Why Formula 1 should reject artificial intelligence

In 1999, when I was technical director at Benetton, I started a project to apply linear neural networks - the keys to teaching computers to classify information in the same way as a human brain - to investigate the relationship between car set-up and how well the driver rated the handling of the car.

In those days, we wrote our own data analysis software and indeed built our own data acquisition systems as there were no proprietary products that approached our requirements. This made it reasonably easy to embed the electronic setup sheet into the header of the downloaded data.
 I also devised a very precise way of encouraging the drivers to be less subjective in their description using a flow chart that related their difficulty in a corner and their perceived loss of time to a single rating from one to five.

The final part of the infrastructure at that time was a separate Oracle SQL database that held all our data, setup sheets and driver comments. It could be interrogated in many ways. This was useful in that, if you were considering trying, say, a stiffer front roll bar, you could query the database for every time you had changed from a type 1 to a type 2 front anti-roll bar and it would come up with the set-up, data and driver comments.

In itself this was a useful tool, but it still relied on the engineer to make decisions. My dream was to automate this and hence I engaged Sheffield University to work with us on a very early form of machine learning. The aim was that once the system was trained it would look at the data - particularly the understeer curve - and the driver comments and suggest what might be the best set-up change to make.

In reality we were a bit too far ahead of our time and the software available and computing power we could bring to bear on it was nowhere near advanced enough for this first foray into exploiting automatic analytics on 'big' data. Ultimately the system was never used in anger.

Fast forward 20 years and machine learning and artificial intelligence are phrases heard in many different contexts. In Formula 1 we're using Amazon's machine learning tools to bring you the analytics that are starting to appear in real time on the television feed. This will expand in the future to provide far more insight into what is happening in the races.

Within the teams, such techniques, in various guises, are now common place. For example, race strategy software has developed from the extremely simple time-minimisation calculations, through more intense statistical techniques, which examined probabilities in a relatively dumb way, to complex machine-learning techniques employing a mixture of learning and game theory to suggest strategies likely to give a good outcome through changing scenarios. This has led to the number of simulations for this one aspect of race preparation increasing from around 150,000 runs 10 years ago to over 1.75 million today.

Equally, with car set-up, the use of optimisation techniques has advanced rapidly and now the teams arrive at a race track with configurations that approach perfection. Friday practice sessions allow the 'old fashioned' tuning to take into account local environmental factors, such as track temperature and wind direction, to get ever closer to the consummate configuration.

However, we must question whether in sport it is correct to abrogate such decisions to machines. True our sport is a mixture of human physical endeavour and human machine design, but if we approach singularity in this domain specific arena the jeopardy that forms the prime motivator of spectator sport disappears.

Both MIT and Oxford University are setting up departments to study the ethics of artificial intelligence thanks to the philanthropy of Stephen Schwarzman, but in this context we are not really considering ethics. Instead we are considering spectacle, probability, risk and all those other inputs that release endorphins and cause the state we know as excitement.

The neurology of watching sport has been studied in depth. 
An avid fan is immersed in his sport, he almost imagines he is participating in it and wants to be involved in the actions. The human brain therefore becomes a player in the game, but that brain has limitations. The neurons that transmit information through the cerebral cortex fire between 350 and 450 times a second. A good laptop can process over 150 million instructions per second, so when it comes to processing vast amounts of information and drawing conclusions the human comes a distant second to any form of well-conditioned artificial intelligence algorithm.

It is this excellence that is abhorrent to any fan of sport, and one which I feel will rob us of all those uncertainties that provide the excitement and spectacle. We would never dream of putting stability control on a race car even though it is mandatory on road cars and intervenes whenever it detects the human in control is not up to the job. It is for this very reason we should resist machine learning in the determination of excellence in sport.

Thirty years before our experiments at Benetton, Stanley Kubrick released the classic film 2001: A Space Odyssey. Fans of the film will know how HAL, the computer running the spacecraft, overruled the human in charge. How long before, as Lewis Hamilton turns the knobs on his steering wheel to give more corner entry differential lock to help his oversteer on turn-in at the expense of some front tyre wear, we hear the prophetic voice of a Mercedes F1 car saying: "I'm sorry Lewis, I'm afraid I can't do that. This mission is too important for me to allow you to jeopardize it."