Robots Are Now Beating Us At Reading Comprehension Tests Too


James Felton

Senior Staff Writer

clockJan 16 2018, 13:09 UTC

Vasilyev Alexandr/Shutterstock

At some point, we're going to have to accept that robots are just better than us at most tasks. They can beat us at chess, they can beat grand champions at Go with only three days of practice, and they are better than us at a whole host of other activities including backflips.


We thought that one thing was safe (for now). No matter how advanced they got, we thought we were still the biggest bookworms on the planet. Now it seems robots are trying to take that away from us too.

An artificial intelligence (AI) algorithm has outperformed humans for the first time in a reading comprehension test. The algorithm, which uses natural-language processing, managed to beat human scores on the Stanford Question Answering Dataset (SQuAD). The SQuAD test is a reading test used around the world, made up of over 100,000 questions, based on 500 Wikipedia articles. The test is designed to test whether machine-learning artificial intelligence can process large chunks of information and supply precise answers in response to questions about the information.

The AI, developed by the Alibaba Institute of Data Science of Technologies, managed to beat human scores by a score of 82.44 to humanity's measly 82.305. The company says this raises the prospect that the algorithm could be used to automate human jobs.

A part of the reading comprehension test that robots beat us at. Stanford.

Chinese tech giant Alibaba says that AI could be used to replace jobs in a wide range of industries.


“Objective questions such as ‘what causes rain’ can now be answered with high accuracy by machines,” chief scientist for natural language processing at the Alibaba institute, Luo Si, said in a statement.

“The technology underneath can be gradually applied to numerous applications such as customer service, museum tutorials and online responses to medical inquiries from patients, decreasing the need for human input in an unprecedented way.” 

To add insult to injury, Microsoft says they too have developed AI that can beat our feeble human brains. Their own deep neural network scored 82.650 on the same test.


Microsoft said that the AI could be used to automate parts of even more complex jobs, such as getting the AI to read through legal documents to look for legal precedent, or scanning large documents for medical findings on the behalf of doctors.

It might be time for us to invent new things we're great at, and that robots have no chance of beating us at any time soon. Like feeling feelings, or efficiently digesting carbs.