PhilDiscuss™ - The Philippine Discussions Community Website® 

  • Google’s DeepMind AI Fails At School Test For 16-Year-Olds

  • Forum related only about Science & Technology. News and Update.

Announcement: We are still developing new features, but the website is rock solid and ready to dive. Progress: 99%

Forum related only about Science & Technology. News and Update.


 #13861  by Ootak
 2 months ago (Wed Apr 10, 2019)
There are a lot of people who fear that AI will soon conquer the human world. We have already seen AI winning in the game of chess, assessing MRIs, and sometimes even beating human experts at their jobs.

So the paranoia around it is obvious. However, if a new report is to be believed, artificial intelligence still has a lot to learn as it appears to have failed a 16-year-old level math problem.

School isn’t easy; not even for an AI [cries in binary 🙁 ]
Google’s DeepMind put its algorithms to the test by subjecting it to a high school math questionnaire for 16-year-old students in the UK. To everyone’s disappointment, the deep learning neural networks couldn’t even translate the problems.

DeepMind trained its neural network algorithms with data on Numbers, Algebra, Arithmetic, Comparisons, Measurement, Calculus, Manipulating Polynomials, and Probability, for the test.

The test comprised 40 questions aimed at different algorithms but they could solve only 14 out of 40. One of the incorrectly answered questions was:

What is the sum of 1+1+1+1+1+1+1?
While this is a no brainer for us humans, the neural networks couldn’t answer it.

Although, there was one model which performed slightly better than the others. However, most of them encountered difficulty in translating words, symbols, numbers, and functions in the questions.

DeepMind AI doesn’t really suck
DeepMind’s AI isn’t absolutely useless. Apart from the math test where it sucked, the neural networks have professional Go and StarCraft players. The AI has also helped Google reduce power consumption at its data centers by 15%.

On many occasions, DeepMind algorithms have even left doctors behind in detecting diseases.

No shaming AI, okay?
We cannot shame AI for flunking a math test because such calculations require a higher degree of cognitive skills that we humans have and, therefore, it comes to us naturally. However, machines need to be upgraded to acquire such skills.

Neural networks not only require training on how to comprehend problems but also work out those problems — by applying knowledge of numbers, operators, words and mathematical rules to obtain results.
Topic Reactions