AI nutrition tracking stinks | Edge

This is it Mohsena weekly newsletter sent every Friday from one of the Verge’s top reviewers Victoria song Which analyzes and discusses the latest phones, smartwatches, apps and other gadgets that you swear will change your life. Mohsen It arrives in our subscribers’ inboxes at 10 a.m. ET. Subscribe to Mohsen here.

once again, Artificial intelligence fails to achieve this In some of its promises.

Before my last long trip, I made my usual pre-workout breakfast. Two pieces of dark chocolate Kodiak protein waffles, a tablespoon of peanut butter, and a little honey. On the side is a modest cup of iced coffee with a little soy milk.

I write a newsletter called Mohsen. Admittedly, I dabbled in counting macros — the practice of tracking your protein, fat, and carbohydrate intake — to see if it helped my training. Of course, I spent five training units discovering that this breakfast gives my body the roughly 355 calories, 16 grams of protein, 28 grams of carbs, and 17 grams of fat it needs to feel good during my morning run. and Don’t fall asleep at my desk yet. The annoying thing is having to re-enter the same information into any training or food logging app.

I’m told that artificial intelligence will change that. Recently, the introduction of Ladder, my go-to app for strength training AI-powered feeding features Which promised to make counting macros easy. All I had to do was take a photo, and the AI ​​would do the rest. So imagine how I felt when Ladder AI told me that my carefully prepared breakfast was 780 calories, 20 grams of protein, 92 grams of carbs, and 39 grams of fat. How, when specifically edited to include exact brands and quantities, this resulted in another, equally wrong number.

This, my friends, is exactly why I did no Count calories or macros anymore.

This is an undeniable fact: logging is drilling.

Traditionally, these scoring apps let you search for food options ranging from frozen dinners to raw ingredients. Some even let you scan barcodes. This is simple enough if all you eat are packaged or whole foods. It all starts to fall apart when eating at restaurants or, ironically, cooking at home. Restaurants that post calorie counts often don’t provide accurate details. Although you can import ingredients from online recipes, this is of little help to experienced home cooks who are improvising dinners throughout the week or replacing ingredients on the fly. To get the most “accurate” and efficient records, you need to measure everything you eat, avoid eating out, and eat the same things every day.

It’s bad because studies consistently show that retention Food diary Or using Digital health tracking tools It is associated with greater success in losing or maintaining weight and gaining muscle. That’s why we’re starting to see health and fitness apps turning to AI to make this process less tedious.
There are endless options.

When Oura introduced its chatbot Oura Advisor, it also added the ability to write a description or take a photo of your meals. Once you do this, your macros will be detailed, whether they are highly processed, and how they may impact your overall health. If you use a Dexcom continuous glucose monitor, you can import that data into the Oura app and use it to compare specific meals with glucose spikes.

Likewise, the January app lets you take photos of meals and, based on your demographic data, creates an estimate of how this is likely to affect your glucose levels. MyFitnessPal has also added a ScanMeal feature that lets you take photos to get calorie and total estimates. My TikTok feed keeps advertising a gamified food tracking app Raccoon AI. You can take photos of your raccoon “feeding” while the AI ​​analyzes and records your meal. In addition to photos, Ladder’s AI feature also allows you to dictate or write text descriptions of your meals.

Methods vary, but the premise is: take a photo and let AI do the rest.

Unfortunately, AI is only good at identifying foods based on images. Oura Advisor has routinely mistook my matcha protein shakes for green smoothies. January was able to recognize that I was eating chicken, but he mistook the barbecue sauce for teriyaki sauce and failed to acknowledge that there were mushrooms in the dish. When Ladder’s AI made breakfast, it was estimated that I ate two 7-inch waffles instead of 4-inch protein pancakes, 2 tablespoons of peanut butter instead of one, 2 teaspoons of syrup instead of 1/4 teaspoon of honey, and creamer and Sugar in my coffee. (I never have sugar in my coffee, thank you very much.)

None of these AI features were able to identify when you’ve made health trade-offs. Instead of white rice, I often mix a cup of edamame and quinoa with brown rice for a denser carb. Oura’s AI classified my mixture as mashed potatoes and white rice. Ethnic foods are also crap. Ladder’s AI has scored Curry Dal Makhani with Basmati Rice and Peas as a Chicken Soup. Sometimes the AI ​​correctly identifies Korean rice cakes in gochujang spicy sauce. Other times, I’ve had rigatoni in tomato sauce.

That’s not you I cannot Edit these AI-generated entries. You can. It’s just that this defeats the very purpose of simplifying a tedious process. Instead, it replaces one inconvenience with another. Whatever time you save searching for entries to log in is now spent editing and validating AI errors.

After thinking about it, maybe simplifying food logging is the wrong problem to solve.

For starters, AI can recognize objects in images at scale, however It’s often a nonsense of detail. He can tell the difference between a banana and an apple, but he will never be able to tell what filling is inside the ravioli. It is also not the best at estimating ratios. If you care about accuracy, you’ll always need to care about it. But what is even more frustrating is that applying AI in this way does not address the root problem. Dietary changes are not difficult due to lack of knowledge. We all know the basics. The difficult thing is to apply that knowledge in your life sustainably. It is reprogramming your emotions and behavior. AI can suggest changes, but you will always be the one who has to make them happen.

The goal of food logging isn’t really about hitting a random calorie goal or overall goal. It’s building awareness about what you eat: to know your eating patterns, what can be improved, and practicing mindfulness when you indulge in a bag of Cool Ranch Doritos. Once you get the hang of it, you quit. You may start over temporarily when goals or health circumstances change — but that’s not something most people have to do for the rest of their lives. Ideally, you should stop recording food because you trust your sense of what you eat and when you eat it.

The problem is that app makers never want you to quit.

A “successful” food logging app is one that keeps you engaged forever. Instead of attributing your success to your hard-earned knowledge, you give credit to the tool. You start thinking, well, if it were me no Track everything, all the time, I’ll be back to how I was before. Or, if you are struggling, perhaps the proposition is that if AI makes difficult things easier, then perhaps achieving your goals will also be easier. (Spoiler: It won’t happen.)

In fairness, there is Something The idea of ​​taking a photo of your food and having AI tell you a useful insight. I really don’t know what this insight is. Maybe it would be enough if the AI ​​told me that my home-cooked meal is a nutritional masterpiece. Or that I’ve had a 15 percent increase in eating glazed donuts in the past 30 days — maybe it’s time to think about what triggers stress eating. Or, “Hey girl, you’ve been eating some pretty impressive food, but culinary wise sadNumber of grilled chicken breasts. Treat yourself to white rice.

All I know is that an AI shouldn’t ask me to take a photo of my breakfast and then waste the next 15 minutes bullying it into correctly identifying what I ate.