Can AI Really Judge What's on Your Plate?

Meta's latest venture, Muse Spark AI, claims it can evaluate the nutritional value of your meals and provide dinner suggestions based on your fridge contents. I decided to put it to the test, starting with my favorite Japanese bento box lunch.

Evaluating Lunch with AI

My lunch, a bento box featuring seared salmon, rice, mixed greens, and raw salmon, was rated by Muse Spark at 7.5 out of 10. It flagged high sodium content, warning me to watch my intake for the rest of the day. This AI model attempts to break down the meal's components and caloric value based on a photo. However, it couldn't precisely determine ingredient weights or oil types, estimating the total calorie count at 760.

While the AI indicated a good balance of Omega-3 and micronutrients, it noted a lack of fiber, vitamin C, and calcium. Despite its effort to analyze my meal, the AI stumbled when trying to generate a labeled image of the meal, producing unintelligible text.

Dinner Suggestions and Limitations

Next, I challenged Muse Spark with a photo of leftovers and miscellaneous ingredients from my fridge. It suggested three options: tomato-braised chicken, spaghetti alle vongole, and Japanese-style oyakodon. These recipes aimed to compensate for the nutritional gaps identified at lunch, such as fiber and vitamin C.

Muse Spark advised rinsing canned tomatoes to reduce sodium and avoiding additional salt. The AI's recommendations were insightful but lacked specificity in recognizing ingredient modifications, like mistaking yogurt-covered strawberries for freeze-dried ones suitable for a smoothie.

The Verdict on AI Guidance

While Muse Spark offers intriguing meal planning assistance, its limitations in ingredient recognition and labeling can't be ignored. Can an AI truly replace a human's nuanced understanding of food? For now, it seems Muse Spark provides inspiration rather than concrete solutions. I ultimately chose the oyakodon for dinner, storing the other suggestions for another day.

Despite the AI's dietary caution, my love for soy sauce remains unchanged. The real bottleneck isn't the AI's model. it's the infrastructure supporting its learning about diverse cuisines and meal preferences.