Doomed to Fail the Turing Test?
Last week, David Boyle, a British author and journalist, wrote an Op-Ed for the daily newspaper The Guardian recounting his experience as a jury member for the Loebner Prize, an annual Turing test competition.
Alan Turing was the brilliant English codebreaker of the infamous Nazi Enigma encryption method and inventor of modern computing and artificial intelligence (AI). He famously argued that AI will have reached maturity the day a human being couldn’t tell, in a conversation, whether the other party was a computer or another human. This has come to be known as the Turing test.
David Boyle put forth an interesting and somewhat provocative takeaway from his stint as a jury member: artificial intelligence (AI) programs fail the Turing test because they’re too perfect. As Boyle writes, “It isn’t the infallibility of people that makes them human, after all. Quite the reverse; it is their sheer fallibility – their ability to make mistakes, be quirky, make relationships, love and care. It is more unnerving, not more reassuring, to be phoned by a robot that is nearly human than it is to be phoned by an obvious machine.”
That had us talking at Yseop. First of all, because as an AI company, the Turing test remains the ultimate goal. And secondly, because as a natural language generation (NLG) company, we’re acutely aware that NLG only succeeds in the eyes of our clients and the general public when it writes flawlessly.
Are we therefore doomed to fail the Turing test so that our solution continues to be commercially successful? There’s a lot of value in perfection. Public opinion tends to think that robotization, whether it’s in the agricultural or industrial sectors, or today in the services sector, is just about cutting costs, but it’s about more than that. Automation using robots – and AI is the latest wave of that process – is about automating boring, repetitive, and menial tasks to let human beings focus on what only we do best: personal relations, creativity, and so on. It’s also about carrying out those tasks perfectly. In the automotive industry for example, robotization has not only made car factories a safer place for workers, it has also made for safer cars. A robot in a factory is capable of repeating the same tasks countless times a day, perfectly every single time. That means higher quality products, like safer and more efficient cars.
The same can be said for AI and NLG. NLG technology allows companies to automate the data to insight workflow with software that explains insights in plain written language. Few people would find it interesting to spend their entire work day pouring over raw data, graphs, pivot tables, etc., and writing reports for their boss on what’s going on, why things are happening, and what’s the best course of action. People will get bored. People will get burned out. And mistakes will be (and are being) made – whether in the writing or in the reasoning – and mistakes can be costly. That problem is compounded by the shortage of data scientists on the market. The value of automating that process – and automating it perfectly – with just the click of a mouse thanks to a powerful AI natural language generator can’t be understated. An expert AI system, following a rules-based (deterministic) approach guarantees that your best practices – and your best practices alone – are used. There is no guessing involved. And the NLG component writes out the conclusions reached in flawless English, making the insights accessible to everyone.
Does that mean we will never pass the Turing test?
Of course not, but the focus for AI tools like ours today is perfection. Perfection is what businesses need and it is a lot easier to take something that writes perfectly and force it to make mistakes then it is to take something that is inherently flawed and make it perfect.