Reading Robots

Reading comprehension has long been a topic of pride for schoolchildren. Being able to read at a higher level than others their age makes them the coolest kids in the class,* the apple of their parent’s eyes, and outright geniuses in the playground.

But as we roll into 2018 … that could all be about to change.

This week, US multinational technology company, Microsoft, and Chinese multinational conglomerate, Alibaba, announced that they have both independently developed artificial intelligence (AI) software that has the ability to read and comprehend to the same level as adults.

The test was performed against the Stanford Question Answering Dataset, and used a series of made up questions about a set of Wikipedia articles. An average human is expected to receive a score of 82.305 – you’d be forgiven for not knowing if that is a good score or not – and for the first time, AI not only equalled this score, but surpassed it – just. Microsoft’s AI scored 82.650 and Alibaba’s AI scored 82.440.

At first, the merits of this achievement may remain unclear. But it is important to remember just how much potential this breakthrough brings. Just like humans, computers must be able to understand subject matter before offering answers. And while they can easily ingest data that is pre-prepared for them (think databases) it is much harder with loose, unqualified data like that of an article.

Reaching score parity with humans marks a major milestone for search engines, conversational AI, and natural linguistics because for the first time we can simply offer a computer (well, not just any computer) text to process and ask it questions directly. Think “how did Harry catch the Golden Snitch in his first Quidditch match?” In the past, unless someone (a human) had posted the answer online, you’d be out of luck. Now, the answer can come from AI reading the text and understanding it well enough to offer the answer, with no human intervention.

You don’t have to be a computer scientist to understand how useful this might be in the future. Novels aside, imagine the next time you’ve broken down on the motorway and can’t face flicking through the 600 page car manual to find out how to fix the hazard light… wouldn’t it be easier just to ask?

While machine comprehension is nothing new – Facebook, Tencent and Samsung have all previously attempted it – these latest breakthroughs sees both Alibaba and Microsoft step-forward as leaders in the space. But that certainly doesn’t mean the machines are taking over just yet. With scores just 0.1 and 0.3 greater than the human result, the margin isn’t great enough to offer a true advantage.

However, with China claiming it wants to be the leader in AI by 2020, and Russia’s President Vladimir Putin saying whoever wins the AI race will “become the ruler of the world,” who knows how long it will be before machines surpass us meaningfully in areas other than comprehension.

*Probably not true.