Google is within reach of being able to create machines that have enough human-like common sense that they could chat, hang out and even flirt with us. That’s according to artificial intelligence brainiac Geoffrey Hinton, who was acqui-hired by Google two years ago.
“It’s not that far-fetched,” Hinton told the Guardian last week. “I don’t see why it shouldn’t be like a friend. I don’t see why you shouldn’t grow quite attached to them.”
This comes nearly a year after Ray Kurzweil, Google’s director of engineering and another noted artificial intelligence optimist, told us he expects computers to have enough emotional intelligence to have a romantic relationship with a human by about 2029. At the time I was highly skeptical of the notion of being able to convert emotions, which we can barely express in words, into code that a computer can process.
The problem, as I see it, is that such technology would require a dataset (i.e. defined meanings for something as elusive as human emotions) that we have yet to satisfactorily define for ourselves. Two plus two always makes four, but every dictionary, poet and human being probably defines “love” with a different set of words and experiences.
What Hinton is working on at Google, however, is a very interesting intermediary step to creating something somewhere in between Apple’s Siri and science fiction’s Samantha from “Her” or Ava from “Ex Machina.” He’s working on translating thoughts into code, something he calls “thought vectors.”
Initially, I had the same reaction to this concept that I had to Kurzweil’s notion of encoding emotions into binary — the dataset to do such a thing is not available because language, our current tool for expressing our brain’s electrical impulses in a meaningful way, is too fluid and amorphous to be shoved into the rigid digital logic boxes that computers use to process data.
I think of language as being like a river, a majestic thing whose source is usually hidden, either in high mountain snows, or deep underground springs, or drifting rain clouds. From that source, it flows wildly wherever it will — it typically follows a defined path, but it is capable of shrinking and swelling and carving a whole new trajectory at any time.
It is possible to capture a small portion of a river, to stick it in a bottle and give it a static, defined form — much as an algorithm might do with bits and pieces of language to feed it into an artificial intelligence for further processing — but in the process it is deprived of its beauty and power. You could bottle an entire river and stick those millions of bottles in a warehouse, but you could not call the contents of that warehouse a river. Even if you pulled off such a thing, the source of the captured river would continue to flow, slowly regenerating it, making it impossible to ever capture the entirety of the thing.
Then again — if I’m being totally real here — I have to admit that I’m biased, because what Kurzweil and Hinton are working on is a threat to me and my livelihood. Perhaps the truth is that grandiose river metaphors are the most potent weapon I have against the notion that an artificial intelligence may be able to write this article before I’m old enough to start collecting social security.
So it terrifies me to concede that Hinton may actually be on to something. He’s working on the problem that I’m attempting to explain with rivers, which is how to get beyond words and grammar and language to actual meaning. In fact, Hinton’s system uses a feedback loop to identify errors in word usage and continually refine word choices until the system is using them correctly the way humans use them.
In other words, if it works, such a system could continuously be capturing the whole “river.”
The nuance of human communication isn’t lost on Hinton either. He thinks that higher-level, more subtle forms of communication like irony and flirting should also be possible for computers to grasp based on the same principle.
“Irony is going to be hard to get,” he told the Guardian. “You have to be master of the literal first. But then, Americans don’t get irony either. Computers are going to reach the level of Americans before Brits.”
Speaking of irony, I could see how the first computers that master human language and communication might actually help us meatbags communicate better among ourselves and with our computers. Just imagine a resource that could actually tell you all the different things the phrase “meatbag” might refer to and the likelihood that you’re interpreting its meaning correctly.
In the long run, such a system might actually enhance human language and communication and lead to a new Renaissance in poetry and literature.
Or, it might just put me out of a job, in which case I’m on board with Elon Musk, Stephen Hawking, Bill Gates and the other smart people preaching caution in our approach to artificial intelligence. Just let me know if you see the robots either starting to bottle up the Rio Grande or sitting on its banks composing a metaphor.