Boker Tov: What Has Sabich Got to Do With AI?

The lesson

The only light in the shuk

It is ten to five in the morning. The shuk, our market, is still dark. My stand is the only one with the light on. The sign above my counter says Boker Tov. It is Hebrew. It means good morning. I pour you a glass of sweet tea, and I pull out the one stool at my counter. You are my first customer today. You came for my sabich, and you will get it. Fried eggplant, a hard-boiled egg, potato, hummus, tahini and amba, all tucked into a warm pita. But I am serving a second dish this morning. That dish is artificial intelligence. Here is the menu. First, where AI begins. Then, how machines learn and create. Then, machines that plan for themselves. Last, what it means when we give them hands. My promise is simple. By the last pita, the technology news about AI will make sense to you.

What AI is, and what it is not

Let me clear the movies out of your head first. AI is not a small person living inside a machine, thinking and plotting. It is a machine that learns from examples. A stand keeper learns the same way, from years of mornings. Rain coming at six tonight? I know the shuk will be quiet, so I fry fewer trays of eggplant. Friday before a holiday? I stuff 300 pitas instead of 200. The eggplant crates arrive before the lights are on, and I sort them. The firm ones go to the oil. The tired ones go to the bin. A regular tells me my amba was too sharp yesterday. I adjust, and today I cook a little better. That is the whole package. I predict, I classify, and I learn.

Some say the name sabich comes from the Arabic word for morning. A morning food, for a morning stand. Mornings taught me everything I know.

Now picture one straight line, and put every machine on it. At one end sit the rule followers. A calculator obeys its keys, and it never gets better at arithmetic. At the other end sit the learners. Every learner improves with experience. I live at the learning end. So does AI. Feed a machine enough examples, and it acts as if someone had written it a rule. The examples taught the learners how to predict, classify or group things.

You already live at the learning end

You already live at the learning end, whether you notice. Your phone unlocks at a glance. It learnt to read your face from millions of photos. Your map warns you of a jam ahead. It learnt from years of morning traffic. Your keyboard suggests your next word. It learnt from you. So when a company says its product uses AI, it nearly always means one thing. Many examples taught the product’s AI, and the AI adjusted itself until it became good.

Real AI, or movie AI?

Now let me test you. I will make three claims. After each one, ask yourself: real AI, or movie AI?

Claim one. Your phone unlocks the instant it sees your face. Real, or movie? I give you three seconds.

Pause here for three seconds.

Claim two. An app writes a birthday poem for your savta, your grandmother, in 10 seconds. Real, or movie? Three seconds.

Pause here for three seconds.

Claim three. A robot feels hurt when you walk past its stall. Real, or movie? Three seconds.

Pause here for three seconds.

Two of the three are learners with enough examples behind them. The third is a story about feelings. No matter how many examples you throw at it, the AI will never understand feelings.