Doug Bonderud

14 January 2021

8 min read

Related
Links

Sweating as a service

By JP Mangalindan

6 min read

The new health economy

By Ryan Lawler & Mimi O Chun

2 min read

Personal fitness is now a $100 billion dollar industry, and the market shows no signs of slowing down. But how consumers are staying in shape in the wake of a global pandemic has drastically shifted how and where those dollars are being spent.

While previously consumers might have gone to the local gym or attended studio classes, more than ever people have turned to at-home virtual classes and connected fitness equipment. At the same time, that shift has opened the door to fitness frameworks driven by artificial intelligence (AI) that take into account their strengths, weaknesses, and overall fitness goals in a way that traditional gyms and training programs can’t.

Supported by wearable devices sporting simple tracking technologies for things like heart rate, overall exercise activity, and sleep patterns, AI fitness solutions are making a play for mainstream fitness consumers by offering a truly personalized training program

Capitalizing on this market demands robust machine learning muscle, however. That’s why equipment manufacturers and fitness companies are using machine learning applications to handle the massive amounts of data processing required to deliver personalized fitness programming.

The new normal: Fitness-as-a-service

A number of companies are harnessing the power of machine learning to help develop personalized fitness applications. Products from FitnessAI, Tonal, and Tempo, for example, all use AI to help users meet their fitness goals. They all incorporate some form of machine learning to collect, interpret, and apply anonymized user datasets at scale, in turn making the potential of personalized fitness both possible and practical.

  • The FitnessAI iOS app uses data from about 6 million workouts to build customized fitness plans. Users simply enter basic biometric and goal data, and the app creates a personalized training program that specifies what exercises to do, what weights to use and how many reps to complete. The app then uses an AI algorithm to suggest a progressive increase in weights and reps relative to user size and strength, in turn providing a more personalized training experience. Applications such as Freeletics fill a similar niche, allowing users to define their own goals and customize nutrition plans.
  • Tonal offers a resistance-based weight training experience with a sleek form factor that adapts to user feedback. Using a combination of digitally controlled magnets and electricity, Tonal offers up to 200 pounds of resistance for both upper- and lower-body workouts along with a 24” flatscreen. Tonal’s AI infrastructure starts with a full-body strength assessment, followed by a workout that matches resistance intensity to user capability. The system then tracks user effort over time and increases resistance automatically to limit the potential of a physical plateau.
  • Tempo replaces resistance-based training with more traditional dumbbells and barbells that come as part of a freestanding cabinet package that includes a 48” touchscreen. Integrated 3D cameras capture anonymized data about body position and form during workouts to help improve the overall effectiveness and personalize training plans. While instructors on-screen demonstrate specific movements, Tempo’s AI-enabled camera system monitors user form and rep counts to offer real-time feedback.
Component smart fitness intro2
Big data, big gains: The AI advantage

Sports science is now a data-driven field. Research into human capabilities, limits, and overall performance has led to the development of generalized programs that help build strength, reduce fat, or improve endurance.

On an individual level, however, performance and potential stray from the mean — every person’s physical makeup is different, meaning that they perform, adapt, and gain strength or endurance at vastly different rates. Traditionally, certified physical trainers filled this gap, and their in-person expertise combined generalized knowledge with client characteristics to shape programs suited to each individual.

But this model doesn’t work for everyone. In many cases time is a factor, since work schedules or child-care commitments may limit the ability of clients to schedule sessions. In others, convenience and comfort play key roles, particularly during the COVID-19 crisis, as individuals may not want to attend in-person classes or have instructors in their homes. Recent pandemic pressures also frustrate the efforts of personal trainers to connect with clients — for example, Zoom fatigue is a real and growing problem.

AI in personalized training offers a way to bridge the gap by leveraging machine-learning algorithms to aggregate generalized physical data, collect specific (and anonymized) information about users and then combine these datasets to create truly personalized training programs.

As noted by Simon Alger, lead data scientist at Freeletics, “the main benefits for users are access to training planning, monitoring and even motivation at a fraction of the cost, which means that more people can be reached than ever before.”

In practice, integrating machine learning technology into fitness equipment requires access to massive amounts of personal user data, such as current fitness level, height, weight, and, in some cases, anonymized images of body shape and type. Then, fitness companies must develop AI outputs that deliver individualized suggestions and lead to sustainable fitness improvements over time.

Three key benefits of AI-driven programming

Specificity

While trained fitness professionals can offer customized physical activity frameworks, AI tools provide access to truly specific workout plans. By using a combination of user-entered information, movement data captured by on-board cameras, and physical measurements taken by sensors that detect physical effort exerted or monitor user form during exercises, intelligent fitness tools can deliver specificity on a level that was previously unattainable.

Scalability

AI fitness frameworks also help improve user outcomes by intelligently moving the goalposts. Consider Tonal — as users get more comfortable at current resistance levels, the device automatically increases resistance to ensure ongoing improvement. This scalability helps provide long-term value for users by delivering dynamic rather than static goal-setting that evolves in tandem with their performance.

Many companies are also pairing their AI offerings with certified, instructor-led classes that allow users to access professional trainer expertise without the in-person component. According to Pete McCall, ACE-certified health and fitness expert, this approach offers “a major competitive advantage over traditional fitness facilities in that they can deliver workouts to an almost unlimited number of consumers.”

Sustainability

One of the most familiar — and frustrating — facets of physical training is the “plateau.” This occurs when current exercise regimens stop delivering measurable results but instead lead to a “leveling out” of personal performance. For many gym-goers or trainer clients, this lack of measurable progress leads to waning exercise interest and eventual abandonment.

AI-powered training programs, meanwhile, offer a more sustainable success approach. By combining current user data with aggregate workout information from millions of other users, AI tools are better positioned to create programs that deliver steady gains and sustainable interest.

Let’s get digital

Technology-driven frameworks offer a new way to create more customized workout routines, but the effectiveness of these models depends on critical, digital components.

The first is big data. This includes any anonymized data collected by fitness devices along with any personalized data users are willing to provide. While generalized data about fitness goals and user-submitted information around metrics such as height and weight are helpful, digitally captured and anonymized body image and movement data points make it possible to create much more robust machine learning models.

Backend computing throughput is also critical. Tuan Phan, a partner at Caplock Security and member of ISACA’s Emerging Technology Advisory Committee, notes that the uptake of evolving fitness frameworks thankfully coincides with the availability of lower-cost compute performance. “The computing costs have dropped significantly,” he says, “What would have been prohibitively expensive can now be done in the cloud.”

Connectivity is also critical. For example, the adoption of robust 5G networks makes it possible for fitness devices to manage the massive bandwidths required for digital data transmission. Coupled with the rise of edge computing initiatives that shift the burden of ML and AI processing away from centralized cloud nodes these connected devices are better equipped to deliver on-demand, personalized guidance and assistance.

It’s worth noting, however, that even a fully-tested and reliable model isn’t enough to guarantee success. While great data helps design better fitness plans and reduce the risk of exercise outliers that don’t match user goals, bias presents a potential pitfall to companies looking to introduce AI-driven fitness programming.

More reps less bias

Bias isn’t a new problem for artificial intelligence — from over-representation of certain datasets to unconscious developer decisions, unintentional bias can cause problematic shifts in AI output.

The same scenario is potentially problematic for fitness programming, where undetected bias could favor specific body types or workout routines and result in customized content that isn’t so “custom” after all.

Phan says that battling bias starts with robust and reliable validation to ensure algorithms aren’t favoring a specific attribute. He also highlights the need to remove attributes that aren’t significant to eventual output, while also correcting for the human tendency to “over fit” models by weighing specific attributes more heavily than others. “If the attribute is marginal and you believe it should be significant,” he says, “you may need more data.”

Much like exercise itself, the biggest benefits for battling bias come from repetition. The more data analyzed and the more often analysis is conducted, the more reliable the results. This is critical for user satisfaction — if customers have provided biometric data to fitness applications with the intention of meeting specific goals and AI-driven tools can’t deliver, it’s hard to justify the substantive cost of these solutions.

Challenge notwithstanding, machine learning technology is becoming an increasingly accessible option for firms looking to build out data-driven fitness functions. According to a recent Gartner report, machine learning technologies have passed the peak of inflated expectations and are now being driven by the democratization of AI frameworks to deliver business value.

As a result, machine learning solutions baked into connected fitness technologies now offer a way for users to receive personalized, real-time feedback about their fitness efforts in order to help them achieve specific goals over time.

While the long-term impact of AI-focused fitness frameworks isn’t certain — some users many find themselves ready to get back in the gym or opt to continue their at-home journey — the rapid expansion of this exercise market vertical makes it clear that truly personalized training with integrated machine learning muscle has arrived, and it’s going strong.

* * * * *

If you have a startup looking to connect with Next, consider joining our Global Innovators Directory. We'd love to learn how we can help you achieve your goals - whether it is through investment, partnership, or collaboration.


Join the Global Innovators Directory