Thanks for submitting!

  • Jo Clubb

Diving Deeper into AI in Sport

Updated: Jul 27

Thanks to everyone who responded to my appeal to canvas industry opinion on artificial intelligence (AI) in high performance sport. I received an array of questions and comments, from both practitioners and researchers all over the globe. This post shares the key themes that emanated from the discussion.

Going forward, I am delighted to be engaging with Zone7 - a leader in AI application for injury risk forecasting and performance management - to discuss each of these areas further. I will be collaborating with co-founders Tal Brown and Eyal Eliakim, as well as Performance Director Rich Buchanan to dive further into each theme.

Questioning The Data Pipeline

Interestingly, there were questions and comments throughout the entire data pipeline, from collection, through transformation, analysis and modelling, to delivery and application (see Figure 1). This reinforces the need to think critically about the entire data process, which is not just specific to AI, but anytime data is collected in sport. Balancing innovation with validation can be a challenge for early adopters, as witnessed with GPS technology uptake in the early 2000s. Yet, as scientists, we should critically evaluate the tools we are using. This includes statistical and modelling techniques.

The data pipeline represented with Collect, Transform, Analyse, and Deliver, with questions shown around each of these levels.
Fig. 1: Industry feedback questions along the data pipeline

There were questions around data quantity and quality of the inputs going into the model. This is reflective of a sports science practitioner’s process across multiple data streams; “how many trials will provide sufficient quality” or “what is the quality of the tracking data given today’s conditions/location” for example. Yet, one thought that strikes me is perhaps not “do we have enough”, but “what value can we gain from the data we already have” …

That said, data quantity in our field warrants much discussion. Underpowered studies, those with insufficient data points, have been suggested as an issue in sports science (Abt et al., 2020). As such, aggregated data sets may provide greater insights. For instance, Rich Buchanan has previously shared on Sportsmith a Zone7 case study in which they performed a retrospective analysis on a dataset from 11 football teams’ datasets that included 423 injuries.

As datasets are aggregated, it is important to maintain the context of each specific sport, team, and athlete. Contextual factors have been identified as a central theme to load management (West et al., 2021), so it was not surprising to see context frequently come up in the responses. How can AI techniques incorporate context, such as the playing position, environmental conditions, coach’s playing style, time in season, and the different technologies used to capture data, for example.

Regarding the modelling process itself, there were comments on transparency. This is a hot topic, especially given the connotations associated with the “black box” approach. Like most topics, social media seems to portray this as a binary argument and yet, I believe there must be much greater nuance. Therefore, it is an important theme to explore further.

Finally, the question of “so what?”. What do practitioners do with AI outputs? There are concerns that hark back to Little Britain’s catchphrase “computer says no”! How should practitioners interact with and apply the outputs? Do they replace or augment decision making? Do they impair or support coach intuition? This topic, in particular, I am really excited to explore alongside the Zone7 team.

Key Themes

From the responses, I have extracted the following six key themes:

Definition: What do AI and ML mean?

Inputs: Data quantity and quality

Context: Incorporating situational factors through human input

Transparency: Validation and access of the algorithms

Outputs: Injury risk and load management as outcomes

Application: How do practitioners apply the outputs?

Defining and explaining the premise of AI itself seems like the best place to start, before we delve into each area. Thank you to everyone who contributed their questions and concerns. This topic is one that necessitates discussion and exploration so let’s keep the conversation moving forward.

Thanks also to Zone7 for engaging in this discussion. Keep a look out for the first response piece coming soon!