A.I IN AFL PART 3/3

Last week Sam Mitchell caused quite a stir when he admitted to using AI in his coaching process and although what was reported contained very limited detail of what he actually uses and what he uses it for, the uproar was ridiculous.

There is no doubt that AI will become a part of Aussie Rules in the near future as team's are always looking for ways to optimise and streamline their approaches and there's not an easier way to do that then through AI means.

On Twitter I came across this comment on the subject which of course sent me down the rabbit hole of this 341 page study titled "Artificial Intelligence-Based Decision-Making Support During Australian Football Matches" and here is part 3 of 3 of my numerous notes from it.

• AI prediction of match outcomes study
• Can machine learning models based on technical performance and not score margin, predict match outcome in real time?
• All models performed well (73.5 – 75.8% prediction rate) v benchmark score-based model (77.4%) with accuracy being at its lowest at the start of the match (45.7 – 48.8%) and increasing to peak near the end of a match (87.2 – 92.7%)
• A DSS is a computer system designed to support the decision making of the user
• A handful of technical PI’s are able to predict match outcome from 78.9 – 95.1% but the models weren’t designed for in-game us, only pre/post-game
• Wanted to test other metrics but scoring-based one’s and opted for meters gained, meters gained per kick, time in possession, inside 50’s, turnover’s, time in forward half, contested possession, effective disposal, groundball gets, marks on lead and rebound 50's
• Defensive stats x 1v1 losses, rebound 50 differential, tackles, 1v1 wins, spoil differential
• Transition stats x intercept from mark, defensive 50 rebound to boundary %, defensive 50 ball movement to wing %, defensive 50 ball movement to corridor %, turnover’s
• Offensive stats x inside 50’s, repeat inside 50’s, uncontested marks, initial inside 50’s, leads targeted, scoring accuracy
• Data should guide you with what to do, not just tell you what to do
• Suggestions were made for it to dig deeper so they added an insights option which provided breakdowns of the subcategories of PI’s contributing to the broadly defined PI’s which formed the suggestion and the coaches liked it
• What might be best is a combined approach where coaches provide a list of specific changes they might make during the game or have made before, coupled with their pre-conceived expectations of the way these would impact the values of their PI’s and instead of suggesting numerical PI targets, the AI could then suggest a small selection of these specific changes which would satisfy the teams needs according to the coach’s preconceived expectations and then the coach picks the most appropriate change in the moment
• Long term, data can be collected on the actual impact of these changes and this database could be used to provide more objective actionable suggestions
• Decision-making stages AI could support include opportunity trigger, understand the opposition, determine the need for action, explore options, evaluate the decision
• Impact AI could have on decision-making constraints x manage information volume, reduce emotion, reduce time pressure, prompt early coach reaction, direct focus
• An AI-based DSS should update regularly, be easy to navigate, use color, be customisable, be transparent in its confidence, provide the ability to manually dig deeper
• Coaches respond to an opportunity trigger, they endeavour to understand the opportunity and then determine the need for action
• If action is required, coaches explore options, take action, and then evaluate the decision
• AI can handle far more information in time-constrained environments than humans and can pick up on patterns not recognised by the coach
• Coaches should consider opportunities within current in-game decision-making/coaching practice’s where they would be willing to be supported by AI and communicate this with those responsible for building the DSS, use the discovery of context specific decision-making processes as an opportunity to reflect on your own practice by asking yourself: h
• How am I becoming aware of an opportunity?
• Am I truly considering the underlying cause of the opportunity before deciding?
• Is taking action always necessary and how do I assess whether action is necessary?
• Where am I sourcing my options from and am I leaning on the experience of others around me to support the exploration of options?
• How/when do I take action and is my message getting across to the right people in an effective manner
• How do I evaluate whether a decision was the right one?
• The discovery of context specific decision-making constraints during matches serves as an opportunity to minimise some of these constraints through means other than AI so ask yourself:
• Are there any ways that I can improve the way I communicate my thoughts to players/assistant coaches?
• What impact does emotion have on my ability to make good decisions and how can I be more objective
• What are the most important pieces that I want to be fed during games and have I communicated this effectively to my coaching team/support staff?
• Work closely with analysts to find common ground on the types of DSS based solutions you think could be useful and provide feedback on any barriers you face when using the DSS
• Give a DSS time to evolve, communicating with the developer about what would help build trust in the system
• For analysts you are the conduit between the DSS and the coach and its effectiveness will depend on your communication of suggestions from the DSS to the coach, work closely with the person designing the DSS and provide suggestions for improvements + understand the underlying processes/technique involved so you can adequately explain suggestions
• For researchers/practitioners/analysts aiming to design/develop/implement a DSS to gain an in-depth understanding of how coaches make decisions in their environment, you need to:
• Build a DSS that fits their current processes
• Consider ways to incorporate additional data types to improve on the current method
• Work closely with coaches/analysts and consider the implementation of a DSS as a continuous process
• Ensure the suggestions the DSS provides are sufficiently actionable for the coach
• Be willing to compromise on perfect science to provide a DSS coaches will find most useful
• Any new system will be competing for attention so incorporate desirable functions of older/pre-existing systems into the new DSS