It’s all A and no I. All intelligence requires agency. None of the machines we’ve built have agency in themselves. Yet these are very useful tools, just as woodworking, machine tools, hydraulics, and electric motors greatly extend what our physical bodies can accomplish.
What we term “AI” can be grouped into algorithms and machine learning.
Algorithms are created by individuals who have agency. “IF this then that,” and “do this next” programming. The intelligence is in the individual, who can reasonably explain the why and the how of the algorithm. Millions of us receive customized email messages from Amazon which recommend products based on our previous searches. These are created by algorithms and software written by intelligent individuals. Glenn Brooke gets recommendations on an odd collection of old books, new business books, backpacking gear, laundry soap, and dog poop bags.
Machine learning is a relatively new kind of tool. Intelligent people create a training data set and a decision model to make predictions or decisions without being explicitly programmed to do so. During the training phase there is a feedback loop so that over time the model gets better at making predictions. There are some similarities to the way we think neurons in our brain work, but machine learning make progress in a different way than human thinking (which is partly why it’s a valuable tool). A machine learning system cannot explain how it arrives at a prediction, but the predictions are useful.
There are many different applications for machine learning. Most are quite specific to a business or research process. Famous examples are the machine learning models that play Go and Chess better than any human or earlier software program. Alpha-Go, for example, was trained by giving the model tens of thousands of games of Go, then played against itself. Alpha-Go plays quite differently than any human Go master. Alpha-Go can’t play backgammon or checkers or Monopoly — but the underlying machine learning technology could be pointed to new training data sets to create a world-class players in those games. Another powerful example of machine learning is DeepMind’s system to predict 350,000 protein structures – a real leap forward for medical treatments.
The ability to make better predictions or decisions from messy, complex data is a powerful asset. The key thing to recognize about machine learning is that no one can really explain in detail how the prediction was made, and the machine learning tool is not self-aware of why it has been optimized a certain way. Intelligence is still required to decide what to do with the prediction that machine learning made.
There is no universal “AI.” Algorithms and machine learning models are exquisitely built to accomplish tailored tasks. You can combine many of them into systems which appear to act intelligently, but all the intelligence is in the creators. The idea that a new intelligence will spontaneously develop when we connect a sufficiently large amount of computational power is simply false. Intelligence requires inference and leaps of connections which are impossible in today’s 0 and 1 binary coding systems. Everything that “appears” intelligent is a function of human designs. Scientists and philosophers have at best working definitions of alive, intelligence, and consciousness which are far short of being able to engineer these things.
Remember: We shape our tools, and our tools shape us. He who shapes the vocabulary shapes understanding.
