Unsupervised strategies for naive animals: New model of adaptive decision making inspired by baby chicks, turtles and insects

Innate biases vary in flexibility and strength. Fixed-action patterns usually have low flexibility and high strength, while early predispositions are weak and plastic. The figure illustrates some innate biases described in the references indicated with specific numbers. Cues position along this spectrum should be interpreted qualitatively. The exact position on the horizontal axis is purely indicative, as predispositions depend on species and context, as shown in this article.

The new model proposed by the researchers suggest that naïve animals like newborn turtles and chicks are not blank slates but are supported by the presence of multiple biases that interact. Researchers found that early biases are surprisingly widespread among newborns of various species. However, these choice biases are not robust and consistent, but instead, weak and transient. They noticed that young animals can combine many weak inbuilt biases to make better choices, such to locate their mother, or target flowers, for example biases about sound, movement, and colour, without the need for learning. The paper also gives a mathematical model to develop and test ideas on enhanced decision making that do not require any prior learning.  

This ‘unsupervised strategy’ has implications for developmental psychology and artificial intelligence where decision-making with sparse evidence is a crucial challenge. 

For instance, the co-occurrence of reddish colour, upward movement dynamics, and changes in speed is a much better predictor of the presence of the mother hen, than each cue considered in isolation.  When two or more preferred characteristics co-occur despite being encoded in different modalities, such as face-like pattern and cluck sounds– the convergence provides strong evidence the stimulus is relevant, because the random co-occurrence of these different stimuli would be extremely rare. 

This unsupervised strategy can help not only inexperienced animals, but also artificial intelligence, as it reduces the need for training. It is surprising how much intelligence is present in animals at the beginning of life that can not only shed light on adaptive behaviour but also inspire the development of new technologies.  

Elisabetta Versace, Senior Lecturer in Psychology at Queen Mary University of London said “Reviewing the literature on inexperienced animals helped to identify an underlying reason for why animals exhibit "soft” preferences.’ At first it is counterintuitive that evolved preference might be weak and transient, but it's this feature that allows them to reduce false alarms in decision making, and to benefit from the multiple cues available in the environment.” 

Benjamin L. de Bivort Professor of Organismic and Evolutionary Biology at Harvard University said: “Modelling this phenomenon has shed new light on experiments conducted over the last 100 years, which can now be better understood as partial clues helping organisms navigate early life.” 

As the model has clear predictions, it will be interesting to test these both on animals and on artificial cognitive systems.   

Cite this article: Versace E, de Bivort BL. 2026 Multiple weak biases support adaptive choices without prior experience: a self-supervised strategy. Proc. R. Soc. B 293: 20251878. 

DOI: https://doi.org/10.1098/rspb.2025.1878