TITLE PAGE


Author:  Laurence Press

Title:   Press, L., Systems for Finding People
	 Journal of Organizational Computing, 2(3&4), 303-314 (1992).

Affiliation:

         Professor, Computer Information Systems, California State
         University at Dominguez Hills

Email:   lpress@venera.isi.edu

Running head:  Finding People

Keywords:

         organizational culture, participation, collaboration,
         computer-supported cooperative work


                               ABSTRACT

Finding a previously unknown person with the skills and knowledge to

answer a question or perhaps to collaborate with is an effective use

of a computer-mediated communication (CMC) system.  This article

discusses two aspects of systems for finding people, system

architectures and organizational implications.


The architectures considered are special interest groups, centralized

servers, and decentralized systems.  The organizational implications

are the value to organizations of people-finding systems, management

incentives for individuals to participate in them, and participation

in the absence of apparent incentives.


The deployment and improvement of people-finding and other CMC systems

will bestow a marginal advantage upon cooperative individuals and

organizations with cooperative cultures.  As a result, CMC may

marginally alter organizational and human nature, nudging us in the

direction of a time when nice guys finish first.


.pa
Systems For Finding People


How do we use computer networks?  Early proposals for networks

envisioned remote login as the key application (Marill & Roberts,

1966).  As early networks became reality, electronic mail, file

transfer, and access to remote databases were added to the list of

applications (Roberts & Wessler, 1970).


Information retrieval has turned out to be of great value, and today

many use computers to retrieve structured and unstructured information

from local and public databases.  Cuadra (1989) lists over 2,000 on-

line databases, and MIS departments often open data files to members

of their organizations.  While such structured information is of

obvious value, it is used primarily for operational tasks like order

processing and billing or for relatively specialized tasks like

literature or legal-precedent search.


Unstructured data may contain valuable information, but locating

relevant items is difficult.  Visions of the future such as the

"personalized newspaper" at the MIT Media Lab, Brand (1988) promise

that artificially-intelligent agents will one day find relevant

unstructured information for us, and Malone, Grant, Turbak, Brobst,

and Cohen (1987) have experimented with a system in which simple,

rule-based agents search for relevant semistructured messages

(containing fixed fields and text), but such systems are not yet in

general use.


While information retrieval and access to remote computation are of

value, electronic mail, or more generally computer-mediated

communication (CMC) has turned out to be the most common application

to date.  CMC is now accomplished over a worldwide matrix of connected

computer networks and conferencing systems (Quarterman, 1990).


We can consider two types of CMC transaction: mail messages for a

specific person and queries in which you are trying to meet a

previously unknown person with skills or knowledge you need.  In the

former case, you may or may not know the email address of the person

you wish to communicate with.  If you do not, the Internet has several

systems for finding addresses.  If the person's name and host are

known, the Finger protocol (Harrenstien, 1977) or a message to the

system postmaster may be used.  If not, "white pages" services such as

Whois (Harrenstien, 1985), Knowbot (Droms, 1990), and NetFind

(Schwartz & Tsirigotis, 1991) may help.


The focus of this article is the second case, where you do not know in

advance who you are looking for.  In my own experience, CMC enables me

to "meet" someone new who helps me with my work at least once a week.

(I know of a married couple who met on a bulletin board system and

understand that Compuserve has a regular listing of "cupcakes" --

couples that met on Compuserve).  Therefore, I am more interested in

an agent or system which can help me find relevant people than one

which retrieves addresses or information.


This article discusses two aspects of systems for finding people,

system architectures and organizational implications.  The next

section presents three alternative architectures:  special interest

groups and centralized and decentralized systems.  That is followed by

a discussion of organizational implications.


SYSTEM ARCHITECTURES


We can think of the problem of finding a person who can provide

information or collaboration as follows:  the information seeker (or

an agent) formulates and submits a query, hoping there are system

members capable of providing a useful reply.  Ideally, the system

would locate each member who could provide unique information

regarding the query, and no others.


Three possible approaches to this problem are:  posting a query to a

subset of the system members who voluntarily participate in a special

interest group (SIG) such as a teleconference, mail list or USENET

News group, a centralized person server which matches a query against

profiles of system members, and a decentralized system in which system

member's agents screen queries to determine their relevance.


Special Interest Groups


The first of these, posting a query to a SIG, is common today.  System

members identify themselves as being likely to have information on a

topic by joining the SIG.  When a user posts a query, it is broadcast

to all members of the SIG, each of whom decides how to respond to it.

While widely used, SIG systems pose some problems.


When a query is submitted, system resources are used in making each

member of the SIG aware of it.  Next, each SIG member spends time

screening the query.  The screening time might be as little as reading

a header line and deciding not to read further, or it might mean

reading the entire query.  Some subset of the SIG members then respond

to the query, which again takes their time and system resources.


In the ideal case, there would be only one response, and it would be a

complete, exact answer to the query; however, that is seldom achieved.

There may be no response, which is appropriate if no system member is

able to answer the query, but inappropriate if there is a member who

could answer, but does not see the query or does not read it.  There

may also be multiple, redundant responses.  I recently posted a query

to a news group asking for source code for a data compression

algorithm, and received essentially the same answer from 8 people.

For seven of these, the time they took to formulate replies, the

system resources needed to deliver them, and the time I spent reading

them were wasted.


The situation is analogous to type I and II errors in statistical

hypothesis testing -- the system can err by not finding a person who

could answer the query or by finding several people who provide

redundant information, wasting time and resources.  Of course, queries

with a single correct answer like "what is size of the national debt"

or "who is the purchasing manager at the XYZ Company" are not typical.

In general, queries will have more complex, open-ended answers, so

replies fall on a continuum from completely relevant and unique to

partially redundant to fully redundant.  In the case of a single-

answer query, there can only be one useful reply; in the case of more

complex queries, the relevance of a reply depends upon both it's

content and the time at which it is received.


Centralized Person Servers


The second approach to the problem of finding a person to answer a

query is a centralized person server (see Figure 1).  A query is sent


              ----- insert figure 1 about here ----


to the server, which identifies system members likely to have

information or skills relevant to the query.  Let us consider several

architectural alternatives.


The server could either be designed to automatically pass the query on

to the member(s) exceeding a system or query-specific relevance

threshold, or it could pass their descriptions back to the requester,

along with an explanation of why they were selected.


In the latter case, the user could consider the selections, optionally

request more information on the suggested correspondents or perform

further queries, and finally contact people at his or her discretion,

cutting the time spent on irrelevant queries and replies, but

requiring more of the requester's time.  Given the current state of

the art of procedures for determining relevance, and the requester's

knowledge of the request, I would expect this approach to be superior

to one which fully automated the matching of queries and people.  In

nearly all cases, a centralized server would be more efficient than a

SIG; however, it could be more prone to "type I" errors, in which

relevant system members do not see the query, depending on the design

of both systems and the behavior of system members.


The server must have access to profiles of the interests, skills and

experience of the system members as well as a procedure for

determining the likely relevance of a member to a given query.  Note

that the profiles need not necessarily be stored on the same machine

as processes the queries.  A network enables resource discovery

(Schwartz, 1991), in which one machine automatically searches or

queries others.  Examples of systems using this architecture are the

Knowbot and NetFind, mentioned above, and Archie (Deutsch, Emtage, &

Heelan, 1992) which compiles a list of files available for anonymous

access using the Internet File Transfer Protocol.


A profile could be as simple as a set of keywords, or it could include

semistructured documents similar to resumes or bibliographies, or it

could include the full text of documents written by the members.  The

design of profiles would be a function of the application context.  In

a large public network, only keyword identifiers with a short

biography might be feasible.  The keywords would be searched for

matches with the query, and the user would scrutinize the biographies

in deciding who to contact.  (At one time IBM had a system for the

selective dissemination of information (SDI), in which members

automatically received copies of magazine and journal articles as a

function of self-defined keyword lists).  If the system were serving a

single organization, it might be feasible to utilize structured and

semistructured information in evaluating query relevance, for example,

locating "any person with contacts in the XYZ company."


A wide range of options exists for search procedures and query

formats.  A simple system could be built using current document

retrieval software with boolean keyword queries matched against

keyword profiles or fields in a structured or semi-structured profile.

A more ambitious system might search an inverted file of terms used in

free-format profiles or of documents prepared by members.

.pa
Simple search software would merely decide that there is or is not a

match between each query and system member, while other software might

rank the relevance of each member relative to a query.  In the former

case, the search could continue until a specified number of matches

were found, in the latter case, it would find all members exceeding a

threshold of relevance or it would rank them all.  Free-format

queries, for example finding people who have written documents similar

to others or people with skill similar to others, can be processed by

large computers.  For example, Stein (1991) describes wide-area

information servers (WAIS) which handle such queries on Connection

Machines and Dumais, Furnas, Landauer, Deerwester, and Harshman (1988)

describe latent semantic analysis, a computationally intensive metric

for distance between documents.


Clients and servers may be decoupled by adhering to query protocols

which are emerging as standards.  Three candidates are the WAIS

protocol, SFQL, and CD-RDx.  The WAIS protocol extends the NISO Z39.50

protocol developed for library catalog queries, and it is being used

and developed by an active group of Internet-based researchers (Lynch,

1991; anonymous, 1990).  SFQL (structured Full-Text Query Language)

extends SQL, and it grew out of the need to query heterogeneous

aircraft documentation distributed on CD-ROM (Shapiro, Diamantopoulos

& Cotton, 1991).  It is being developed at the request of the Air

Transport Association.  CD-RDx is being developed at the request of

the Information Handling Committee of the CIA (anonymous, 1991).

Their goal is to enable government agencies to share data.


Existing information retrieval and database management software can be

adapted to the task of finding people, and there have been a few

trials.  The WAIS approach was tested for finding consultants with

specified expertise at the KPMG Peat Marwick consulting firm (Kahle &

Medlar, 1991).  Streeter and Lochbaum (1988) describe the application

of latent semantic analysis in a similar system at Bellcore.  The

California State University System is experimenting with a system for

locating consultants from among faculty members, and Cartermill's BEST

North America and BEST Great Britain are commercial databases

including information on faculty and graduate students at 106 North

American and all British universities (S. B. Hesson, personal

communication, January, 1992).


A more flexible, object-oriented approach would also be possible.  In

this case, each profile would be encapsulated with methods for

assessing its own relevance relative to a query.  An object-oriented

approach would allow flexibility in the definition of profiles since

heterogeneous descriptions could be accommodated.  An object-oriented

approach would also lend itself to parallel processing, with multiple

profiles evaluating their relevance to a query simultaneously.  New

software and techniques would have to be developed for an object-

oriented system, and it would probably not be possible (or necessarily

desirable) to devise a single metric of likely relevance for all

classes of profile.


A Decentralized Approach


The object-oriented approach to a central server brings us close to

the third alternative, a decentralized system in which member's

filters or agents examine queries to determine whether or not to make

them known to their owner.  This can be thought of as an extension of

the centralized, object-oriented architecture with the methods for

determining relevance executing in a process (probably running on a

personal workstation) associated with each member.  It is different;

however, in that the decision to present or not present a query to a

given member is made locally, without knowledge of the relative

relevance of the query to other members.  It also requires that

communication resources be used in presenting a query to each agent

and there is the likelihood of "type II" errors, with several

redundant replies to a query.  On the other hand, the agent screening

queries will run in the background, using otherwise idle resources.


A version of a system such as this could probably be implemented using

an experimental system like the Information Lens (Malone et al. 1987).

With Information Lens, semistructured messages may be addressed to

"anyone."  Each system member defines rule-based agents which screen

these messages.  This mechanism can be used to screen messages and to

classify those which are accepted for reading.  The rules are written

by the user, and they operate on information in the fixed-field

portion of a semi-structured message.  Since the message formats are

extensible and users write their own rules, it should be possible for

users to define rules which incorporated self descriptions.  Queries

sent to "anyone," would be checked by the potential recipient's agent

using the self-description rules.  The agent would screen out queries

which its user would probably not be able to answer.  It might also be

possible to implement rules which categorized queries as to the

likelihood with which they could be answered.  Whether with

Information Lens or another system, it should be possible to implement

query screening agents.  As in a centralized system, an object-

oriented approach could facilitate heterogeneous self-descriptions and

screening methods.


Of the three approaches to finding people who can answer a query, only

SIG systems are in wide use today, and in spite of their costs and

tendency to make errors, they are valuable.  There have also been a

few experiments with centralized systems, and the technology is

available to deploy others.  Assuming such systems could be built

brings us to organizational issues.  The following section discusses

the value to organizations of people-finding systems, incentives for

individuals to participate in them, and participation in the absence

of apparent incentives.


ORGANIZATIONAL IMPLICATIONS


An organization often benefits from cooperative, internal CMC as

outlined above.  In every transaction someone "gives away" information

and someone else "gets" it.  The organization is enriched because both

members have the information, and they know of each other.  If person

A posts a query, and person B is able to answer it completely and does

so, the organization will be better off if the value of the answer to

A is greater than the cost of supplying it.  If the query/response

system were perfect, and only B saw and replied to the query, then the

"profit" to the organization would be the value of the information to

A less the communication cost and the cost of B preparing the reply.


Of course the system will not be perfect.  As we saw above, there are

communication costs and, more important, the system may cause several

people to read the query, and several people may prepare and submit

redundant answers, which will have to be read.  To the extent that

this happens, the system is inefficient.  Such occurrences are

analogous to friction in a mechanical system.


In spite of of their relatively high "friction," SIG systems are

proliferating and many people find them worthwhile.  As SIG systems

improve and person servers are implemented, efficiency will increase,

making CMC still more valuable for the organization.  Other new

communication technology and infrastructure will have the same effect.

As the cost of cooperation decreases, organizations with a high

propensity to cooperate will enjoy an increasing competitive

advantage, thus economic selection will tend to favor the survival of

organizations which develop an ethic or value system favoring

cooperation.


While organizations will benefit from increased cooperation, what

about the individuals in them?  Why should person B reply to person

A's query?  Does person B even have an incentive to make his or her

qualifications known to the system -- won't doing so mean being

bothered to read queries and taking time to answer them?  Person A

might even be a competitor of person B.  Perhaps misinformation would

be in B's best interest.  The situation is analogous to the "tragedy

of the commons" in which it "pays" for an individual to abuse or over

use common resources like grazing land, lakes, or the atmosphere

(Hardin, 1968).


Incentives to Participate


In the face of possible disincentives for individuals to participate,

management has several options.  Profit sharing and other mechanisms

which reward group rather than individual performance increase the

value of participation to individuals.  Rewarding group performance

has a secondary effect of increasing the emphasis on cooperation in

the general organizational culture, further increasing the propensity

to participate.  This indirect effect can be amplified by using hiring

practice, training, executive communication, and other techniques to

increase the general value placed upon cooperation in the culture of

the organization (Schein, 1984).


Management also has the prerogative to exercise its authority to

mandate participation, making it part of an employee's job and

monitoring compliance.  The effectiveness of group rewards and

attempts to produce a cooperative culture will be strongest in

difficult economic times, when the organization is threatened and

individuals have relatively few alternatives.


Changing organization structure, for example, by flattening it, may

also have some influence on participation.  The effect of organization

structure (the topology of communication channels) on propensity to

communicate has been investigated in small group experiments by

Bonacich and Schneider (personal communication, January, 1990).  It is

also possible to create economic markets within and among

organizations, by paying royalties or consulting fees.  Perhaps the

first suggestion along these lines was Nelson (1973), which proposed

the Xanadu Network, a hypertext publishing and conferencing system

supported by a system of copyright and royalty payments.  Potential

consulting fees are the incentive to participate in the Cartermill

systems mentioned above.


Participation without Incentives


While management actions such as these may be needed to overcome a

reluctance to participate, there are also many examples of seemingly

irrational cooperation or altruism.  This section examines possible

causes of this behavior.


One framework for explaining it is offered by Maslow (1954) who

postulates an hierarchy of needs: physiological, safety, social,

esteem, and self-actualization.  A person may be preoccupied by a

lower level need, but as it becomes satisfied, its importance

diminishes as the next level moves into focus.  People with

unsatisfied physiological or safety (security) needs, would be less

likely to take the time to respond to a query than those who were

focused on higher needs.  If you are insecure in your job, you might

not be willing to spend time helping a coworker without explicit

credit or compensation.


Higher level needs, for example for esteem, explain some apparently

uneconomic contributions.  As any aspiring poet can tell you,

unpublished information does not yield esteem.  Voluntarily submitting

a description of your skills and experiences to a person server would

afford an opportunity to achieve esteem as would answering queries.

Frequent contributors to public SIG discussions, such as those on

USENET, gain esteem by becoming known among to SIG members.


From the point of view of someone posting a query, anonymity might be

desired in order to prevent erosion of social status.  For example,

Karabenick (1987) has found evidence that in an academic setting,

people are more likely to request help if they may do so anonymously,

presumably in order to hide their ignorance, and Stodolsky (1989)

suggests that mechanisms to protect expression, for example anonymity

or psuedonymity, where each member has one, and only one pseudonym,

might encourage information sharing.


However, information contributions which bestow neither economic

advantage nor social status are common on today's SIG systems.  These

might be explained as a recognition of long-run economic calculations.

For example, Kenner (1989) recounts an anecdote in which someone on

the BIX network helps him with a problem and he passes the favor along

by helping someone else.  He explains this altruism as follows:


   [Members of the BIX Network] routinely exchange services for
   free.  It's the regnant assumption that, whether I can ever be
   of service to [the person who is assisting me], I can surely
   be of service some other time to somebody, and these pooled
   favors level out.

But why doesn't the tragedy of the commons stop one from contributing?

When Kenner sees a query which he can answer, it is still locally

irrational for him to do so.  While some may use Kenner's

rationalization, others may find themselves contributing without being

able to explain why -- it just feels right.


The evolution of apparently irrational altruism has been studied in

depth by biologists, who may offer some clues.  We often have

difficulty giving rational, symbolic explanations for skills and

concepts which are critical to survival.  Perhaps this is because

surviving members of our species learned these concepts in a pre-

verbal era.  For example, everyone can differentiate between living

from non-living entities, yet "life" is difficult or impossible to

define.  Similarly, to some people, at some times, cooperation, taking

the time to answer a query on BIX, might just "feel right."  Lovelock

(1979) states that what feels right may be genetically determined:


     It may be that we are also programmed to recognize
     instinctively our optimal role in relation to other forms of
     life around us.  When we act according to this instinct in
     our dealings with our partners in Gaia, we are rewarded by
     finding that what seems right also looks good and arouses
     those pleasurable feelings which comprise our sense of
     beauty.

Biology offers theories which might explain the evolution of

altruistic feelings and hence behavior.  One is the theory of kin

selection, in which "a gene which causes its carrier to perform an act

which puts it at risk may nevertheless increase in frequency if the

result of that act is to help relatives of the actor, who may carry

identical copies of the gene" (Maynard-Smith, 1982).  For example,

animals are known to give warning calls when a predator is near, at

their own peril.  Sherman (1977) has studied this phenomenon in ground

squirrels, and after testing six alternative hypotheses to explain it,

concludes that of the six, the warning of kin is the most important

and perhaps only function of these calls.  Could it be that the

feeling which leads to cooperative behavior among individuals within

an organization is a case of misguided kin selection, evoked by

contact in the workplace and common organizational symbols and

culture?


There are also evolutionary theories explaining reciprocal altruism

among unrelated individuals.  Risky, cooperative behavior has been

observed in unrelated animals, for example, by Packer (1977), and

modeled using game theory in Axelrod and Hamilton (1981) and Axelrod

(1984).


Axelrod and Hamilton analyzed the Prisoner's Dilemma game and showed

that a strategy which involves some cooperation (instead of always

defecting) is evolutionarily stable if the game is repeated many

times, the players recognize each other, and the probability that they

will play each other in a subsequent game is sufficiently high, given

the payoff matrix.  One such strategy is tit-for-tat (TFT), in which

the player always cooperates on his or her first encounter with an

opponent, and thereafter does whatever the opponent did on the

previous encounter, i. e., begin cooperatively, but immediately

retaliate for a defection and reward a return to the cooperative fold.

Axelrod invited game-theory experts and amateurs from around the world

to participate in a computer-simulation contest for such strategies,

and TFT won, even though the other strategies were computationally

more complex. (The TFT algorithm required only 5 lines of code while

the 62 other entries ranged from 6 to 152 lines).


Axelrod and Hamilton speculate that uncertainty as to whether others

were relatives (kin selection) could get the cooperative ball rolling.

Furthermore, they show that once TFT is established, it will resist

intrusion by an all-defection strategy.  Gregory (1982) notes a TFT-

like pattern in the behavior of people in clan-based societies with

gift-exchange economies.  Gift exchange establishes a relationship

between the participants, with a gift establishing a debt that has to

be repaid.


TFT did not defeat any competing strategy, but its total score was

highest.  Axelrod states that "TFT won the tournament, not by beating

the other player, but by eliciting behavior from the other player

which allowed both to do well," and goes on to give examples from

every day life (business, arms control, etc.) where such behavior

works to everyone's benefit.


Simon (1991) also offers an evolutionary theory of motivation to join

organizations and work without shirking.  As partial explanation for

such cooperative, often altruistic behavior, he speculates on the

evolution of a tendency to balance self-assertiveness with docility,

where being docile is "to be tractable, manageable, and above all

teachable."  Simon reasons that a degree of docility would on the

average increase fitness, and hence be selected for.


If there is some instinctively felt propensity to cooperate, we would

expect that propensity to be stronger in some individuals than others.

We would also expect it to be felt more strongly at some times of life

than others, due both to circumstance and maturation.  Achieving

material success could pave the way for altruism, and it may also come

more easily at certain stages in life, just as sexual maturation or

the ability to acquire language have a well defined place in the life

cycle.  (Also, as pointed out by Axelrod and Hamilton, older

individuals would have lower probability of encountering others in the

future, shifting their game-theoretic assessment).


Lovelock suggested that we may be genetically programmed to recognize

beauty, and we may also be genetically programmed, to varying degrees,

to cooperate.  Helping someone who posts a query on a computer network

might feel good because that feeling tends to propagate your genes.
.pa
CONCLUSION

CMC infrastructure is proliferating rapidly, lowering the cost of

communication and cooperation.  One way CMC systems are used is to

find previously unknown people who can provide information or

expertise.  We discussed three general architectural alternatives for

building such systems.  Of these, SIGs are already in wide use, and

the technology is available to deploy systems using centralized

servers.


While this form of cooperation benefits the organization, it may not

benefit the individuals in the organization.  We outlined several

management options for increasing individual's incentive to

participate.  We also discussed reasons people may participate without

incentives -- to gain status or because it was part of their

instinctive nature.


Individuals, organization, and tools co-evolve.  We shape our tools

and they shape us.  Imagine organizations without the clock, electric

light, telephone, or automobile.  Tools also change our individual

nature.  At the point where a tool is pervasive and no longer noticed,

our sub-conscious self has changed.  We perceive our environment

differently, therefore we are different.  If the propensity to

cooperate has a genetic component, and CMC bestows a marginal

advantage upon individuals and organizations that cooperate, we might

expect some evolutionary pressure in favor of cooperation.  Perhaps

CMC will marginally alter organizational and human nature, nudging us

in the direction of a time when nice guys finish first.

.pa


                                       Person
                                       Server
     ZDDDDDDDDDDDDD?     query     ZDDDDDDDDDDDDD?
     3             3DDDDDDDDDDDDDD>3             3
     3 User with   3               3  Matching   3
     3 Query       3    matches    3  Procedure  3
     3             3<DDDDDDDDDDDDDD3             3
     @DDDDDDDDDDDDDY               @DDDDDDBDDDDDDY
                                          ^
                                          3
                                          3
                                   ZDDDDDDADDDDDD?
                                   3             3
                                   3  Member     3
                                   3  Profiles   3
                                   3             3
                                   @DDDDDDDDDDDDDY


Figure 1.  A Person Server.  A query is matched against member
profiles, and the user is informed of the identity and characteristics
of members likely to posses relevant information.  Alternatively, the
system could automatically forward the query to matching members.  In
an object-oriented system, the matching procedure would be
encapsulated with the profile.

.pa

REFERENCES

.pa
1.  Axelrod, Robert, "The Evolution of Cooperation". Basic Books,
    New York, 1984.

2.  Axelrod, Robert and Hamilton,William D., "The Evolution of
    Cooperation". Science, Vol 211, pp 1390-1396, March 27, 1981.

3.  Bonacich, Phillip and Schneider, Sherry, "Communication
    Networks and Collective Action". personal communication,
    1990.

4.  Brand, Stewart, "The Media Lab". Penguin, New York, 1988.

5.  Cuadra, Carlos, "Directory of On-Line databases". Elsevier,
    1989.

6.  Dawkins, Richard, "The Selfish Gene". Oxford University
    Press, Oxford, 1976.

7.  Deutsch, Peter, Emtage, Alan, and Heelan, Bill, "archie - An
    Electronic Directory Service for the Internet". FTP:
    archie/doc/whatis.archie@archie.mcgill.ca

8.  Droms, Ralph., "Access to Heterogeneous Directory Services".
    Proceedings of the InfoCom Conference, June, 1990.

9.  Dumais, Susan T., Furnas, George W., Landauer, Thomas K.,
    Deerwester, Scott, and Harshman, Richard". Using Latent
    Semantic Analysis to Improve Access to Textual Information".
    Proceedings of the ACM National Conference on Computer-Human
    Interaction, 1988.

10. Gregory, C. A., "Gifts and Commodities". Academic Press,
    London, 1982.

11. Hardin, Garrett, "The Tragedy of the Commons". Science,
    December 13, 1968, pp 1243-1248.

12. Harrenstien, K., "Name/Finger". RFC 742, December, 1977.

13. Harrenstien, K., Stahl, M. and Feinler, E., "NICName/Whois".
    RFC 954, October, 1985.

14. Hesson, Sylvia B., Cartermill, Inc., Baltimore, MD, (410)
    563-2378, personal communication, January, 1992.

15. Kahle, Brewster and Medlar, Art, "An Information System for
    Corporate Users: Wide Area Information Servers". FTP:
    pub/wais/doc/ wais-corp.txt@think.com, April, 8, 1991.

16. Karabenick, S. A., Computer Conferencing: Its impact on
    Academic Help-Seeking, The Second Guelph Symposium on
    Computer Conferencing, pp 69-76, Ontario, Canada: Guelph
    University, Department of Rural Extension Studies, 1987.

17. Kenner, Hugh, "Out my Computer Window". Harpers Magazine,
    November, 1989.

18. Lovelock, J. E., "Gaia". Oxford University Press, Oxford,
    1979.

19. Lynch, Clifford A., "The Z39.50 Information Retrieval
    Protocol: An Overview and Status Report". ACM SIGCOMM
    Computer Communication Review, pp 58-70, Vol 21, No 1,
    January, 1991.

20. Malone, T. W., Grant, K. R., Turbak, F. A., Brobst, S. A.,
    and Cohen, M. D., "Intelligent Information Sharing Systems".
    Communications of the ACM, 1987, 30, 390-402.

21. Marill. T. and Roberts, L. G., "Toward a Cooperative Network
    of Time-Shared Computers". Proceedings of the Fall Joint
    Computer Conference, 1966.

22. Maslow, Abraham, "Motivation and Personality". Harper and
    Row, 1954.

23. Maynard-Smith, John, "Evolution Now". W. H. Freeman and
    Company, San Francisco, 1982.

24. Nelson, Theodore, "A Conceptual Framework for Man-Machine
    Everything". Proceedings of the 1973 NCC, AFIPS Press,
    Montvale, NJ, 1973.

25. Packer, C., "Reciprocal Altruism in Papio Anubis". Nature,
    265, 441-3, 1977.

25. Press, Larry, "Collective Dynabases". Communicatios of the
    ACM, in press.

26. Quarterman, John S., "The Matrix: Computer Networks and
    Conferencing Systems Worldwide". Digital Press, Bedford, MA,
    1990.

27. Roberts, L. G. and Wessler, B. D., "Computer Network
    Developments to Achieve Resource Sharing". Proceedings of the
    Spring Joint Computer Conference, 1970.

28. Schein, E. H., "Coming to a new Awareness of Organizational
    Culture". Sloan Management Review, Winter, 1984, pp 3-15.

29. Schwartz, Michael F., "Resource Discovery and Related
    Research at the University of Colorado". ConneXions, pp 12-
    29, Vol 5, No 5, May, 1991.

30. Schwartz, Michael F. and Tsirigotis, Panagiotis G.,
    "Experience with a Semantically Cognizant Internet White
    Pages Directory Tool". Journal of Internetworking Research
    and Experience, Vol 2, No 1, March, 1991.

31. Shapiro, Neil R., Diamantopoulos, Elias, and Cotton, Paul,
    "CD-ROM Disc Interchangeability Standards: Beyond ISO 9660
    with Structured Full-Text Query Language (SFQL)". ATA/AIA 89-
    9C Monograph, April, 1991.

32. Sherman, Paul W., "Nepotism and the Evolution of Alarm
    Calls". Science, Vol, 197, pp 1246-1253, September 23, 1977.

33. Simon, Herbert A., "Organizations and Markets". Journal of
    Economic Perspectives, 5:25-44 (Spring 1991).

34. Stein, Richard Marlon, "Browsing Through Terabytes", Byte
    Magazine, May 1991, pp157-164.

36. Streeter, Lynn A., Lochbaum, Karen E., "An Expert/Expert-
    Locating System Based on Automatic Representation of Semantic
    Structure". Proceedings of the IEEE Conference on AI
    Applications, pp 345-349, San Diego, CA, March, 1988.

37. Stodolsky, David, Computer-Supported Cooperative Work: The
    Question of Personal Integrity, International Conference on
    Information Systems, Work, and Organization Design, Berlin,
    July, 1989.

38. --, "WAIS Interface Protocol, Prototype Functional
    Specification". April, 1990.  Anonymous FTP:
    /pub/wais/doc/protspec.txt@think.com.

39. --, "Standard for the Exchange of Digital Information on CD-
    ROM". DCI Information Handling Committee, Intelligence
    Community Staff, Washington, D. C., 20505, Version 3.1, May,
    1991.