The Mathematician Rat—An Evolutionary Explanation


JG is a rat, a Cricetomys gambianus or Giant Gambian Pouched Rat; she is also a Hero Rat, a landmine detector at Apopo in Tanzania. In December 2009, she performed uncharacteristically badly and puzzled everybody as Hero Rats don’t make mistakes. What was the problem with JG? Had she lost it? Had the trainers made a crucial mistake?

Apopo in Morogoro, Tanzania, trains rats to detect landmines and tuberculosis and the little fellows are very good at what they do. In Mozambique, Apopo has so far cleared 2,063,701 square meters of Confirmed Hazardous Areas, with the destruction of 1866 landmines, 783 explosive remnants of war and 12,817 small arms and ammunitions. As for tuberculosis, up until now the rats have analyzed 97,859 samples, second-time screened 44,934 patients, correctly diagnosed 7,662 samples and discovered 2,299 additional cases that were previously missed by the DOTS centers (Direct Observation of Treatment, Short Course Centers in Tanzania). More than 2,500 patients have since been treated for tuberculosis after having been correctly diagnosed by the rats.

In December 2009, I was working full time at Apopo in Morogoro. I wrote their training manual, trained their rat trainers, supervised the training of the animals and analyzed standard operating procedures. At the time of writing, I still do consultancy work for Apopo and instruct new trainers from time to time. Back then, one of my jobs was to analyze and monitor the rats’ daily performance and that’s when I came across the peculiar and puzzling behavior of JG in the LC3 cage.


LC3 is a cage with 10 sniffing holes in a line and the rats run it 10 times. On average, 21 holes, randomly selected by computer, will contain TNT samples. We train rats in LC3 everyday, recording and statistically analyzing each session. We normally expect the rats to find and indicate the TNT samples with a success rate of 80-85%. Whenever the figures deviate from the expected results, we analyze them and try to pinpoint the problem.

On December 19, we came across a rat in LC3 that did not indicate any positive samples placed from Holes 1 to 6. She only indicated from Holes 7 to 10. In fact, from Hole 1 to 6, Jane Goodall (that’s the rat’s full name) only once bothered to make an indication (which was false, by the way). From Hole 7 to 10, JG indicated 10 times with 9 correct positives, only missing one, but also indicated 11 false positives. Her score was the lowest in LC3 that day and the lowest for any rat for a long time. What was the problem with JG? She seemed fine in all other aspects and seemed to know what she was doing. Why then did she perform so poorly?

Written By: Roger Abrantes
continue to source article at


  1. Why then did she perform so poorly?

    I’m not sure the article really answered this question.   Did this rat “beat the system” by finding a better strategy (for herself) than the one the trainers had intended?   Was she smarter than all the other rats then?  Or just smarter than the trainers?  Is this rat fully trained, and just goofing off, getting the best reward-for-effort ratio?  Or has she simply revealed a weakness in the training programme.

    Somehow this reminds me of the white mice in Douglas Adams’ Hitchhikers Guide to the Galaxy…..

Leave a Reply