AI in Its Infancy - How Helpful Are Chatbot Training Plans?

1. KI: Support for trainers
2. Training plans for cycling and fitness
3. Top sports: AI-based coaching system
4. Freeletics success: AI conquers mass sports
5. The more valuable data, the more effective AI becomes
6. Artificial intelligence warns of dangers during training
7. Cheaper than personal trainers - but not an adequate substitute
8. "Explainable AI" will be the decisive further development

"I don't believe that trainers can be fundamentally replaced," says Alexander Asteroth. He is one of the authors of the AI study conducted by the German Federal Institute for Sports Science. "Rather, artificial intelligence provides tools to support trainers in their work. In the area of training planning, AI makes predictions about possible performance developments based on training. These can be correct. But they can also be completely wrong."

ChatGPT, for example, became synonymous with artificial intelligence at lightning speed after its release in November 2022. With the help of this, the chatbot answers a wide variety of questions within seconds: "What is HIIT?" or "How do I train for a marathon?" for example. After registering free of charge at https://chat.openai.com/ to conduct human-like conversations.

The chatbot answers questions with remarkable speed and clear, straightforward language. This saves searching through various websites, such as those offered by Google. Even with more precise follow-up questions on the same topic, the hyped AI provides detailed answers, and the information is presented in a clear and structured manner. The system is fed with data from websites, Wikipedia entries and books, so that it sometimes provides quite useful answers. However, the AI has limits and shows weaknesses when it comes to individual, athletic training.

Prof. Dr. Alexander Asteroth explains why AI sometimes gets it wrong when interpreting training data: "Quite strikingly, if you look at data, you could come to the conclusion that more training leads to stronger performance development. According to the motto: a lot helps a lot. Put simply, an AI could therefore generate training plans that are as extensive and intensive as possible. But this could result in excessive training and injuries," warns Asteroth. His conclusion: "You always have to critically question what artificial intelligence suggests. And for that, you need expertise. And this is what - real - trainers usually bring to the table."

Furthermore, Prof. Asteroth notes that AI training plans in cycling, for example, are usually just "modified standard plans" like those found in books on training theory. "This may have to do with safety considerations. Manufacturers are not forthcoming about what exactly their training plans are based on: Is it really 'machine learning' or has it been hard-coded?"

Many high-reach fitness Youtubers are currently showing off machine learning in their video blogs how the chatbot spits out workout plans in no time, based on information such as age, training goal and available time to train. The general consensus is that the results are impressive, but should be treated with caution because these plans do not take into account individual characteristics such as injuries.

For this reason, Matthias Fischer, a sports scientist and personal trainer from Heidelberg, is skeptical about training plans created by AI: "I have a more specific stance on this, because I think artificial intelligence is still in its infancy. I think before you create a training plan for a person, you should look at them as a complete biopsychosocial system. Every human being is too complex to train according to a standardized plan," says the owner of the company CAPECS® sports consulting in Heidelberg, Germany, in an interview with ISPO.com.

"If an athlete wants to build muscle in the thighs, for example, enters his key data such as age, weight, body fat percentage, then an artificial intelligence can certainly already create reasonable training plans. But in practice, as I see in my daily work, you have to examine the entire body and take a medical history, not only for optimal results in muscle building, but especially for pain prevention and rehabilitation. That's where I see the limitations at the moment."

When asked which top athletes are using artificial intelligence to improve their performance, ChatGPT provides concrete answers; for example, that top British track and field athlete Katarina Johnson-Thompson is using an AI-based fitness coaching system developed by the world's leading U.S. provider "Vi." Using data from wearables such as smartwatches and fitness trackers, this system creates personalized training plans for Johnson-Thompson, a medalist. Serena Williams, LeBron James and Usain Bolt also train with AI, according to ChatGPT.

AI coaches are also on the rise for normal athletes, as the success story of Freeletics proves. The fitness app startup founded in Munich in 2013 shows that AI has long since found its way into popular sports. The focus of training at Freeletics is on strength and endurance without equipment: for example, as high-intensity training and calisthenics (intensive physical training in parks, where you train with your own body weight). The virtual trainer is, according to Freeletics, "the most advanced digital, AI-based personal trainer available."

Artificial intelligence is becoming more and more prevalent in sports at all levels of training control, because smart tools are making more and more data and statistics available. In addition, information, experiences, and feedback from users around the world are all flowing into AI-based apps. The availability of valuable data is the reason AI can evolve; the more data, the better the results.

The app "Enduco", for example, first records the training of its users* and integrates performance data from other sources. Taking into account the user's current form and athletic goals, the AI-controlled coach generates a "customized training plan." After the workout, the AI coach adjusts the training plan based on the user's current condition. Based on deviations, this plan is then adjusted accordingly for the future.

Artificial intelligence is even capable of predicting injury risks. Data scientist Alessio Rossi from the University of Pisa and his team

have analyzed Italian professional soccer teams over an extended period of time. They measured each player's training load using parameters such as GPS and video analysis, as well as heart rate, lactate levels and subjective perception of exertion. The researchers fed all of this information to an AI to identify patterns in the load data and ultimately predict injuries. The AI was able to calculate probabilities that a player might get injured in the next few days or weeks and also provide clues as to why injuries might be imminent.

Using a combination of, for example, camera data and information provided by wearables, AI can even solve strategic or tactical situations. This is the conclusion of the report "Artificial Intelligence for Top-Level Sports in the Area of Conflict between Big and Small Data" by the German Federal Institute for Sport Science.

Currently, AI tools are mainly used in the fitness sector. And they can do more than just create general training plans. For example, the app "Mirror" helps its users to perform movements correctly during fitness training. AI provides inexpensive ways to workout plans, helps with workout management, and supports motivation, at a much lower cost than a personal trainer.

Despite this fact, human trainers do not have to worry about their raison d'être, explains Munich-based personal trainer George Tsantalis: "With complicated movements such as squats or deadlifts, an AI cannot intervene directly if they are not performed correctly. The AI can't see the movement and thus can't intervene in real time at any time, as a personal trainer would." In general, though, Tsantalis sees a lot of potential in what AI can do in the long run: "As the technology improves almost every day, it becomes much more competitive and can certainly be more helpful to us personal trainers and the fitness industry in the future."

Prof. Asteroth also relies on human coaches, but also sees the potential of virtual coaches: "Like many others, I'm skeptical about AI-generated training plans. But in the future, as they evolve, artificial intelligence has potential in sports. That's why I'm working and researching it. But right now, I just don't see AI getting there yet."

The scientist does not want to commit to what development potentials can be expected in the coming years, but Prof. Asteroth clearly defines what they should be able to do: "Artificial intelligence must provide explainable training plans. Modern AI approaches are data-driven; the system's machine learning develops its predictions from data. But these remain completely unexplained. Risks, for example, remain open. What happens if I train this way or that way?"

"Explainable AI" will play a very big role in the future, in Asteroth's view. Most modern machine learning methods are black-box algorithms, i.e., the predictions are made without explanations as to why exactly this prediction result occurs. But this, he said, is quite crucial to being able to assess whether or not an AI-generated training plan can be trusted. The AI would not only have to provide a plan, but the rationale for it as well. That is not happening at the moment.

Prof. Asteroth sees the potential of AI primarily in the form of support systems for trainers, whether in competitive sports or in the fitness sector: "Particularly in the professional sector, athletes are already fully trained, and here we may need alternatives for new training models. AIs can provide these. From my point of view, AI should never be a fully automated system, but always a supporting one. Because we are dealing with people!"