Performance Data Primer: Ethics
Just because you can…
Sport science has become remarkably good at measurement.
We can measure how much an athlete runs, how quickly they accelerate, how much force they produce, how their heart responds to training, how their autonomic nervous system responds to stress, and increasingly, what happens while they sleep.
The technological barrier to collecting information about athletes continues to disappear.
But perhaps the more important question is becoming increasingly difficult to answer:
Just because we can measure something, should we?
The ethical conversation surrounding athlete monitoring has struggled to keep pace with the technological one. We spend considerable time debating validity, reliability, signal-to-noise ratios, and whether a metric can predict injury or performance.
Those are important questions.
But there are questions that come before them.
Why are we collecting the data?
Who benefits from collecting it?
Who owns it?
Who gets to see it?
What decisions will be made because of it?
And perhaps most importantly:
Does the athlete understand and have a meaningful voice in that process?
The athlete cannot simply be the source of the data.
They need a seat at the table.
The other third of an athlete’s life…
Sleep tracking provides an interesting place to start because sleep is fundamentally different from most of what we traditionally measure in sport.
A force plate captures a few seconds.
A GPS unit captures training.
A dynamometer captures a testing session.
A sleep tracker follows the athlete home.
Müller and colleagues describe sleep as both highly sensitive and inherently intimate. Unlike steps or exercise, sleep occurs largely outside conscious awareness and typically within private space. Some sleep technologies can even inadvertently capture information about another person sharing that environment.
That distinction matters.
When an organization asks an athlete to wear a device overnight, we have crossed an important boundary.
We are no longer simply observing the athlete at work.
We are observing part of their private life in the name of improving their work.
That does not automatically make sleep monitoring inappropriate. There are legitimate reasons why understanding sleep may help athletes recover, perform, and stay healthy.
The potential usefulness of information does not automatically justify collecting it.
The question cannot simply be:
Would this data be useful to us?
It also needs to be:
Is collecting it reasonable to ask of the athlete?
When monitoring becomes surveillance
There is an important difference between monitoring an athlete and surveillance.
The distinction is not necessarily the technology.
It is the relationship surrounding it.
Consider two athletes wearing exactly the same sleep tracker.
One understands why the information is being collected, helped decide what information will be shared, knows who can access it, and regularly discusses the information with the performance staff.
The other was handed a device during preseason and told everyone on the team wears one.
Technologically, these situations are identical.
Ethically, they are very different.
Müller and colleagues make a similar argument around relationality. They note that collecting information about another person can create a knowledge-power imbalance, particularly when the person being monitored has not meaningfully agreed to that monitoring.
High-performance sport already contains significant power asymmetries.
Coaches control playing opportunities.
Organizations control contracts and scholarships.
Medical and performance staff influence availability decisions.
Now add physiological data.
An athlete might technically have the ability to decline monitoring.
But if everyone else on the roster participates, how voluntary does that decision actually feel?
Consent is more complicated than clicking I agree.
When the number becomes the athlete
There is another problem with monitoring that has less to do with privacy and more to do with interpretation.
Numbers have authority.
A sleep score of 62 feels objective.
An HRV value that falls two standard deviations below baseline feels meaningful.
A countermovement jump that drops 8% feels actionable.
But the existence of a number does not automatically mean something important has happened.
The Müller paper discusses the medicalization of everyday sleep: ordinary experiences become reframed through scores such as sleep quality and sleep efficiency.
In extreme cases, individuals can become so preoccupied with improving wearable-derived sleep scores that the pursuit of "perfect" sleep becomes counterproductive. The paper describes the emerging concept of orthosomnia, in which individuals place excessive trust in their wearable data—even over their own experience.
This is without us even getting into how inaccurate “sleep trackers” are anyway…I digress…
In it all, there is a lesson here extending far beyond sleep.
Measurement changes the thing being measured.
Once we give an athlete a readiness score, they know whether they are supposedly ready.
Once we tell them their HRV is suppressed, they know they are supposedly fatigued.
Once we tell them their force plate profile is abnormal, they know something is supposedly wrong.
The dashboard is no longer simply describing the athlete.
It may begin shaping how the athlete understands themselves.
That carries responsibility.
Don't get blinded by what you can track
This may be one of the greatest risks of modern sport science.
We become so fascinated with increasingly granular information that we lose sight of the person generating it.
Sleep score.
HRV.
Player load.
High-intensity accelerations.
Force-time characteristics.
Asymmetry.
Wellness.
Readiness.
Each provides another piece of information.
But more pieces do not necessarily create a clearer picture.
Sometimes they simply create more pieces.
An athlete can have suppressed HRV and feel great.
They can have a poor sleep score and perform exceptionally well.
They can have an asymmetry on a force plate and have demonstrated that same strategy for years.
And sometimes the most valuable piece of information in an entire athlete-monitoring system is remarkably sophisticated: "How do you feel?"
The point isn't that subjective information should replace objective information.
Data should inform conversations rather than replace them.
A seat at the table
This is where Jess Ellis' discussions around medical paternalism become particularly relevant.
Traditional paternalistic healthcare operated largely around the idea that the clinician possessed specialized knowledge and therefore should determine what was best for the patient.
Modern healthcare has increasingly moved toward shared decision-making.
Yet sport science can unintentionally recreate the same paternalistic relationship.
We have the data.
We understand the data.
Therefore, we will decide what is best for you.
The intention may be completely benevolent.
We want to protect the athlete.
We want to optimize training.
We want to reduce injury risk.
But benevolent intent does not eliminate paternalism.
The better model is not simply data-driven decision-making.
It is data-informed shared decision-making.
The athlete should understand what is being measured, why it is being measured, what the limitations are, who has access to it, and how it might influence decisions about their training, rehabilitation, availability, or performance.
And importantly, athletes should have a seat at the table when it comes to interpretation.
That doesn't diminish sport science.
It makes the process better.
AI raises the stakes
The ethical questions become even more important as sport science moves from collecting data toward predicting outcomes from it.
Machine learning makes an attractive promise.
Collect enough information about enough athletes and perhaps patterns will emerge that humans cannot see.
Injury risk.
Readiness.
Recovery.
Performance.
Return to play.
But machine learning does not magically transform imperfect information into truth.
It scales whatever assumptions exist within the system.
A recent systematic review of AI ethics in sport identified four recurring concerns: fairness and bias, transparency and explainability, privacy and data ethics, and accountability. Privacy and data ethics appeared in 22 of the 25 included studies.
Wearable AI creates similar concerns around informed consent, biased training datasets, data aggregation, and opaque automated decisions.
This becomes particularly dangerous when predictions begin influencing decisions.
Imagine an algorithm tells us:
ATHLETE X — 78% INJURY RISK
What happens next?
Does their training change?
Do they practice?
Does the coach see it?
Does the athlete see it?
Could it influence selection?
What if the prediction is wrong?
And perhaps most importantly:
Who is accountable for the decision?
The sport scientist?
The physician?
The coach?
The vendor?
The algorithm?
There is a tendency to treat an algorithmic recommendation as though responsibility has been outsourced to mathematics…it hasn’t.
My newest pet peeve…THE PROMPT BOX ISN'T A PRIVATE DATABASE
Artificial intelligence has made working with performance data remarkably easy.
Export a spreadsheet from your athlete management system.
Upload it to ChatGPT, Claude, or another AI platform.
Ask it to identify trends, write code, build a report, summarize an athlete's history, or look for relationships you may have missed.
Within seconds, you have an analysis that previously might have required hours of work.
But convenience can obscure an important question:
What exactly did you just give the model?
A spreadsheet containing an athlete's name alongside sleep, wellness, HRV, training load, injury history, or rehabilitation data isn't simply a dataset.
It is information about a person.
And uploading that information to an external AI service means potentially moving athlete data outside the systems and governance structures your organization established to protect it.
The fact that information can be copied and pasted into a prompt box does not mean we have permission to put it there.
This is particularly important because athlete datasets are often more identifiable than we realize. Removing a name is a good start, but de-identification is not necessarily the same as anonymity. Team, position, injury, testing dates, game participation, and other variables can make an athlete recognizable when combined.
The same ethical questions we ask before collecting data therefore need to follow the data throughout its lifecycle.
Who is receiving it?
Where is it being processed?
How is it retained?
Could it be used for purposes beyond the original intent?
Has the organization approved that platform?
And most importantly:
Does the athlete know their information is being used this way?
AI can be an incredibly useful tool for sport science. We shouldn't avoid it.
But there is a significant difference between asking an AI system to write code for analyzing countermovement-jump data and uploading identifiable countermovement-jump data from your entire roster.
Use synthetic data.
Remove unnecessary identifiers.
Work within organization-approved environments.
Understand the data policies of the tools you use.
And treat athlete information with the same care whether it is sitting inside an athlete management system, an Excel spreadsheet, or an AI prompt.
A new tool doesn't create a new standard for privacy.
If anything, the ease with which AI allows us to move and analyze information means our standards need to become higher.
Garbage in…garbage out…
There is also a more fundamental problem.
Algorithms learn from the information we provide them.
If our inputs are noisy, poorly validated, biased, incomplete, or disconnected from the outcome we are trying to predict, a sophisticated model does not rescue the process.
It simply produces a sophisticated answer to a poorly constructed question.
A model built from GPS data, HRV, sleep scores, force plates, wellness questionnaires, and training load may appear extraordinarily comprehensive.
But complexity should not be confused with understanding.
Before asking whether machine learning can find a pattern, we should still ask the first-principles question:
What mechanism would make this information relevant to the outcome we are trying to understand?
Otherwise, we risk producing increasingly complex models while becoming increasingly disconnected from the athlete standing in front of us.
Data stewardship
Maybe ownership isn't even the most useful way to think about athlete data.
Perhaps stewardship is better.
The athlete generates the information.
The organization may collect it.
The practitioner interprets it.
The technology company may process or store it.
Each participant therefore assumes responsibility for protecting it.
Müller and colleagues argue that responsible implementation requires considering the vulnerabilities of users, involving users directly in development, and clearly communicating what technologies can—and cannot—provide.
That philosophy translates directly to sport.
Before collecting another stream of athlete data, perhaps every performance department should be able to answer a few simple questions:
Purpose: Why are we collecting this?
Necessity:Do we actually need it?
Access: Who can see it?
Transparency:Does the athlete understand how it is being used?
Agency:Can the athlete meaningfully participate in decisions arising from it?
Interpretation: What can this metric actually tell us—and what can't it?
Action: What decision will change because we collected it?
Accountability:Who is responsible when that decision is wrong?
If we cannot answer those questions, another wearable probably isn't the solution.
The forest or just the trees
The future of sport science will undoubtedly involve more data.
Better sensors.
Continuous monitoring.
Computer vision.
Integrated databases.
Machine learning.
Artificial intelligence.
That progress is exciting.
But technological capability cannot become our only measure of progress.
The goal of sport science was never to build the most sophisticated dashboard.
It was to help athletes.
And athletes are more than the collection of variables we can extract from them.
Sometimes the most advanced thing a sport scientist can do is recognize that a measurement isn't necessary.
Sometimes it is acknowledging uncertainty in the data.
Sometimes it is asking the athlete what they think.
And sometimes it is choosing not to collect information simply because we can.
Don’t forget to see the forest through the trees. Take a step back and consider the big picture.
The athlete should never become a passenger in a system built from their own data.
They should have a seat at the table.
This blog was inspired by:
Jess Ellis’s vocalness and work on ethics and medical paternalism in sport.
TOMMY OTLEYPhysical therapist, sport scientist, educator, and researcher working at the intersection of rehabilitation and performance.
READ MORE
As sport science becomes increasingly data-driven, every measurement, model, and prediction comes with assumptions. This edition of The Film Room explores the assumptions hidden within data collection, processing, injury epidemiology, and prediction—while arguing that the goal isn’t to eliminate assumptions, but to understand them and let their limitations shape our confidence in the decisions we make.