Musings - All Computer Models are Wrong

This week my note is simply a musing - something I thought about on the bike and wanted to share..... One thing that I am certain of is that nothing is certain! Life is a collection of random interconnecting experiences that we try and make sense of through rationalisation and explanation. The truth is that much of what we think we know is wrong, and where we are today in our lives is the result of a series of opportunities and decisions that link together to form a life path. In his excellent book 'Fooled by Randomness' Nassim Nicholas Taleb explores the role of chance in life and the fallibility of human knowledge. Humans love to be able to explain and provide meaning to everything. This leads us to find causality where there is only coincidence, and assign precise explanations where there are only loose facts. For a minute think about how some of the decisions you have taken in your life have led to your current life path. Now imagine an alternative history where you make different decisions. You may have married a different person, not finished your degree (twice like me!), you may have turned down what ended up being a dream job, and so the list goes on. Our lives could be very different from the ones we are leading and have developed very precise and convincing narratives about. The development of computing power has provided us with the ability to model how certain situations might play out in the future. By looking at past data, we can find patterns that might help predict the future. Human nature is pretty constant so the theory that history repeats itself has considerable merit. Computer models are also pretty good at being able to provide probabilities to describe uncertainty. Monte Carlo simulations are computational algorithms that can mix and muddle data to produce different versions of history. They are widely used in financial markets to model risk but are they any good......? As with all computational models, including physiological models of athletic performance, they are useful but not sufficient. In the 2008 financial crash we saw market events that should have had only a 0.3% chance of happening and yet we saw these events for multiple days in a row. In the current situation with COVID 19 we have seen the usefulness, but also the limitations of using computer models to make decisions about lockdowns, social distancing and quarantines. Rewind a couple of weeks and we all witnessed how a well thought out and sophisticated (we hope) computer model totally messed up exam grade predictions. People love computer models because they can provide an explanation. They can also allow people to abstain responsibility. They can make decision making lazy. " I just followed the science" How does this in any way relate to cycling? Pretty much the whole cycling training community uses a computer model to prescribe, measure and analyse training. Training Peaks, WKO5, Trainer Road, Strava ......all use computer models to tell you:
  • How fit you are
  • How tired you are
  • How ready for racing you are
  • What training is best for you
Don't get me wrong, I love these models, but if there is one thing I will tell you for free, it is that there is no precision in these models. They are as prone to error as the risk management systems used in the city and the system used to allocate exam grades. They are useful but not sufficient. Let me give you an example. Let's say you build your fitness to a chronic training load (CTL) of 100. You feel progressively fitter as you build from 50 to 60 to 70 to 80 to 90, having organised your training well and taken enough time for recovery and adaption. You complete your big event and go on holiday, relax your diet and sip Pina Coladas on a beach for 2 weeks. You get back home and check your CTL and find it has fallen to 80. You go out and ride your bike and feel awful, nowhere near as good as you felt at 80CTL on the way up and this feeling persists for multiple rides. Yet the model tells you that not only are you as fit as you were on the way up, you are now super fresh for a great performance. It should be clear to you all that what you do in your every day life effects your performance. Despite peoples' best and futile efforts to record everything they do to make sure they have a 'truthful' CTL (and I have seen pretty much everything on a Training Peaks diary including things that are best left in the bedroom, I kid you not!) there is no way that the system can accurately take account of every stress and strain in your life. This is why the subjective and qualitative feedback that you provide to your coach is the most important element of what we as coaches see. Think about HOW DO YOU FEEL? We provide all of our athletes with unlimited coach interaction and a very immersive and interactive fortnightly Zoom call that encourages discussion. I can often tell how fatigued one of my athletes is just by looking at them on the Zoom call, looking at their eyes and body language. The moral of my little story is - hey talk to your coach and tell them HOW YOU FEEL every day. That is the only way that we are going to develop a full understanding of you as a unique individual. This allows us to prescribe training that is both individual to you and also has the best chance of you adapting and improving given your current physical and mental condition. Over time we develop a sixth sense and can spot the early sign of form and fatigue through what you say and how you say it. Merely looking at graphs and charts tells us very little about the experience of the workout or ride or how you feel more generally. Have a great weekend riding everyone and remember.......fill out your post workout feedback! Rob Wakefield / Founder & Coach / 07779136840 Propello