Applied Animal Behaviour Science is about understanding animals’ mental, emotional and physical capacities in real life. We met with two animal scientists, Marianne Wondrak and Ariane Veit, to discuss their recent work with pigs (Sus scrofa domesticus) to categorize pictures of human heads [1]. 

The work was motivated by the ever-ongoing quest to discover surprising cognitive abilities in pigs, including animal memory and social cognition. The team of Wondrak and Veit was especially interested in the visual domain of pigs. They were greatly interested in learning how pigs learn about humans, for instance, if they would learn to discriminate different people visually.

To this end, they conducted an experiment with 33 free-ranging pigs that lived for almost two years in close relationships with humans on an Austrian farm. The pigs were trained to discriminate between two views of the heads (front and back) of ten different women presented on a tablet computer screen. Amazingly, they found that 31 subjects reached a high (> 80%) and stable learning criterion, with some pigs having accomplished this in less than half of the trials. 

They also ran a generalization test, in which the pigs were asked to spontaneously discriminate the two female head from a set of 16. The latter test achieved an astonishing result.

As they found, “the average score of 84% correct responses suggests that the pigs had not simply stored the training pictures together with the respective contingency in their memory but have learned the task by open-ended categorization.” 

Wondrak and Veit also did not shy away from challenging the still dominant view about the poor visual-cognitive abilities of pigs, which some of their tests could also confirm. A few challenges are of particular interest in computing, as per Marianne’s experience in field trials:

  • We need to develop new and smarter transponders to gather data 24/7 and in outdoor settings. How to cover 5-6 hectares (10,000 m²) of a pig’s habitat with such devices is an open issue.
  • It would be nice to have means to easier track positive context in pig behavior, e.g., snout-to-snout contact, snout to body and that without external factors, such as availability of food, light (or even music) and with no interactions with people when measuring.
  • Touch screen technology needs improvement, and we need a fundamental understanding of their long-term aim in animal behaviour.
  • In future field trials, new and better ML algorithms are needed for automatic face/body recognition of pigs at the performance level comparable to what can be done with automatic recognition with microchips today. Algorithms needed are especially critical for videos taken outside under different weather and visibility conditions.

As this paper shows, there are exciting challenges in computing with animals, and further more ML algorithms and smart communication transponders. We hope this blog inspires other researchers to delve further into this area, and explore how computers can support the animals around us.

[1] Marianne Wondrak, Elin Conzelmann, Ariane Veit, Ludwig Huber, Pigs (Sus scrofa domesticus) categorize pictures of human heads, Applied Animal Behaviour Science, Volume 205, 2018, Pages 19-27, ISSN 0168-1591, https://doi.org/10.1016/j.applanim.2018.05.009.