Thursday, February 25, 2016

GPS Evidence: GPS location data as an element of predictive technologies.

      This blog has examined and discussed GPS tracking evidence in both legal and practical settings. In recent years, GPS evidence creation has become omnipresent in technological devices. These same devices include information about individuals and their surroundings. That information is stored both for the convenience of the individual and for the profit of others who want to market both goods and services to the owners and users of the technology. At this writing, most of the applications don't intelligently use all the information either available to them or data available to actually predict user actions in ways that will assist the users. This blog is about how GPS evidence help can establish what I would call geocontexts, that is, contexts derived from both current geolocation, past geolocational history, and supporting data.
       Take, for instance, the activity of cell phone predictive texting or voice recognition. Current programs are blind, deaf and dumb to their surroundings. They focus on either a few keystrokes, within the message. Newer programs try to predict what the next word will be based on who the message is addressed to. But all ignore the basic context clues that humans rely on every day to attempt to decipher person-to-person face-to-face communication. Geolocation is one of the major communication clues.
        A face to face conversation in a specific geological location (together with its identifying functional context) give humans a great advantage in predicting how a conversation will proceed. Often, people will complete each others sentences because the overall circumstances make the content of the conversation so clear.
         If two strangers dressed in white jumpsuits were ushered into a stark white windowless room, seated across from each other, then asked to predict what the other might talk about, the answer would probably be “could be anything.” A cell phone app or voice recognition software faces the same issue. However, two people talking at a supermarket probably are talking about food, work or family. Two people talking at a car dealer are probably talking about cars. People at a sporting event are probably talking about sports. These people are aware of what is going on around them, the likely subjects to come up in that environment, and even probably have an idea about how the conversation will go. Certain conversations take place at certain times of day, different at 8 a.m. at the front door of a school than at 1:30 a.m. at a bar.
          In the same light, voice recognition and text predictive software could, if taking advantage of all the contextual sensor information available, more reliably predict possible inputs and give much more accurate choices by narrowing down the likely word choices to ones that make sense in the context. For instance, a program could build lists of possible word choices or phrases usually texted or from car repair shops - “e” choices would include “estimate” “exhaust” etc., These targeted libraries would be much more accurate than choices picked from the entire universe of the English language. Voice recognition could be weighted towards an auto repair specific vocabulary, rather than choosing possible words from an entire universe of vocal sounds.
          Since we are all creatures of habit, our devices could learn that, based on our GPS tracking history, we work out at the gym on Tuesdays at 7 am, so texts or voice recognitions that occur either just before or while we are traveling (hopefully, as a passenger) would skew towards exercise talk. If our devices are aware of our schedule and our location, they might actually be able to suggest messages, or send them for us-if I'm late to the dentist, but on my way (as detected by my device) it sends a message telling them when to expect me.
          Trips at certain times to certain places on a regular basis also suggest context, context that can be used by devices (and created by learning) that can serve to limit vocabulary to your choice of destination. Weekly trips to certain restaurants will probably have talk of menu choices, seating choices, reservation times and possible companions. Specific areas can be gleaned from past voice or text driven conversations.
           Devices some day may actually share where both the sender and the desired recipient are, and use that data to make intelligent decisions about both message context (for accurate text/ voice prediction) and intelligent message handling. For instance, messages about what to get at the grocery store can wait to be delivered until the driver going to the grocery store has stopped; whereas messages about a scheduled sporting event that was canceled due to bad weather should be passed immediately to those on the way of the event, but passengers should be notified prior to attempting to notify the driver. The device would know of the scheduled event; know the direction of travel toward the event. While sensors to determine which phone (I.d.ed by phone number) is in which car seating area (thus designating the driver) have yet to be invented, GPS technology might tell who is driving based on a historical driving pattern.
           Where two distantly located people are talking will likely suggest a conversation-two business contacts, one a supplier and one a customer, are likely to limit the conversation to the overlap of the two businesses. A restaurant owner talking to a paper supplier will likely be talking about toilet paper and paper towels. More advanced systems might even identify locations of whole families, relate their schedules based on GPS history, and suggest conversation topics based on all those locations. A text from a minor child at school mid-day, with subsequent exchanges between working parents at their workplaces suggests illness related discussions, and who will pick up the child from school.

         The entire point of this post is that predictive text, voice, and someday autonomous systems have much to gain from GPS generated context evidence. While the universe of human activity is highly varied and complex, the specific activity of specific human beings is much more narrow, more constrained and more predictable than the species as a whole. When tracked and recorded by GPS evidence, autonomous predictive systems can accurately predict the context and content of text and voice behavior, and therefore offer a more accurate, quicker, and less frustrating user experience.

No comments:

Post a Comment

Comments are always moderated-your comment will not appear unless it is, in the eyes of the author, relevant.