Mental Collectives Via Use of the Remembrance Agent (or “Serendipity is simply too necessary to be left to likelihood”)
Thad Starner
Might 1996
Augmented Reminiscence
Computer systems are excellent at storing information and performing repetitive
capabilities in a short time. People, alternatively, might be excellent
at intuitive leaps and recognizing patterns and construction, even when
not actively attending to an occasion. Thus, an interface the place a
wearable pc helps the person keep in mind and entry data appears
worthwhile. Generally, 95% of pc time is devoted to phrase
processing. Since wearable computing permits such ease of entry to a
keyboard, this proportion could also be even larger for such machines. Nevertheless,
phrase processing requires about 1% of the cpu energy of the
system. As a substitute of losing the remaining 99%, an data agent
can search the person’s private textual content database for
data related to the present context. The names and quick
excerpts of the closest matching information might then be displayed. If
the search engine is quick sufficient, a constantly altering listing of
matches may very well be maintained, which might enhance the chance that
a helpful piece of data will probably be recovered. Thus, the agent can
act as a reminiscence support. Even when the person largely ignores the agent, he
will nonetheless are likely to look at it each time there’s a quick break in his
work. In an effort to discover such a piece atmosphere, the Remembrance
Agent was created.
The Remembrance Agent
The advantages of the Remembrance Agent (RA) are many. First, the RA
supplies well timed data. If writing a paper, the RA may recommend
related references. If studying e mail and scheduling
an appointment, the RA could occur to recommend related constraints. If
holding a dialog with a colleague at a convention, the RA may
deliver up related work based mostly on the notes taken. For the reason that RA
“thinks” in a different way that its person, it usually suggests mixtures
that the person would by no means put collectively. Thus, the RA can act as a
fixed “brain-storming” system.
The Remembrance Agent may help with private group. As new
data arrives, the RA, by its nature, suggests information with
related data. Thus, the person will get strategies on the place to
retailer the brand new data, avoiding the frequent phenomenon of a number of
information with related notes (e.g. archives-linux and linux-archives).
The primary trial of the prototype RA revealed many such inconsistencies
on the pattern database and urged a brand new analysis
undertaking by its groupings.
As a person collects a big database of personal information, his RA
turns into an professional on that information base by means of fixed re-training.
A purpose of the RA is to permit co-workers to entry the “public”
parts of this database conveniently with out interrupting the person.
Thus, if a colleague desires to find out about augmented actuality, he merely
sends a message to the person’s Remembrance Agent (e.g.
thad-ra@media.mit.edu). The RA can then return its finest guess at an
acceptable file. Thus, the person is rarely bothered by the question,
by no means has to format his information, and
the colleague feels free to make use of the useful resource (versus knocking on
an workplace door). Data switch could happen similarly.
When an engineer trains his alternative, he can even switch his RA
database of data on the topic in order that his alternative could
regularly obtain the advantage of his expertise even after he has left.
Mental Collectives
Presumably essentially the most hanging use of the Remembrance Agent is its
means to seemlessly share information in a piece group. As a substitute of
merely utilizing one member’s notes, the database is expanded to incorporate
the members of a small work group. This permits private expertise to
be shared rapidly and conveniently. For instance, such an interface
is helpful if one member of the workgroup is in command of repairing the
group’s computer systems. When the group member receives new details about
an obscure bug within the working system, he places it in his private
data information which might then is obtainable to the remainder of the group.
If different members then expertise this bug, the suitable file may
be urged to the member. This helps determine the issue rapidly
and vastly reduces the overhead concerned in diagnosing and assembling
a well-formed query to the remainder of the group, whose acceptable
member could also be unavailable.
To experiment with this concept, three volunteers have been assembled.
These volunteers have pooled their information into one RA database which
runs constantly of their textual content processor (emacs19). Initially, logs
had been to be stored on which of the RA really helpful information was accessed, however
attributable to issues with the software program, a efficiently operating group system
was not obtainable in time for the week lengthy examine deliberate.
Implementation
The present Remembrance Agent makes use of the Savant data retrieval
system developed in-house by the Jan Nelson and Bradley Rhodes.
The Remembrance Agent runs by means of
emacs, a preferred textual content editor. The person interface is programmed in
elisp, and the outcomes are offered as a 3 line buffer on the
backside of the window. A number of concerns have gone into the
design of the RA. First, the RA shouldn’t be distracting except
uncommon circumstances come up. To that finish, the RA doesn’t use
boldface or highlighting and is run at a low precedence. Secondly, if
the RA recovers one thing of curiosity to the person, the total textual content is
accessible with a fast key mixture (e.g. C-c 2 to get the file
related to the second line). The RA is configureable to go looking
in any dimension of context. On this case, the very best ranked end result from a search
on the final 10 phrases (first line) and the very best 2 on the final 100
(second and third line) are displayed. The native context search is
achieved ever 10 seconds whereas the bigger context is searched out of part
each 30 seconds
Determine 1 exhibits the output of the Remembrance Agent. The reference
database for this display shot was Brad Rhodes’s e-mail archives. The
first quantity on every line of the RA output is just a file label for
comfort. For instance, to view message 2, the person would merely
press “Management-2” The second quantity on every line refers back to the
“relevance measure” of the message.
Preliminary Outcomes
A major artifact of this check is that every participant doesn’t
have an equal sized database to share with the others. Thus, it will probably
be anticipated that the one with the most important database can be sited most
by the Remembrance Agent, and such is the case to date. Nevertheless, the
usefulness of the system in on a regular basis e-mail was instantly proven.
Upon receiving an e-mail request for assistance on file io beneath UNIX, the
RA instantly urged a latest serendipitous message from a
completely different newsgroup. One other focal point was that it was usually
not clear whose reminiscences had been being recalled at any given level. This
could also be attributable to working relationships of the check topics however brings up
a viewpoint problem. What occurs when your RA brings up different’s
impressions of the identical occasion? Does this enable for simpler
collaboration? Will different’s viewpoints be by chance accepted as
one’s personal?
Utilizing this technique on a a number of particular person database has revealed
a number of areas of enchancment for the system. Frist, the server nature
of savant have to be improved to permit sooner replace speeds. Higher
descriptor traces must be included to point out of whose reminiscences
a given recall comes. Lastly, a number of enhancements within the emacs person
interface have been recognized. As soon as these are mounted, a bigger check
could also be tried with detailed logging on what data is
referenced and when this data is offered throughout completely different
customers.