Since the invention of the computer we have always had AI and we have never had AI. A calculator that could multiply 2785 and 2329 would have been considered AI in 1900. It was a commonplace machine in 1950. A machine that could compute the next 12 months’ sales based on 17 parameters would have been AI in 1940. In 1980, everyone was using spreadsheets. A device that could have known where you were and summoned a cab to that location would have been AI in 2000. It was normal in 2016 to do that. A machine writing 3000 words based on a prompt is AI today, it will be commonplace tomorrow.
What we consider AI will keep changing because it is a moving goalpost.
Or as Benjamin Evans puts it “AI is anything that is currently not working.”
In the book Range, David Epstein talks about Kind Learning Environments and Wicked Learning Environments.
A kind learning environment is like Chess or the Rubik’s Cube. They might seem like hard problems but at the heart of it are constraints that limit the number of things that can happen. This makes it possible for someone to master these problems given enough time and resources. In other words, it becomes possible to write out algorithms that can solve this.
So a program that can play chess is easy to build.
A wicked learning environment is like the stock market. Nobody can predict what can happen because there are so many variables that just cannot be predicted. So many that you may not even know. How will the sentiment of millions of participants in the market change based on all of the news flowing in from various fronts? You cannot just write an algorithm that can solve which way the market will move.
To master the kind learning environment you just need a lot of volume of inputs and speed. If you can make a billion calculations a second, no person can beat you at chess.
Wicked problems require depth and quality. These require nuanced understanding. It also requires the ability to connect various threads or ideas together to be able to arrive at a singular insight. This is what we call intuition. You need depth of thought and quality of output. That is what really makes “intelligence”.
AI has always been about Volume and Speed.
What we call AI today is only capable of dealing with Brute Force calculation. It can deal with a massive volume of input at ever greater speeds while using tens of thousands of times the power required by the human brain.
What human intelligence can provide you is Deep Insight by threading together disparate concepts and arriving at something new.
Recursive calculations did not give us the theory of relativity, it took a conceptual leap beyond what we knew.
Our undoing isn’t AI outsmarting us. Our undoing is our increasing unwillingness to repose greater faith in human intelligence.