Coach Tip Tuesday: Is AI Coaching & Sports Training Software Actually Intelligent?

There have been so many changes in the endurance sports space over the last several years, and how athletes approach and think about training is one of the big things that has changed.  With the economic changes that have occurred in the last several years, athletes have been seeking less expensive ways to train for the things that matter to them and still see results.  One of the new and less expensive options available to athletes is AI-generated (Artificial Intelligence-generated) training.

‍

But do less expensive training options, such as AI-generated training plans, have the same value as “traditional” training plans designed by human coaches?  Or, does less expensive actually mean “cheaper,” aka of inferior quality?  Is AI (Artificial Intelligence) training software actually intelligent?

‍

While I am not opposed to technology (proof of this is that you are reading this conversation on a technological device!) and use it in my daily coaching life (my preferred coaching/training platform is Final Surge), I have a level of skepticism when it comes to newer technologies.  There are several things I’m always seeking to feel out from a new technology:

• Is it easy to use?
• Is it easy to understand?
• Is it actually helpful?

‍

It takes time to determine the answers to these questions.  As such, it’s not often that I will publicly adopt a new technology early.  While I may test it out, it actually takes a long time for me to endorse something as being worthwhile for athletes to use.

‍

AI-generated training plans have not yet proven themselves to me.  Yes, AI is new, glitzy, and glamorous.  AI may very well be here to stay.  But this doesn’t mean that its current iteration is one that has a positive impact on athletes or the goals that are so important to them.

‍

What Is an AI Coach?

Training plans developed by AI are generated using biometrics and performance data from thousands of athletes who have individually completed millions of workouts.  The platform’s software then uses data mining to process and sort through this massive data set to identify patterns and relationships in the data.  Machine learning is deployed to use the data and algorithms in a way that imitates how humans learn, gradually improving the accuracy of the software.

‍

Thus, the software “learns” from that massive data set and compares it to the data that is uploaded by an athlete using the platform.  The software makes a recommendation of training workouts for the athlete based on this comparison and any filters, preferences, and parameters that the athlete inputs themselves into the program.  As the athlete progresses through the training plan, the platform’s software will continue to assess the athlete’s data and compare it to its aggregate data set, making changes to the athlete’s training plan as it deems fit based on what the data indicates is best.

‍

Training platforms that leverage AI technology to generate training plans often sell themselves as “coaches”, but I honestly think that this is inaccurate.  AI training platforms are computer programs, and I don’t think that there is currently truly such a thing as an AI Coach (or, perhaps more accurately, a Computer Coach).  My definition of an endurance coach is more expansive than what AI training platforms are currently capable of doing; right now, AI training platforms can only process data and make data-based recommendations.  

‍

An endurance coach can also do this, and they can also account for all of the nuances and details of your entire life, not just the data sets that are uploaded from your workouts.  Additionally, you can speak to a human endurance coach (either in-person, phone, video, email, text, etc.).  This - the ability to communicate - is probably the most critical piece of my definition of an endurance sports coach.  It is via this communication that a human coach can actually provide coaching, not just workout planning.

‍

Rethinking AI-Generated Training: The Unseen Stress of Data Dependency and Algorithmic Limitations

While the term “AI-generated training” is what is circulating, what is honestly more accurate to say is “algorithmically-generated training.”  An algorithm is a set of rules to be followed in calculations or other problem-solving operations.  While some humans are capable of processing algorithms, the vast majority of algorithms are programmed into a computer.  Thus, the computer does the “thinking.”  In endurance sports, humans have developed algorithms to take data and to generate predictive training based on those data sets; this is what they are calling AI-generated training.  Simply put: AI-generated training cannot function without data inputs.

‍

In my eyes, this is a major limitation, and here’s why I find this to be problematic: It puts immense pressure on athletes to always record data, and to always record data exactly correctly.  If athletes do not do this, their workouts will not be planned appropriately for them.  Most software has a tolerance range for no or bad data of up to about 5%.  This means that in a typical training situation, an athlete can forgo recording data once a month if they are using AI software to generate their training.  For all intents and purposes, all workouts need to have properly recorded data.

‍

Think about it.  Every single workout, every single time, must be recorded.  Not only must every single workout be recorded, but it must be recorded correctly.  No mistakes can be made.  You cannot forget your heart rate monitor.  You can’t forget to zero-out your power meter before riding.  You cannot forget to use Drill Mode when you’re swimming in the pool.  If something - anything - goes wrong, all of your future training will be wrong.  This is because the data that that future training is based off of is bad data.  

‍

How does that sound?  Does that sound pleasant and enjoyable, or does that feel stressful to you?  From my experience coaching athletes, I can say with certainty that this need for exact data every single workout is not a good thing.  I only require exact data and no room for error 5-10 times per year for athletes who train by heart rate or power, and those 5-10 workouts (out of literally hundreds of workouts that they complete in a year) and their rules cause an immense amount of anxiety for them.  Therefore, I know for sure that having this pressure to have exactly correct data all the time will cause a lot of excessive, undue stress and angst for athletes.  

‍

Since humans are responsible for utilizing the devices that record data, there is always going to be a certain amount of errors in the data.  This means some of the data will be incomplete, if not missing entirely.  Thus, in addition to this unnecessary stress on the athlete themselves, training that is designed based on incomplete or bad data opens the door to so many undesired pathways, including to injury and unmet goals.  

‍

I’ve said it for many years (long before AI was even a thought, let alone a real thing): Bad data is worse than no data.  But an AI training plan cannot have no data, because if it doesn’t have data, it cannot write future training for you.  Furthermore, it can’t decipher when data is bad.  Even if you know that the data is bad because of something that went wrong during the workout, you cannot change the algorithm.  You cannot speak to the algorithm and tell it to disregard the data; it cannot account for subjective feedback like that.  It is just a computer sitting somewhere in the world crunching numbers, and only numbers.  

‍

Beyond Data: Embracing the Human Touch in Athletic Training and Coaching

Only a human can account for the humanity in an athlete.  While computers are designed and created by humans, they are not alive.  They are not sentient (not yet, anyway).  They are not capable of interacting with a human in a meaningful and authentic way.

‍

A human coach, however, can do these things - and more.  While humans are not perfect, technology (aka AI) also isn’t perfect.  (Just think about how many times something technology-related didn’t work as you wanted it to or how it was intended today.)  Computers, and all technology, are only as smart as the humans who program them.  And since computers are not sentient, computers cannot get to know you.

‍

But humans can get to know you - you as the unique person you are.  A human can (and does) see you as more than just the data you produce and the data you record.  A human can listen to subjective feedback and take it into account.  You can express how you feel about your training to a human and (correctly) expect that you will be understood.  A human coach can look at your data and your subjective feedback, and then see the data through a different (more accurate) lens.  That same human can then make predictions and plan training in a more holistic and more all-encompassing way than by using data alone.

‍

It should be noted that yes, it is possible for a human coach to use AI technology in their coaching.  However, I’ve seen a dependency on the technology (and by extension, the data) by coaches using AI technology.  In one particularly illustrative example, a coach who uses AI software told me, “I wouldn’t know how to be a coach without AI because coaching is hard.”  If a coach needs AI technology to coach, then that indicates that the coach doesn't have the knowledge that she should have.  A good coach can leverage technology, but it is not their supreme guiding light; a good coach can coach with or without technology.

‍

An athlete using AI-generated training indicates that that athlete has more trust in computers and technology than they trust humans.  Why trust a computer (which is designed by humans) more than you trust a human?

‍

The Bottom Line

AI-generated training has burst onto the endurance sports training scene with force.  It claims wild promises of individually-curated training that will help athletes reach their goals.  But this technology is limited in its scope because of what it needs in order to function: vast quantities of precisely and accurately recorded data. 

‍

AI-generated training has not yet proven itself to be anything more than a very sophisticated, nicely packaged, and well-marketed equation.  It’s my personal belief that athletes who have goals that are important to them deserve more than an equation.  They deserve authentic relationships, training designed based on all aspects of their humanity (not just the data they produce), and someone who cares just as deeply as they do about them staying healthy and reaching their goals.

‍