This posting is about ‘the search for missing information.’ In particular, we compare a TED Talk by Eli Pariser on the ‘filter bubble’ to the neuro-engineering work I do on what is called cortically-coupled computer vision (C3Vision). In Mr. Pariser’s talk, he warns of the potential danger of filtering online searches tailored to individual persons’ interests. For instance, he cites an example of searches performed by two of his friends for ‘Egypt’ during the start of the Arab Spring earlier this year.

The results that came up for each friend differed. In particular, one friend’s results were laden with information about the protests before the fall of Mubarak, while the others were not. In other words, Google had filtered results on ‘Egypt’ as appropriate to what their algorithms determine is ‘important information’ for each user. This search is an example of what Mr. Pariser calls the filter bubble. One friend’s bubble centers on protests, while the other’s does not.
But the internet is such a vast web of information that it would be wholly chaotic if we did not have some sort of ‘relevance’ filter. Mr. Pariser seems to agree, but he concludes his talk with a request that the novelty of unexpected information be maintained in searches on the internet. In his words, “We need [the internet] to introduce us to new ideas and new people and different perspectives.” So the real question then is how to fine-tune search algorithms for information that could of ‘interest’ to its users – perhaps even without them knowing it.
This is where the work I do on C3Vision comes into play. C3Vision is a brain-computer interface (BCI) used to search through huge amounts of pictures for objects of interest. Right now, this is developed under the wing of defense-related research. But if we think of Mr. Pariser’s requests for better filtering then we might see a use for C3Vision there as well.
For instance, as a BCI, C3Vision notes neural signatures of stimuli that are of interest to the subject. These neural signatures are covert, i.e., there is no physical response with a hand or arm gesture, as with clicking a mouse. Furthermore, this signature happens in the first 500ms (0.5 seconds) after the stimulus is seen. So not only is the neural response pre-motor (and therefore much faster than mouse-clicking), it is so fast that the subject need not be distracted from his/her current search task when he/she registers neurally that a given stimulus is of interest to him/her.
This is not meant to be an advertisement for C3Vision or the company that develops it, Neuromatters. Rather, it is meant to illustrate that there are ways of improving the search algorithm when it comes to finding novel information. More importantly, this might be a way to refine the search when we don’t even know what information we are searching for. In other words, it could be a way to search for the missing information.