somewhat similar to chess.

However, chess is far less complicated than Go, mathematically speaking. Although the number of possible moves in chess is ludicrously high, in Go, there are around 1 x 10170 possible configurations thats the number one with 170 zeros after it. To put it another way, there are more configurations than there are atoms in the universe.

AIs have long been able to beat expert human chess players at their own game, as was most famously demonstrated by world chess champion Garry Kasparovs defeat to the IBM supercomputer Deep Blue in 1997. AlphaGo has spent a long time playing the game Go against older versions of itself, and its recent tournament victories has cemented its rise as a formidable intellect in this regard.

To the relief of human Go players everywhere, Mr Lee did manage to win the fourth match, despite being left speechless after losing the second to AlphaGo, which played a nearly perfect game. According to BBC News, Mr Lee said that he has never been congratulated so much because Ive won one game.

Mr Lee’s sole win is particularly impressive given that, just beforehand, AlphaGo was granted divine status by the Korea Baduk Association, a governing body for the game of Go. This daunting ranking implies that the AIs gaming ability has bordered on god-like.

Although it is somewhat of a stretch to say that AlphaGo will become sentient and take over the world, this AI wunderkind is a perfect example of why many predict robots will soon take over many human-run jobs, particularly those that involve tasks with repeating patterns. It begins, as some may say.

Many have started speculating what challenge it, or indeed similar AIs, may take on next. Some have suggested poker or even video games like StarCraft, far more advanced games that humans will have a distinct advantage in for now.

Read more: http://www.iflscience.com/technology/googles-alphago-beats-go-champion-4-1-landmark-victory-artificial-intelligence

Technorati Tags: , , , , , ,