Putting science into predicting the next big thing

Originally, self-driving cars were programmed to handle a vast range of potential scenarios. If the car in front suddenly brakes, then stop. If an object blocks the road, swerve or stop to avoid it.

If this happens, then do that. The original innovators built a multitude of “if-then-else” type decision algorithms to ensure a smooth and safe ride.

Then they realised that the scenarios were ad infinitum. If they continued down that path, it would be inconceivable to ultimately place a self-driving car on populated streets. It was simply impossible to programme the infinite number of if-then scenarios that a car may encounter in an uncontrolled environment.

Prediction problem

However, when the innovators reframed driving as a prediction problem, it all of a sudden made sense. Rather than programming endless ifthen statements, they started to allow artificial intelligence (AI) to make predictions based on a simple question: “What would a human driver do?”. (I certainly hope they asked, “What would a “good” driver do?”.)

A few months ago, when I was in Tokyo, I went to a lab to watch driverless cars in learning mode and see how the prediction technology worked. Fitted out with a variety of sensors and cameras, the cars were put on the roads to observe and record what drivers did.

Low kerb appeal

They looked like they had been decked out by a group of teenage nerds working away in their basement, but the absence of kerb appeal belied the rigorous sophistication of the innovation.

The cameras covered every possible angle, meaning they could see substantially more in their periphery than you or I ever could.

As these cars took to the road, they collected vast quantities of human driving data. AI allowed the cars to recall and think through what the drivers were doing, thus enabling them to predict how humans would react and to do the right thing as a result.

Goodbye endless if-then statements. Say hello to the era of prediction.

Do you want to know how to spend your time better? How to get more return from your marketing spend? How to maximise your operations?

Let me predict that the answer is “Yes!” With this shift to the prediction era, you can.

Declaration in advance

AI, in its essence, is a prediction technology designed to declare in advance of certainty, what will happen. Since it’s impossible to have guaranteed accurate information about the future, we need to make such predictions. For example, how much inventory you should have on hand is a prediction.

Prediction is the input to a host of activities. What medicine should I take? What’s the best way to get to my destination? What should I wear?

Who’s likely to buy my product or service? Who’s likely to travel on my airline or stay at my hotel? What actions are my employees likely to take next? What changes should I make? If I make a change, what will be the outcome? At the core of each of these questions is a prediction. Prediction is useful to assist you in making plans about uncertain future developments. Are you in the trap of, “if this scenario happens, then we will [fill in the blank]?”

If so, instead of waiting for the future to be clear before responding to something, predict and act.

Tasks reframed

Moving forward, lots of tasks will be reframed as prediction problems and you are going to start using prediction as an input for things you never did. This is not only going to happen via AI, it will become mainstream management practice. Prediction is the bridge away from opinions and educated guesses, and the path towards increased certainty. AI is able to analyse vast data at speeds and costs that were once unheard of.

It is accelerating the rate of prediction accuracy, plummeting the cost and making it a conversation for the boardroom. It’s also a giant leap away from the existing (and limited) optimisation and utilisation models created by human hands.

That’s probably a good thing. Philip Tetlock has spent decades studying how people make predictions and, as he puts it, “Experts are only about as effective at predicting the future as dart-throwing chimpanzees.”

People are often spectacularly bad at forecasting the future. But they don’t have to be.