What French Bread Teaches Us About Artificial Intelligence (AI) And Machine Learning (ML)
I’ve never been a cheese fan, but in the days before COVID-19 I had the opportunity to visit France, home to some of the best cheeses in the world. So I did my best to adapt and tried all kinds of cheeses while I was there.
Throughout my cheese journey, I came upon the fact that French bakeries carry two kinds of baguettes that look identical but are fundamentally different in composition because of the type of yeast used in each. One is called pain au levain and the other is called pain à la levure. I was taught this easy tourist-y mnemonic for remembering the difference between the two: the former ends in “n” and that means “natural yeast.” The latter means “chemical yeast” which is a more common approach today.
I mean, they’re both just loaves of bread. What’s all the hoopla and how do you tell the difference?
As I mentioned, they look more or less identical. But when you taste the bread, the difference is obvious. As a dumb American, my easiest reference for pain au levain is the sourdough breads I’ve enjoyed during my trips to San Francisco. That slightly sour taste comes from the natural yeast.
I bring up this tale of two breads because at the time of my personal CheeseFest, I was grappling with the fact that folks who can speak machine, like me, have always been unusually good at discerning whether or not an interaction they were having was powered by a computer. For example, the pauses in diction from a natural-sounding voice on the other end of a phone call, an obvious “tell” that I’m talking to a live human. Or, when watching a Hollywood blockbuster, it was almost always clear to me which scenes involved a computer graphic character or a synthesized backdrop because there was always some flaw that revealed it was created digitally. There was a certain distinct “smell” to this kind of computer interaction that was always easy to detect. It wasn’t a cognitive smell that might indicate that something is natural, like a pain au levain. It was the exact opposite: it reeked of being unnatural. They might look the same, but you could always taste the difference when looking for it.
The smell of the machine is disappearing.
That same year I was enjoying my cheese, Google had just released its new voice assistant that sounded eerily like a real human being with just the right kind of pauses and “ums” and “uhs.” And I began to wonder when the discernible smell of computer-based interactions was going to eventually disappear.
How was this possible? The advent of machine learning, or so-called “ML,” had achieved a ripeness in maturity that was boosted in the early 2010s with the dawn of inexpensive and constantly improving Graphics Processing Units (GPUs). The way that GPUs process information turned out to be the perfect architecture to accelerate machine learning methods and move ML from science fiction to the consumer market. If you just listen to the way that Apple’s Siri voice assistant evolved from iOS 9 to iOS 11 on their research site, you can literally hear that ML quantum leap in the transition from the iOS 10 and iOS 11 voices.
So while in France, my new recognition of the two kinds of breads that looked the same but tasted different felt like one of those things you get to enjoy as a human taking careful note of the world. In a similar vein, I felt like modern ML advances had started to strip away the ability for us to detect the difference between an AI versus a human being. We would become a world with one bread to rule the world that smelled and tasted human, but really wasn’t. A couple years later, The Guardian shared an article that was written completely by a computer and I certainly couldn’t tell the difference.
I guess that’s why I began to write about how to speak machine with the hopes that there would be more folks out there who can more aptly, and figuratively, smell what’s going on right now during this fourth industrial revolution. Over time I figure I will stop caring, but while this shift is happening I want to remember how there was a difference. The smell of the machine is disappearing.
