ideambulate

I’d like to peek around the next corner in technology.  Specifically, what technological challenges and solutions will arise following the advent/ascension of the mobile web. Let’s fast forward about five years, give or take some technological optimism.

Assumptions

• Broadband wireless internet is widely accessible in the developed world.
• Mobile audio and visual data capturing devices with day-scale battery and storage ranges are available at prices reasonable for amateur bloggers/lifestreamers/artists.
• Consumer-targeted voice, eye, or gesture based user command interfaces for the above are also around.
• No earth-shattering breakthroughs in computational speed or information processing occur.

I posit that since people will be gathering information about the real world at an incredible rate, the successful parsing and filtering of this information will a primary challenge.  To put it a different way, the challenge will revolve around the generation of classification/contextualization metadata about objects, people, and events.

From how things look now, it will be relatively easy to use video clips of someone’s face to automatically look up their public online identity since faces tend to be highly structured and are well-modeled.  Less “regular” information about complex scenes or objects might be a lot tougher to tease out automatically.

Since I’m not convinced that complex-object-recognition software will develop quickly enough, I’m guessing that we’ll at least temporarily turn to human-assisted classification.  That’s where an Asynchronous Bayesian Classification Database comes in to help bridge the gap.

Chain of Conclusions

• Lifestreamers would naturally generate their own commentary on the things they observe. (i.e. A user says, “Oh look, an elk!”)
• This commentary will end up associated with the observation-data-stream. (i.e. Video feed surrounding the above utterance marked with tag “elk”)
• Lifecast aggregations will accumulate massive stores of observational data linked with human-generated classification metadata. (i.e. a huge online pile of elk-related video, audio, and text)
• A Bayesian scheme that continually refines correlations between observational data and lifecaster commentary could be invaluable for building robust classifications. (i.e. Taking into account that some people can’t tell an elk from a moose)
• These databases could then be used to estimate the expected interpretation of visual/audio/textual information related to a given metadata query, or vice versa.  (i.e. The software is able to tell some other user in the future, “The thing you are looking at is probably a male elk.  Petting it is not advised.”)

Outlining the relevant portions of a picture or voicing extra audio commentary could also go a long way towards minimizing the amount of effort needed to make the system work.  You know, circling the region of the video feed where the elk is standing, or saying “What a nice brown coat of fur,” so the system knows you aren’t talking about the trees also in the picture.

I certainly hope this sort of mini-David-Attenborough exposition will be incorporated into the lifestreaming culture, as it would greatly enrich the world’s stores of computer-parseable real-life data.  Luckily, most lifestreamers are probably quite happy to describe their world to others in an accessible fashion.

In turn, these ABCDs could serve as the building blocks for more sophisticated AI-based recognition and analysis tools to help us filter through the rising tide of information.  Training wheels for an augmented reality system.