The recent article by Solvi, Al-Khudhairy, and Chittka (2020) titled “Bumble bees display cross-modal object recognition between visual and tactile senses” reminds me of Molyneux’s Problem (or Question), which was first posed by seventeenth-century philosopher William Molyneux (Locke, 1689). The problem is based on a thought experiment in which someone born blind has their sight returned and the question is whether or not they would be able to distinguish different shapes by sight alone. The underlying question is whether years of tactile interaction with shapes provides the appropriate neural representation for visual recognition. The target article did not test blind bumble bees after returning their sight, but instead used a well-known cross-modal testing paradigm borrowed from decades of testing in humans and other animals, but never before in bumble bees.

Bumble bees were trained to find sucrose solution placed in a hole in one of two objects, either a cube or a sphere. Whichever object did not hold the sucrose, held quinine solution. There were two training conditions, lighted and dark. In the lighted condition, bees could see the objects, but were kept from touching the objects by a plastic shield. In the dark condition, bees could not see the objects, but were able to touch the objects.

After training, bees were tested in either lighted or dark conditions in a fully crossed design with the training conditions. The lighted to lighted and dark to dark combinations of training and test conditions assessed whether or not bees could distinguish objects by vision (lighted) and touch (dark) alone. The results were that bees spent more time studying the training-reinforced object during test than expected by chance. Thus, bees were able to recognize objects unimodally with either vision or touch. The other two conditions, lighted to dark and dark to lighted, assessed whether or not bees were able to transfer object information between vision and touch. Again, the results were that bees spent more time studying the training-reinforced object during test than expected by chance. Thus, bumble bees possess the ability to cross-modally recognize objects and therefore possess an appropriate neural representation for cross-modal transfer.

In 2011, Held et al. published a paper with the closest test of Molyneux’s Problem to date. Five people who had been born with visual conditions such as cataracts had treatment of those conditions performed when they were adults, thus restoring their sight. The experimenters were able to test the five people on a cross-modal novel object matching task (similar to what the bees did) within 48 hours of treatment. Patients performed successfully when they were trained with touch and tested with touch, which was expected, because of the lifetime of experience with touching objects. Patients were also successful when they were trained with vision and tested with vision, suggesting that within 48 hours of treatment, the visual system had calibrated enough to perform object recognition. The interesting finding was that patients were unsuccessful when they were trained with touch and tested with vision (cross-modal). The authors did not include the other possible cross-model condition where bees would have trained with vision and been tested with touch. The results suggested that the initial answer to Molyneux’s Problem was “no”, people with restored sight cannot recognize objects by sight that they had previously only known by touch. However, when three of the patients were tested approximately one week after initial testing, it was found that they now performed successfully. It is worthwhile noting that patients had not practiced the specific cross-modal matching task in the intervening time. Thus, in these patients, some natural stimulation of the visual and tactile sensory systems in real-world interactions (for approximately a week) appears necessary for cross-modal transfer between vision and touch to be calibrated and allow for successful task performance.

The results with the patients in Held et al. (2011) is in line with the theory that some neural representations are not sensory-specific, but instead may be organized to process characteristics of objects, such as their shapes. An example is a neuroimaging study in humans using cross-modal priming or habituation (James et al., 2002). Before fMRI scanning, subjects were exposed to two sets of novel objects, one by touching and the other by sight. During scanning, subjects were shown pictures of three sets of objects, the two pre-exposed sets, plus an unexposed set. As is typical in habituation experiments, the pre-exposed sets of objects showed reduced brain activation compared to the unexposed set. The most important finding was that the two pre-exposed sets, exposed by touch and then seen (cross-modal) versus exposed by sight and then seen (unimodal) showed the same level of habituation, specifically in a region of the lateral occipital cortex. This finding has been interpreted as evidence for a shared or fully-integrated neural representation of shape across both the visual and tactile sensory modalities in humans.

As the authors of the target article point out, their results do not speak strongly about the exact nature of the neural representation of objects in bumble bees, beyond the idea that it is sufficient to produce cross-modal transfer of information between vision and touch. One analysis that may shed more light on the nature of the neural representation of objects in bumble bees would be to compare cross-modal and unimodal conditions directly, rather than comparing each of the four conditions to chance performance. A t-test performed on the data provided in the supplementary materials showed that the proportion of cross-modal contact duration (with the rewarding object) was significantly less than the proportion of unimodal contact duration (t(85)=3.11, p=.001). This result suggests that transfer within a sensory modality is more efficient than transfer across sensory modalities. Such a finding is a step toward ruling out a shared neural representation for vision and touch in bumble bees. However, the main conclusion of the target article stands: insects with relatively small brains possess the capability for cross-modal transfer of object shape that is similar, if not the exact same, as found in large-brained animals.