Artificial intelligence continues to underperform in areas where human intuition excels, as demonstrated by its struggles with a seemingly simple task: accurately identifying food. Despite advancements in image recognition, AI-powered nutrition apps consistently miscalculate meals, undermining their promise of effortless macro tracking.
This week, The Verge ‘s Victoria Song tested Ladder’s new AI nutrition feature, hoping for a streamlined way to log her pre-workout breakfast of Kodiak protein waffles, peanut butter, honey, and iced coffee. Instead, the AI inflated the calorie count by over 100% and misidentified ingredients, suggesting she consumed significantly more food than she did. This is not an isolated incident. Apps like Oura, January, and MyFitnessPal all exhibit similar inaccuracies, often mistaking ingredients, misjudging portions, or failing to recognize ethnic dishes.
The fundamental problem lies in the nature of food logging itself. While AI can distinguish between broad categories (banana vs. apple), it falters with specifics (filling inside ravioli, sauce type). Even when corrected, users spend more time fact-checking AI errors than they would by manually entering data. This negates the intended convenience.
More importantly, the reliance on AI distracts from the core issue: sustainable dietary change. AI can suggest adjustments, but it can’t enforce behavioral shifts. Food logging is most effective as a temporary tool for building awareness, not a lifelong dependency. The goal isn’t hitting arbitrary targets, but understanding personal eating patterns and fostering mindful consumption.
The business model of health apps incentivizes perpetual engagement. Rather than empowering users to self-regulate, these tools encourage dependence by implying that success requires constant tracking. The true insight AI could provide isn’t calorie counting, but honest feedback – whether celebrating a nutritious meal or acknowledging a pattern of stress eating.
Currently, AI nutrition features are more frustrating than helpful. Instead of simplifying the process, they create a new layer of annoyance, requiring users to correct nonsensical errors. Until AI can reliably identify food without constant supervision, the best approach remains trusting your own judgment and occasionally indulging in a bag of Cool Ranch Doritos.
