AI started out as a term to cover Intelligence in the handling of problems. After multiple failed waves of AI technology, it has increasingly become data-driven, with no place for intelligence.

Artificial Neural Networks (ANN) is a good example. It is thought by some that a device can be programmed using data and sent out into the world to handle the problems it encounters. There are three problems here – the world is a complex place, so some learning on the job would be beneficial, the data is out of date as soon as it is gathered unless everything is stable, and someone has to decide on the appropriate inputs and outputs. Deep Learning, a version of Machine Learning, uses a simple algorithm to decide which inputs and outputs to keep, by throwing away any inputs that do not change the outputs. This has its own problems. Some inputs, like the presence of fire escapes, only rarely affect the outputs, but results can be catastrophic if they are not present.

Some purveyors of AI use the expression “The answer is in the data”. This is fine for simple problems, like how many hammers a hardware store should stock. The companies that sold button up boots saw a steadily rising sales graph, until disruption in the shape of a zipper came along, and destroyed their business model. Rather than concentrating on data, their time would have been better spent on looking for disruptors – specifically patents for fastening devices related to boots.

boots

The answer is in the data sounds very much like driving while only looking in the rear vision mirror – a practice which would quickly land someone in the ER or the cemetery, or your business down the drain in turbulent times.

rear view vision

Some problems are clearly unsuited to the data approach – Climate Change, where the world is changing, Anti-Money Laundering, where there is an intelligent and nimble adversary, Strategic Planning, responding to a rapidly changing world. Even some of the problems chosen to work on by people in the AI area –  Autonomous Vehicles – are far too open for a programmed approach to be successful. A narrow winding road, with a steep drop on one side and a wall of rock on the other and lots of blind corners, and no centre lane marking (Berry Mountain), does not look like something that can be navigated by a vehicle trained for urban roads, and without any ability to assess a new situation and respond to it. Or drive in fog, to add an extra twist. The recent encounter of a Tesla with a business jet in a carpark demonstrated that, after six years of development, the vehicle is still not aware of the size of the navigable aperture required for it to proceed.

AGI in the form of Active Structure has the ability to make new connections, to look up a word it doesn’t know in a dictionary, to surmise and test strategies for new conditions it has never encountered before. There are of course limits on how much intelligence it can display – it has millions of connections, you have billions (but it is a bit focused and it doesn’t sleep nights). An advantage it holds over humans is that it does not have a Four Pieces Limit, so while it may be slower than a human, it can manage much larger and more complex problems than a human can (the human “chunks” the problem – keeps large parts static so they are not overwhelmed, which destroys the nature of the problem and limits the usefulness of the solution).

The four piece limit