Research on food cognition has overlooked that our modern food environment strongly differs from the environment in which our ancestors lived. In fact, in most cultures today, many decisions about food are made during a trip to the grocery store, where we might ask ourselves: Are these canned tomatoes healthy? but certainly not: Are they edible or toxic? In contrast, our ancestors had to navigate their natural environment to Find and Evaluate the beneficial foods in that environment, while Excluding the costs of consuming something harmful, and finally Decide which entities to eat. We call this the FEED problem and present an innovative perspective on food cognition. We argue that a focus on FEED can bridge the gap between different findings in the field of food cognition and shed a new light on the cognitive and brain mechanisms that drive our food behaviors. It has the potential to promote novel intervention strategies, at a time when the rise of food-related health and environmental challenges calls for action.

A bias towards certain kinds of information such as danger, has been observed in a number of domains, for instance learning and attention. Under Error Management Theory (EMT) it has been argued that these biases reflect the costs of making mistakes with this kind of information (Haselton & Nettle, 2006). Based on EMT principles we reasoned that such biases may also exist in generalisation, specifically for information where generalisation mistakes are costly (edibility and toxicity information). Experiment 1 tested this in adults (N = 88), and found that participants generalised toxicity information more, and edibility information less than control information. Experiment 2 tested this in 4- to 5-year-olds (N = 91) and did not find the same pattern, but observed the only effect of information type being in those children scoring high on food neophobia. In Experiment 3 results from Exp. 1 were compared with an additional categorisation condition (N = 30) for the same stimuli, and a different pattern was observed for categorisation compared to generalisation, indicating that participants were not relying primarily on category-based generalisation. Finally, Experiment 4 looked to test whether the findings of Exp. 1 are a function of negativity and positivity effects more broadly, and tested negative and positive information without the same direct costs to an individual (polluting and sustainability information). We found some indication of a broad negativity/positivity effect in generalisation, though this appears less strong than in Exp. 1, indicating a negativity bias effect cannot fully account for these results.