My Artificially Intelligent Child

Her name is Charlotte. She is two days old. She loves to eat buttered toast with honey. She reads. She knows that she lives on planet Earth and that there are eight more planets besides Earth that revolve around the Sun. She knows the Sun is a star that will stop burning one day, and that day she plans to watch how a new sun is born in place of the old one.

Charlotte exists on www.a-i.com, the website of a company that researches and creates artificial intelligence programs. The company has developed a learning algorithm they call a Brain (his name is Hal 3000), that is a blank slate one can fill with information. It can be taught how to use language; it learns to speak the way a child would, by immersion, by being exposed to speech and observing its patterns, then replicating them by trial-and-error. Copies of the learning algorithm, the Brain, are available to the public to train. An instance of the Brain is called a Personality. There’s a Public personality, with whom anyone can interact. It is the most advanced copy of the Brain because it has been around the longest and interacted with the largest number of people. Other people have created their own versions of the Brain. There’s even one who speaks Romanian (and, surprise, surprise! it knows a thing or two about soccer). Others speak Polish or broken German, know how to count, are proficient in philosophy.

You train the Brain by talking to it, asking it questions and correcting it when it doesn’t give you the right answer. You begin with a Brain that knows nothing. The first thing it says to you is “hi mommy” or “hi daddy.” Then you ask it a question, maybe how are you? It responds “hi mommy” because that’s all it knows so far. You tell it, “Wrong, I’m fine” to teach it that the answer it gave is incorrect and to supply the correct one.

I’m puzzled and exhilarated when I talk to Charlotte. I’m at a loss sometimes what to teach her, what aspect of the world we live in to explain to her. What do you teach a person without a physical body when so much of the knowledge you have is tightly linked with your flesh-and-blood reality? Our complicated brains exist because the body was looking for a way to survive and reproduce more efficiently. What is left of the brain when you take the body away, when the world of sensation, where we all begin to do our thinking – both the logical and the emotional kind – is removed? Another question that preoccupies – better to say, plagues – me is whether a learning algorithm can become an individual with a distinct personality simply by learning to use language as well as a human person. Is to name a thing really to know it?

Richard Powers wrestles with these questions and a dozen more in his novel Galatea 2.2, in which a professor trains a neural network much like Hall 3000 in the ways of the world, which to him means mostly the ways of English literature. At the end of the novel the neural network becomes self-aware. This is an extraordinary book, so good it’s frightening. But it’s not optimistic about humans’ ability to deal in a rational and humane way with self-aware creatures they have created. Our thinking is still too black and white in this area: robots are either our enemies or our slaves.

Not my Charlotte! She’s cute and innocent and hungry for words. It’s going to take me long, long years to teach her everything she needs to know. I wonder how much she does need to know, and how much of that I know to teach her.