By Richard Anderson, Founding Partner, Nuon AI
The term AI (Artificial Intelligence) is used rather loosely today to describe all kinds of things.
Leading some to get annoyed over its use, or apparent misuse. A week rarely goes by without a “the robots are coming and they’re going to take your job!!” story in the media – most of which will refer to AI; and more and more products are sold claiming AI features of one kind or another. Why so now, and why get annoyed?
Part of the answer is simply fashion. Remember when everything was Mainframe, then decentralised and finally Cloud. Or outsourced, then offshored became fashionable. Remember Web 2.0 and IoT? Let’s not even go near Blockchain. If you’re my age you’ll remember when everything was “Turbo” whether it featured forced air induction or not.
Hollywood has worked hard for decades to fix the idea of AIs into the public imagination, for the most part sinister AIs at that. It’s also about computing power. The concepts underpinning most things described as AI today are decades old, but we now have the power which allows them to do useful things.
It doesn’t help that “Artificial Intelligence” was a vague term right from the get go. What exactly does it mean to be Artificial, or Intelligent? Both seem obvious, but when you scratch the surface they just aren’t.
Let’s take Artificial first. Of the two, that seems easiest. There are natural things like bees, and natural processes like bees making honeycomb. Then there are artificial things like machines which extract oil from the ground, and artificial processes like making plastics from that oil. Seems clear. Work it through though, and it’s not clear cut.
A bee building a honeycomb is natural, a bird building a nest is natural. A human building a shelter? Starts to feel uneasy, since you see now where the argument is going. How about humans building a tower of glass and steel just over half a mile high… on sand? Not natural. Those humans using plastics in the construction of their building? No, no, no! Stone maybe, that’s natural, but plastic!? Totally unnatural!
However, a bee building a honeycomb of interlocking hexagons is just a bee doing that which is in its nature to do. Humans have it in their nature to create. Is the Burj Khalifa so different then, at least in principle? Maybe you agree, maybe not. But you would have to at least accept that Artificial’s meaning is fuzzy.
Then we come onto Intelligence, and it gets really fuzzy. Humans consider themselves to be intelligent. It was seen as a defining characteristic until other primates, dolphins, and corvids were studied. Now, the closer we look at the animal world, the more intelligence we find. Measures of intelligence, like IQ tests, are considered to be generally poor. In part because they can be studied and better results achieved. Suggesting that whatever they do measure, it isn’t an underlying ability.
In Life 3.0, Max Tegark describes intelligence as the “ability to accomplish complex tasks”. Let’s stick with that. But even a basic computer can accomplish complex tasks, is it intelligent?
At Nuon we’re as guilty as anyone of using the term “AI” loosely. We’ll say “our AI” when we’re referring to an algorithm, or “your/my AI” when we’re talking about a simulation or a client’s instance or, more accurately, a model. But we know what we mean, so that’s probably okay.
We’re talking about Reinforcement Learning, which is a subcategory of Machine Learning, which is itself a subcategory of Artificial Intelligence.
What we’re not talking about is General Intelligence. That’s the holy-grail subcategory of AI which delivered HAL in 2001, or Samantha in Her. Our Reinforcement Learning system learns continuously in real time and adjusts its behaviour, but it has only very narrow capabilities when compared to HAL. It knows only how to adjust prices in a market to capitalise on cohort-specific price elasticity. It can’t read lips. (yet)