Google’s DeepMind AI Scored Less At Math Tests Than Highschool Students

AI has been learning to match from assessing MRIs to a game of chess, and when it comes to human experts, it even beat them a few times. However, if the most recent reports that talks about artificial intelligence are correct, the math level of Google’s DeepMind AI, for example, is worse than the one of the 16-year-old sophomore.

The algorithms of Google’s DeepMind were recently tested on a high school Math test, and the results showed that they could not even have the problems translated.

DeepMind AI scored poorly in math tests

DeepMind trained a neural network on the Math curriculum of a 16-year-old UK student in order to test and analyze the capability of its AI in the domain of Mathematics in a bid. They trained the machine on Arithmetic, Algebra, Comparisons, Calculus, Numbers, Measurement, Probability, and Manipulating Polynomials and they synthesized the data.

However, they were disappointed because the models they tried flunked the test. So, how much the algorithms scored? The different algorithms could only solve some 35% of the 40 questions DeepMind threw at them when the percentage should actually have been 100.

However, we can’t avoid saying that one of the models was slightly better than others, but they failed on most occasions to translate the questions, complete with numbers, words, functions, and symbols into actual operations for getting the results.

The fact is that 14 out of 40, no matter where in the world, is considered a failure.

AI systems need to improve at math

Several cognitive skills are required by Mathematics, skills that we humans have and can use automatically, while for this job machines need to be enhanced.

Training them to make sense of a problem, with all the words, numbers, arithmetic operators, variables, and so on would not be enough to make math geniuses out of them.