Paradigmatic Considerations for an AI Interface for a Wearable Computer
By ai-depot | March 15, 2003
AI Implications
Basic AI Implications
It is fairly simple to devise basic AI agents for the LC. For example, an AI agent would distinguish data which is time-dependent and should be updated, and update such data automatically. This includes simple environmental data (such as current weather and location); clearly defined and recurring decision-making data (current news, stock prices, movie listings, grocery specials, etc.), and professional information.
It is not much more difficult to devise basic AI agents for the PC. A commonly mentioned example within the wearable computing community features face recognition software. This feature is automatically activated when someone is introduced (presumably triggered by audio cues during ongoing, routine analysis of speech continuously input via a microphone), and remains on stand-by to cue the user with the associated name at any later point. (It may also scan the environment for faces in order to be ready to cue the user on demand, or may routinely present the name whenever the face is recognized… The latter strategy is problematic, as we will discover.)
One can also anticipate an AI which manages (transparently to the user) the interface between the LC and the PC. For example, a to-do list or shopping list within PC storage (managed or generated by a larger complex PIM application) could trigger the PC AI to request that LC AI agent conduct a search for prices, or construct a minimizing (or satisfactory) itinerary to accomplish various errands efficiently. Once the data has been selected and written to the LC storage (perhaps to a file the name of which was dictated by the PC in its earlier request), the PC can retrieve the data and use it within its algorithms to help the user schedule time or itinerary, etc.
Whenever a computer presents data or information on demand of the user, we can regard it as basically non-problematic: The interface design requires only the usual care; distraction caused by an AI’s notifying user of results will not be an issue. It is not initially obvious, but great difficulty arises in designing a satisfactory interface for AI agents for the PC when notification is not on demand of user.
Thinking About Thinking
We do not really understand much about how humans think, but some aspects of cognition, and especially creative or intentional thought, are relatively obvious. Intentional thinking requires, minimally, time; we mean, of course, a span of time without interruption. Let us distinguish between distraction and interruption: Events may be distracting, but we have an ability to concentrate which can minimize the effect of distracting stimuli. I will define an interruption as a distraction which disrupts intentional thinking, but note that frequent (or continuous) distraction often has the cumulative effect of disrupting thought, too.
Let us think about distraction in terms of cell phones first, as these are relatively simple and familiar examples of an interface design. We all know that driving a car and handling a cell phone simultaneously are apparently dangerous. Handling a cell phone represents a distraction which impairs driving ability, to a far greater extent (it seems) than merely conversing with a passenger in the car. A stimuli is more distracting when it is novel, of course, and cell phones are still somewhat of a novelty, so the distraction effect may subside with time. Handling a device physically is also known to be distracting; some states in the U.S. have mandated that cell phones may be used in cars if they are mounted in such a way that the hands remain free.
Sounds and moving visual stimuli also have a very high potential for distracting us, no doubt because they sometimes represented data critical to our survival as we evolved. Sometimes our intentional thinking should be interrupted. Essentially, the primary issue is: To what extent should an AI agent be granted the potential to interrupt a user? In conjunction with this, what is an appropriate level of distraction, and what are the distraction quotients of the possible forms of notification?
There is a second issue: Most of us are familiar with telephony devices capable of storing commonly dialed or critical phone numbers. Perhaps most of us have also noticed that if we store even a very well-known and familiar number, we soon forget that number. Previously we associated a person or institution with a string of numbers, but now we associate that entity with a smaller set of button presses (such as "upper" and "3"). This new association apparently displaces the older, longer string. The brain expends considerable energy, and appears to have a strategy of minimizing energy expenditure, so it is not clear whether this tendency to offload memory processing can be voluntarily controlled.
The question arises: To what extent can an AI agent augment user’s own memory without being self-defeating? Using our example: If users know that facial recognition software will recall the name (possibly supplemented with a summary of previous encounters or their context), will they lose their ability to remember new faces?
This particular question is actually not new. When books began to proliferate, many lamented that no one would remember what would henceforth be stored in written form. (Those who denigrate the ability of an oral culture to transmit details reliably probably do not understand that preliterate humans recall entire conversations word for word, but this is a fact.)
Of course in using books, we recall a general sense of what we read, and when we need more details, we return to the book. But what would happen if everything we read, everything an onboard camera and microphone took in, were stored locally, and could be searched for and recalled instantly? What would our human memories retain of what we read or heard or saw?
Strongly reinforcing this concern, a story recently surfaced about a new disability: a severe, disabling PDA over-dependency. This phenomenon occurred in young persons in several physical locations, including England and Japan. According to one study, about 1 in 10 PC or PDA users between 20 and 35 suffer very severe memory problems.
I am more than 50 years old. When I use a PIM, my noting appointments or errands reinforces my memory for these events, which for me occur in a larger context (managing my time, or picking up groceries for a meal I’m planning — which in turn occur in a yet larger context of maintaining a friendship or creating time for my own "work"). For many younger persons, appointments and errands noted in a PDA (or a computer’s PIM) may be just random flotsam, instantly forgettable (and forgotten).
Younger persons naturally lack a large-scale intellectual framework (or orientation) within which to organize relatively large-scale concepts. Such frameworks come with maturity, if they come at all. To what extent do younger persons today either experience what happens to them as insignificant (not worthy of storing in memory), or experience everything as isolated phenomena?
Many observed have remarked that large numbers of young workers are not at all emotionally invested in their work. Without addressing the separate issues of whether this is appropriate, it certainly contributes to making work-related data forgettable. Emotion is another factor that strongly affects the creation and recall of human memories.
Interface design may contribute to the difficulties experienced by young workers: Newer designs provide no context for entries, whereas my very old (in computer years) PIM design does. Information can be remembered; it is hard to remember mere random data — unless it is data one frequently and actively (whether or not consciously) manipulates. Note that manipulation implies structure and a larger contextual framework.
Perhaps one reason my own PIM (ECCO) helps me remember data, instead of "offloading" information from my memory, is that all data is intrinsically organized, since it functions as an outliner by default. Appointments do appear on a calendar view, but they are also notes "under" the item of an address book view, and may be subitems on a larger-scale "notebook" view. My strategy for using ECCO reinforces this: I normally begin my week by reviewing a “notebook” view (i.e., an organized selection of my data, which while it may be adjusted, persists over time) called "Weekly Plan." After manipulating the various hierarchically organized items in this view, I usually drag specific subitems to an address book (or, conversely drag address book entries under an item), and then drag an address book item to the calendar view (window), either as a to-do list item or as a specifically schedules appointment. (Any four views of notebooks, calendar, and address book, can be seen simultaneously in windows.)
This interface design helps my memory in at least two ways: It provides a larger context for everything I need to do, or might do (or everything I care to enter into my PIM). It also encourages me to manipulate the data, increasing my involvement with data items and making them more familiar. (By the way, I also have a "Daily Plan" notebook view, for those exceptionally busy days.) Admittedly I myself set up these notebook views, but my PIM’s design made it possible, as well as encouraging it by the way it automatically integrates and structures my data.
Some theorists blame the new and extremely debilitating memory problems on information overload, rather than over-dependency on external memory aids. Regardless of which explanation proves better, one thing is certain: What we do not retain in our own memory, we cannot think about.
To solve problems, we need to be able to manipulate multiple data concepts (small or large) which we retain in our own memories over some period of time. That is, researchers of creative thought have known for some time that persons who solve significant problems typically spend very long periods of time obsessed with their particular question, and try to fit all new incoming data to the problem in search of a solution. (I am using "obsessed" in a value-free context here, not meaning any so-called neurosis as defined by clinical psychiatrists.) Einstein, for example, claimed to be obsessed since childhood with the question of what he would experience if he were riding on a beam of light.
An old supporting reference I happen to remember is entitled something like "The Case of the Floppy-Eared Rabbits: Serendipity Gained and Lost" by someone with a name like Barber or Barbour. I remember it because, decades ago, I used this article in teaching a class, and I was very disturbed when I later learned (reviewing their responses to test questions) that almost none of the college students knew what "serendipity" was, and yet they had not bothered to look it up or ask about it during the class discussion of the article. My emotional reaction to this discovery not only influenced me to withdraw eventually from the path that would lead to an academic career, but also tended to fix all the incidental details in my memory. That’s how memory works: We remember what we care about, as well as whatever details we work with frequently.
Perhaps this explains why so many persons regarded as very intelligent and creative also happen to be unfashionably passionate about their own field and many other aspects of the world. Unfortunately, it is hard to imagine an AI interface that could appropriately elicit strong feeling about information it presented. It may be worth noting that I devised my own Daily Plan and Weekly Plan notebook views of my PIM’s data while I was attempting to establish my own startup company, so I was deeply invested in everything I was trying to accomplish.
Tags: none
Category: essay |
