Don’t you just hate it when you’re looking for support for a service or app you bought, only to be greeted by some monosyllabic robot ? Ok, that can happen just as well when dealing with outsourced tech support, but at least you know you’re talking to a real person. Well, that might change sooner than you might think. The singularity is getting closer by the moment. Just take a look at Google’s new chatbot which according to the developers has moderate “natural language understanding”. In other words, it can roll with the punches and continue the conversation by itself without following predefined question – answer. Of course, after a while you can still tell it’s not human (fails Turing test), but that doesn’t mean it isn’t entertaining. Have a look at how it answers to “what’s the purpose of life?”.
The bot was developed by Oriol Vinyals and Quoc Le, both researchers at the Google Brain project. They claim their AI works by learning from previous sentences and predicting what comes up next. Essentially, the bot is trying to second guess where you’re going with your questions and frames the answers accordingly. However, after each question it re-arranges itself.
“We experiment with the conversation modeling task by casting it to a task of predicting the next sequence given the previous sequence or sequences using recurrent [neural] networks,” they wrote in a paper describing the experiment posted on the arXiv pre-print server. “We find that this approach can do surprisingly well on generating fluent and accurate replies to conversation.”
To learn “human speak”, the bot was fed with thousands of movie subtitles and a dataset from an online tech support chat. For instance, here’s how it behaves as a tech support assistant.
Not bad at all, quite lively. If you stick to simple, to the point questions – like when you’re trying to solve your tech problem – the bot is up to it without too much hassle. It might even fool some. When it gets too philosophical, it sounds like a drunken droid though.
Nevertheless, it’s still pretty freaking amazing! It doesn’t make those blatant noun-pronoun errors you typically see in other bots, and actually has a bit of “personality” (borrowed from the datasets – does the bot think he’s a movie star?). Just kidding. After all, it can’t think. This is still a mindless software, a child’s play, but most certainly a taste of what’s to come in the future as we see bots made to interact with information or even the physical environment not by acting on predefined instructions, but by improvising based on previous experience.
“In other words, it does not simply look up an answer by matching the question with the existing database,” they wrote. “In fact, most of the questions above do not appear in the training set.”
“It is surprising to us that a purely data driven approach without any rules can produce rather proper answers to many types of questions,” the researchers add.
There’s still a long way to go until we see the singularity breached, but until then these sort of artificial intelligence renderings might prove extremely useful.
“The model may require substantial modifications to be able to deliver realistic conversations,” the researchers wrote. “Amongst the many limitations, the lack of a coherent personality makes it difficult for our system to pass the Turing test.”