“Science gathers knowledge faster than society gathers wisdom.” Isaac Asimov
I recently attended a conference about the use of artificial intelligence (AI) in business applications. Speakers represented legacy firms like Walmart, where the use of AI might be surprising, newer companies like Google and Pinterest, where AI is expected, and many others, large and small.
It was clear from much of the conversation there that AI is virtually equated with machine learning (ML). And personalization is taken for granted as the primary application of AI. While the scientists and engineers had a deeper understanding, virtually every product and business person narrowly focused on ML and personalization.
This is a great disservice to the field.
Let’s start with personalization: Amazon and Netflix led the charge in personalized recommendations starting in the early 2000’s and 2010's. These have been a huge driver for growth and profit for both companies. The very point of personalization is to drive profits by increasing user’s engagement with your product. Is this the fulfillment of AI’s vision?
Let us back up a minute: what is an artificially intelligent system? It can read as a person would read, except a lot more and a lot faster. It can draw connections that a person would make, except more of them over more complex relationships. It has instant access to what you might think of as the giant idealized conceptual graph of human knowledge. The purpose of making a system intelligent is so that it can behave as you would expect and want it to, if it understood the data, content, and context of your interaction with it. In other words, intelligent systems are the very pinnacle of good human-computer interaction.
So, yes, I really would like Amazon to find me books that I would like to read! On the other hand, I’m not sure if I would like Amazon to find me books that will maximize Amazon’s profits over time. Perhaps Amazon will keep me coming back by steering me to a series of books that I might not really enjoy all that much, rather than showing me an old masterpiece where it doesn’t make much margin. Same thing for Facebook — of course I want it to prioritize interesting news. But it might make more sense from Facebook’s point of view to keep me slightly anxious, so I’ll come back and check my feed multiple times a day, and perhaps click on some paid advertising. (And if you think Facebook would not purposely tamper with your emotions for its own profits, think again.)
Now let’s go back to machine learning: intelligent systems that improve their performance with experience are said to be learning. Just like we do! So that’s got to be good right? Well, again, it depends on whether they are doing it with our best interests in mind. Maximizing personalization of your shopping experience can be good or bad, depending on how and, even more importantly, on why it’s done.
Imagine for a moment an intelligent system which is not trying to learn how to show you things you are most likely to buy. There are many other things it could be doing. Off the top of my head it could teach you languages, tutor your kids in math, help you avoid car accidents, help you decide whether to take the muni or BART to get to work on time, wake you up a bit early if there’s a traffic slowdown on your way to work, monitor dangerous fire conditions, help shape policy to slow global warming, or help with important healthcare decisions.
So let’s drop the focus on personalization and machine learning: the future is in creating systems that help all of us live a better life as good citizens of this world.
This is a follow up to my previous article on AI.