Rob Telson - AI everywhere presentation, page-12

One thing that is not immediately apparent from the demonstrations is that the Akida technology works with multiple types of sensors, so it's not just limited to visual use cases. The majority of the demos that I have seen do the one-shot learning from video sources, most likely because it has the most direct impact on a presentation. Rob does show the auditory example for voice detection, so I've seen at least two types of one-shot learning examples.

Aside from any hardware mimicking the five senses as valid input for Akida, I believe it would also be possible to create additional "simulated" sensors based on metadata resulting from pre-processing a data set. For example, there are algorithms that can analyze text to determine contextual information and mood from a blob of text and, in turn, use that "mood" metadata could be used as a faux sensor.

I think the device could learn to detect spam, negative or positive postings, or even "to do" items for creating scheduled appointments. There are definitely a lot of applications for this technology beyond the demonstrations that the company has presented thus far.