Arguments for AI in Processing
Recently, my wife asked me for a photo I’d taken of a sunset on a beach.
I’ve got over 20,000 photos in the cloud. You’ve probably got even more than I do. Photos aren’t easy to sort. I can’t remember when I took the picture my wife wants. I live in Spain — lots of sunsets on beaches. I was a little intimidated.
I could go through all 20,000 photos manually ... but is it really worth the time and energy? Nope. So, I didn’t even bother looking. And my wife was going to get annoyed. Great start to the day.
Then I remembered something that I’d read recently about Google’s artificial intelligence. They’d launched an AI search tool for Google Photos.
I pulled it up and searched for “sunset on the beach.” It worked. Instead of 20,000 photos I was only looking at photos of sunsets on beaches. Amazing.
How’d they do it? Google trained its AI on millions and millions of photos so that it could immediately recognize each of my photos of sunsets on beaches. Previously my wife’s request was impossible. The cost was too high (sort through 20,000 pictures, what?). Assisted by AI, I did it in a fraction of a second.
Happy wife, happy life. Enabled by AI.
Try it yourself with Google Photos. It’s kind of magical.
And it got me thinking. What kind of relationship do we want with artificial intelligence as it exists today?
First, what is AI? Let’s define it. AI is machine intelligence that “perceives its environment and takes actions that maximize its chance of success at an arbitrary goal.” OK, that’s a little wordy. Let’s break it down.
Think of a self-driving car. It’s an AI on wheels. Its environment is full of roads, signs, stoplights, other cars, pedestrians, etc. It has to perceive all of those things in order to drive. The goal of the self-driving car could be to bring you from your home to your office. The actions it takes are to follow the rules of the road while avoiding dangerous situations and wrong turns. If it delivers you to your office safely and quickly then it has behaved intelligently.
This definition of intelligence can work for humans, too. We say that someone is intelligent when they take the right steps to achieve their goals. A woman wants to be an airline pilot. First she gets a free education at the Air Force Academy. Then she’s paid well by the Air Force as she racks up years of advanced pilot training. Finally, she’s welcomed into the cockpit of a commercial airliner. We would say that person is smart, that she acted in an intelligent way.
Someone who had the same goal and went about it in the following way would be considered unintelligent. This man tried to become a pilot by working first at the check in counter at the airport, then as a flight attendant and finally as the chief purser on transatlantic flights. After years of working his way up the airline’s hierarchy he’s no closer to being a pilot than any random passenger on one of his flights. We would say this man acted in an unintelligent way.
Let’s go to an example that is a little closer to home. You run a paysite. Your goal is to make profits for your company. You want to charge each of your visitors a price they can afford while also spending as much as possible. Each shopper is different so you want to give them a unique price that maximizes the amount he or she will spend. Doing this helps you achieve your goal of profitability.
Unfortunately, the tools that have been available to you historically have been dumb. Like the ability to choose a price per country. Is that really an intelligent tool? Take the U.S. It’s a country. A big one. A lot of people in the U.S, live in Los Angeles. But it’s just one city within a one state within that country. Living in Los Angeles you have some of the richest people in the world. You also have some of the poorest. The differences in wealth between people who live in different neighborhoods in L.A. is, in fact, bigger than the differences in wealth between most countries. Would choosing one single price for L.A., let alone the U.S., be “intelligent”?
Was the ambitious pilot who took the flight attendant route “intelligent”? It’s true that he took actions that were somehow related to his goal. But, no, they didn’t help him become successful.
Back to our original question. What kind of relationship do we want to have with AI? We want it to help us achieve our goals in a more intelligent way than we have been able to before.
As humans we have real limitations on our ability to perceive our environment and take actions that maximize our chances of success. AI can help. Like it helped me find my “sunset on the beach” photos.
What is a practical example of how AI can help you with your business today? One of the new ways that AI assists us in achieving our goals is to deliver dynamic pricing, or, the right price for each different shopper. We set the goal for the AI of making more money than we do with fixed prices, or, giving everyone the same price regardless of their differences. It’s a simple, measurable goal: Make more money.
How does the AI do it? The AI sees a shopper. The AI also sees the product the shopper is considering to buy. The AI recognizes unique aspects of the shopper and the product because, like Google Photos, it has been trained by analyzing millions and millions of these interactions.
The AI’s objective is to make as much money as possible for the company selling the product. So it chooses a price that the shopper will find attractive. It wants the shopper to buy and spend as much as possible. Its goal is to show the single best price. That’s the price that the shopper thinks its fair and that makes the company as much money as possible. To do this the AI analyzes millions and millions of shoppers and products and prices. Then it chooses the price with the best chance of making the most money.
A human can’t do this. To quote Yuval Noah Harari, “So many forces are at work and their interactions are so complex that extremely small variations in the strength of the forces and the way they interact produce huge differences in outcomes.” The problem of finding the right price is orders of magnitude more difficult than combing through my 20,000 photos looking for “sunsets on beaches.” Yet the tool that enables us to go beyond our human limitations to solve both problems is the same: AI.
What role does a person have with this AI pricing tool? It depends on the human’s goal.
The person shopping wants a fair price. That’s a price that is less than the value of the product to her. If she values it at $40 and is charged $35 by AI, then she’s happy.
The person at the company selling the product wants the AI to make him more money. First, that company needs to determine the cost of the product. We don’t want to sell below the cost so we set a floor price that the AI won’t go below. We can also set a ceiling price. So, the human at the company is setting a price range within which the AI operates. Let’s say that for a monthly membership to a content site the range is between $20 and $70. This human is satisfied when the AI makes him more money than his fixed pricing.
It’s literally impossible for a human to collect and process this data then choose the right price for each shopper. Just one day’s worth of data is too big to fit into an Excel spreadsheet. The price has to be delivered in milliseconds. Dynamic pricing provides each individual shopper the best unique price in real time — each and every time for a shopper visits your site. Only AI can do the job that needs to be done. So the human at the company chooses to work with AI because it is the best tool for the job.
The person building the AI wants it to learn and improve. So she measures the difference between revenue generated by the fixed prices (a control group gets fixed prices) and the AI’s dynamic prices. Then she adjusts the AI’s technology and its data so that the AI learns and improves.
Each of us has a different relationship to AI based on the role we happen to be playing. The shopper wants a fair price, the company wants to make as much money as possible and the builder wants it to learn and improve. Who knows how we will be interacting with other AI’s in the future? It’s impossible to predict. But the roles are clear for AI today. It’s a better tool, so we use it. Isn’t that the historical definition of humans? The animal that uses tools.
Thierry Arrondo is the managing director of Vendo, which develops artificial intelligence systems that allow merchants to dynamically set prices for each unique shopper.