There’s a lot of buzz about using artificial Intelligence in athletic coaching, particularly in its use to create and manage dynamic training plans that automatically optimize prescribed workouts based on feedback.

Whoever cracks this can use the technology on two ways:

  • To dramatically reduce costs.

If a computer can replace a coach the cost saving is enormous, making it possible for anyone to access elite level personal coaching. This would be similar to the positive impact that the reduced costs of power meters has on rider training.

  • To Improve the effectiveness of professional coaches

Computers can work with the coach to quickly analyse and compare the huge amounts of data already available.

Machine learning algorithms sort the wheat from the chaff and provide the coach with insights way beyond their own personal experience.

This allows coaches to focus on the creative, human aspect side of their job

At the moment people generally either self coach or employ a professional coach. AI coaching opens up a third possibility which is potentially much more powerful.

Coaching Methods for endurance cyclists

Self Coaching: This is based on your own knowledge, experience and data and perhaps a close circle of friends and forum posts.

Professional Coach: This is based on a trained and experienced professional’s knowledge, experience and the data from his or her cohort of athletes past and present.

AI Coach: This is based on all public domain and available private domain knowledge, experience and data. All the data is categorized and analysed for trends, causations and correlations.

People like to use computers to manage their personal development because computers aren’t judgmental or critical. They report the facts, good or bad. They don’t sugar coat them and they don’t gossip to friends and colleagues!

The downside of this is that if you want to skip a days training the computer isn’t going to make you feel bad!

So what exactly is Artificial Intelligence (AI)

AI is a catch all phrase that describes the ability of computers to learn. This is more correctly referred to as machine learning.

There are three types of Machine learning:

Supervised:

The computer is given a set of instructions to follow and a huge dataset that has been categorized or tagged by humans. Using this training data the computer learns how to categorize things according to the instructions and can then go ahead and categorize untagged data on its own.

Unsupervised (also known as Deep Learning):

The computer is NOT given a set of instructions to follow, only the huge dataset, and that dataset has NOT been categorized or tagged by humans. Using just the data the computer works out how to categorize AND goes beyond that and may develop advanced capabilities such as the ability to devise strategies and insights.

Semi Supervised

As you might imagine this is a blend of Supervised and Unsupervised machine learning with human intervention where that adds value.

The best AI solutions today are human supervised

Humans have many skills that go way beyond what is possible with computers today so almost all projects use AI to augment these human skills (Semi Supervised) rather than replace them.

For example: AI is used in oncology to rapidly scan thousand of images and categorize the tumour types. But AI isn’t replacing the radiologists who make the actual decisions on treatment plans.

Many products advertised as AI actually use humans as they need much less training data and very little processing power so actually can work out a lot cheaper and more effective than AI.

The idea is that these projects would use humans until it became viable to transition to AI. However in many cases this tipping point is never reached so the reference to Artificial Intelligence is just used as a marketing tool.

How can a computer learn things?

Machine learning describes teaching computers to do things that humans seem to do instinctively, (but actually have been learning since birth).

For example humans can easily tell the difference between a dog and a cat by looking at certain characteristics that are unique or predominant in each.

Machine learning is the transfer of this knowledge to the computer which then uses this as the basis to make logical decisions.

A key advantage of using AI is that computers don’t get tired or bored, once trained their results are consistent over 100 or 100 million classifications.

Here’s the example of the dog and the cat

It’s pretty easy to tell a dog from a cat, you’d probably have a 70% success rate choosing between the two using the information below.

  • Cats ears are pointy, dogs ears are floppy.
  • Cats have long thin tails they use for balance. Dogs have a variety of tail types which they wag to show happiness.
  • Cats have slitty eyes, dogs have round eyes.

Add in another characteristic and you can get the success rate up to 90%+

How about sound? Dogs go woof, Cats go meow. You get the idea!

Supervised machine learning is used to categorize things

In supervised machine learning we program the computer to look at thousands of images and sounds that we have categorised (tagged) ourselves as cat or dog, we tell the computer this one’s a cat that one’s a dog. The computer takes these categorised images and has a go (remember we already told it the answers). When it gets one right it gives itself a pat on the back, when it gets one wrong it makes a note and we can program it to try again until it gets it right.

Eventually it gets pretty good at doing it (but probably never perfect).

Supervised machine learning is used in: Image Categorisation, Speech recognition, Medical Diagnosis, Financial decisions.

Supervised machine learning requires MASSIVE amounts of Data and MASSIVE human resources to classify the training data. Once the model is trained it can save a lot of time and money by automating repetitive tasks such as product categorization or checking scans for tumors.

Unsupervised machine learning means computers get to make objective decisions and discover insights

Unsupervised machine learning (True Artificial Intelligence) is MUCH harder to achieve. This is where we seed the computer with the data set, then let it solve problems on its own and without any human intervention.

Unsupervised machine learning is used in: Advanced game playing computers, Smart speakers, autonomous vehicles, unmanned aerial vehicles, context-aware devices are all examples of unsupervised machine learning.

Unsupervised machine learning requires MASSIVE amounts of data and MASSIVE Processing power but is much more powerful as the computer will develop hunches and see patterns we miss.

The one thing all types machine learning need is MASSIVE AMOUNTS OF DATA. More data always means better results. A better algorithm will not make up for a lack of data.

In the world of cycling coaching this means that only Training Peaks and Strava are likely to have any chance of making AI work.

How can Artificial Intelligence Help coach endurance cyclists today

I’ve looked at some of the tasks we rely on coaches to help us with to see how AI could be used.

Review Athlete attributes and Goal Setting

Machine learning can help here. Goal setting is a critical part of the coaching regime.

Where is my performance now? What do I it to be? What resources do I have and what do I need? Is my goal realistic or do I need to change it? Etc etc.

A coach can do all this based on their experience. A computer can do this based on available data from all other athletes who have followed the same path as well as estimated data based on patterns and trends.

Data Collection.

The more parameters you can measure the better the results. A coach needs to focus on a few key performance indicators (KPIs). The computer can take EVERTHING into consideration and sort the wheat from the chaff.

There is way more data available than any human can interpret without help from computers.

Motivation.

AI can help here by spotting patterns that show lack of motivation or too much motivation causing over-training.

Data Cleansing, Gap Filling, Sense Checking and Error Correction.

80% of a data scientists work is data hygiene and it’s absolutely crucial for meaningful results. Think GIGO (Garbage in Garbage Out).

Data hygiene is a human intelligence task augmented by tools to make it easier.

Bad data can lead you down blind alleys and is extremely dangerous.

Data Presentation

Humans work best when data is visualized as it is much easier to spot trends and patterns in a chart than in a list of figures.

There are countless tools to help automate this process with the best showing you the big picture then letting you drill down into the detail.

Insights and Feedback

This is where the coach really earns his money! The bottom line is that a good coach will use AI enhanced data to help optimize your training regime. AI will be able to do this on its own in the future but at the moment the financial cost is way out of reach.

Strategy

This is a perfect example of where AI can make a big difference now. AI is used to analyze historical data and outcomes to forecast future results.

Many sports already use AI to create strategies and game plans and AI will only become more dominant in this area.

Conclusion

AI offers huge potential to transform athletic coaching. The current state of the art means that coaches have a range of great tools available to make them more effective and 20 years down the line it’s not outside the realms of possibility that computers could replace coaches altogether.