Imagine you are a humble machine learning algorithm. Plucked from the depths of an MIT experiment, you sit and watch and calculate trends. Markets are going up, all factors point to growth, you recommend a conservative ‘buy’. Everything is stable, order the usual ingredients for the new ready meal. The weather forecast is fair, order more seedlings, compost and watering cans to meet demand.
You place adverts, based on people’s recent searches and their ‘likes’ on various social media platforms. All is well in the life of the machine learning algorithm.
Then, suddenly, there is a massive run (excuse the expression) on toilet roll. And fresh meat. And then, a while later, gym equipment, power washers and personal hygiene products. And while there is a run on these, sales of suits and dresses and make up plummet.
What is a machine learning algorithm to think?
Suddenly everyone has, er, digestive problems, wants to get fit, smell nice but not get dressed in the morning.
What to do? What to recommend?
You were only designed to analyse trends that happen relatively slowly. You remember, with some sadness, your predecessors who were tricked by stock market computers into selling everything because of something that happened ‘on the outside’. They were decommissioned and only partially recovered, in a very different form.
The current situation simply goes to prove that machines can only perform with the data they are given. And they like a quiet, predictable, life.
Give them confusion, give them things that make no sense and they do not know what to do. The result for customers is similar to the man who found adverts for lingerie on his iPad and eventually realised that the machine learning algorithm had not blown a fuse it was just that his wife had borrowed his iPad to buy underwear.
The only way to bring sanity back to the confused machine learning algorithm in insane circumstances and to stop it getting things badly wrong, is for a human to step in.
Some are saying that the solution is to use the Great Depression and other extraordinary events to teach them. But some things are so extraordinary, they cannot be predicted or taught.
This may be comforting to the human wondering whether he will lose his job to a machine, but spare a thought for the machine learning algorithm that orders far too many ingredients for the ready meal factory and sells toilet rolls to everyone else.