What AI-powered lifestyle coaching might look like for you and me
At some point, your next workout routine or weekly nutrition suggestions will likely come from an algorithm. And that’s a good thing!
I’m quite optimistic about technology in general. While I see all the ways it can be abused, I also see all the ways it can be used to make our lives better. Whether we want to do the former or the latter is fundamentally our choice.
Why do I mention this? Well, this is a post about algorithms. And algorithms seem to have gotten a bit of a bad rap lately. They’re biased, stupid and have no shortage of immoral qualities. This usually stems from reports on cases where algorithms have been deployed as dopamine engagement machines to optimize short-term satisfaction, or trained on inadequate and biased data and left to their own devices without human supervision.
But if you tell algorithms to help optimize people’s health outcomes, and you have a sympathetic and well-founded definition of health combined with great datasets and the right humans in the loop, they will certainly be able to do just that. This is what we want to do at Entirebody. The only dopamine rush our algorithms will give you is the one you get when you reach your goals and improve your well-being.
General algorithmic approaches to solving problems have no moral properties. Only people do, and by extension the use cases we choose to pursue and the utility functions we define for our solutions. If we pursue good use cases with good utility functions, technology and algorithms is a force for good.
With that out of the way, let’s talk about exactly how we can deploy AI-powered fitness and lifestyle coaching algorithms to make the world a better place, providing access to such services to those who would otherwise not have it!
We ended our last blog post in this series by claiming that while the full-fledged robot personal trainer will probably remain within the realm of science fiction for quite some time yet, the profession of coaching is ripe with small tasks where AI can both amplify human coaches, or replace them entirely. In this post, we delve deeper into those tasks.
Two ways to Rome: coach-facing and client-facing
Like we have alluded to previously, there are two ways to deploy AI for fitness and lifestyle coaching: as a tool for coaches to work more efficiently with their clients to optimize their results, and as a standalone digital coach that interfaces directly with consumers.
These are not mutually exclusive use cases; there is a lot of overlap between them, and they are likely to co-exist in a digital ecosystem for coaching.
While the endgame for AI in fitness would be to have autonomous systems that work well enough to interface with consumers directly for almost all coaching-related tasks, there are good reasons for keeping human coaches in the loop — albeit less and less as the product matures.
When deploying AI solutions to new problems where there is a significant cost associated with making mistakes, it is usually a good idea to run things through a human filter until you are confident that it is good enough. This also facilitates a feedback mechanism that contributes to making the algorithm better in the first place, with humans being able to correct the algorithm when it makes mistakes.
Matching coaches and clients
So if we assume that human coaches are still going to be a major presence in most cases, a natural starting point for improving coaching using algorithms would be coach-client matching.
Facilitating great coach-client relationships where clients are matched with coaches that are well suited to help them achieve their particular goals is paramount for any coaching service. AI can be applied to do this at scale, by leveraging historical data on coach-client relationships, their respective attributes, and their outcomes.
With the marketplace for coaching services going increasingly remote and even global, algorithms will be a necessity to establish the right partnerships.
Other marketplaces have already been doing this type of thing for years, and here the stakes are even higher because users might be committing to something for a long period of time and investing blood, sweat and tears to improve their health.
Pairing you with the right fitness and lifestyle coach is much more important for your well-being than recommending the next thing you should watch on Netflix. But the problem is actually the same — find the combinations of users/clients and content/coaches that optimize engagement/success — and we have gotten really good at making recommendations in general. With the right data and platform, Netflix for coaching will become a reality.
Recommending and tailoring your next routine …
Now that we have addressed the optimization of coach-client relationships, we can move on to the coaching process itself.
The first step of any lifestyle change — whether it be self-, human- or machine-guided — is to set up some sort of initial plan or guidelines for the two most important pillars of physical health: nutrition and physical activity.
Depending on the condition and goals of a client, this may be everything from a once-a-week spinning session and a few nutrition bullet points to follow, to a multi-activity training regimen designed to maximize performance and a detailed meal plan for every day of the week.
Crucially, any plan needs to take the particular preferences of the client into account, which means that my plan should probably be different from yours even if we have the same goals.
By learning from what has worked and hasn’t worked for coaches and clients in the past, AI can be used to suggest the initial approaches — be it complete training routines and nutrition plans or just personalized guidelines — that are most likely to work for each client.
Like mentioned previously, this can be deployed either towards coaches or towards clients directly. In the former case, a coach would make any appropriate changes based on information that is not captured by the algorithm, approve the plan and ship it to a client. In the latter case, users could just tweak the suggestions themselves to fit their preferences and have at it with no human on the other end should they so desire.
While there is probably a lot of value to be created by simply matching clients with preset plans, routines and guidelines, the natural next thing to do would be to have AI generate brand new, custom tailored workouts and nutrition plans for everyone, or simply make changes and replacements to presets. This will be important for addressing issues like injuries and other physical limitations, as well as all other preferences and constraints.
Another thing to consider is that not everybody knows what their motivation and goals should actually be to improve their long-term satisfaction and quality of life, given their starting point and current state.
Pursuing the “wrong” goals is quite common. For example, where many people tend to overemphasize weight loss as a goal in itself, coaches may often prefer to emphasize mastery and performance in physical activities first, with weight loss being the byproduct of the lifestyle change rather than a holy grail to chase. Given information about a user’s current state, AI will certainly be able to give the same advice by comparing your likelihood of success with different approaches.
… and adjusting it on the fly to optimize results!
But to truly amplify the benefits of coaching using AI, we need to take it one step further. Getting and having a custom tailored plan is one thing. Getting help when the plan needs to be adjusted or when it fails, or getting that friendly — or not so friendly — reminder at the right time is probably the main benefit of coaching when it comes to establishing lasting lifestyle improvements.
What should you do to recover optimally if you have a few terrible workouts? What is going on with your diet when the needle on the scales hasn’t moved for a month? How should you restructure your meals and workouts the next week around the birthday party and your business trip?
Fear not. AI will be able to change the routines and recommendations it provides to coaches and users continuously; adapting to unforeseen circumstances on the fly, and evolving to incorporate new things it has learned.
The trainee who has a few bad workouts will automatically be put back on the right track with a tailored recovery session or two. More calories will be added to the plan to account for the birthday party, and the workout sessions will be restructured around the business trip and the cake — to make sure the cake can both be enjoyed and used to maximize performance.
The weight-conscious will no longer have to worry about the mercilessly unmoving needle on the scales because the AI knows that everything is actually going according to plan and predicts that the healthy target weight will be reached in time. And the religiously performance-oriented will be able to see their workout plans morph before their eyes based on the current context to squeeze out those extra few percent, and their nutrition plan update accordingly.
In summary, AI can act as a personal trainer on steroids (no, not literal steroids …) that follows you up every time you want to or need it. It doesn’t need sleep, it doesn’t need to bill you for every interaction, and it can serve more clients through adding compute power and memory to machines in a data center instead of us having to invent human cloning for PTs.
By now it should be obvious there is no shortage of great use cases for AI in fitness and lifestyle coaching. But how do we actually make all of this work, and how do we make it really good?
We will explore this in the next and final part of this series of blog posts. Without giving too much away, I can reveal that the answers actually have much more to do with data, design and humans than with the algorithms themselves!
