In this paper, the author examines the characteristics of information agency, the work of librarians and of intelligent agents as information mediators, the differences between human and software agents, the possible tasks for software agents in libraries, and speculates on the future of human and software agency. A typical medical library-based information need is presented and the attendant information processes are examined. The author describes the future of information mediation as based on efficient interaction between human and software agents and provides examples of possible collaborative information tasks.
Contents
Introduction
Definitions
How a Librarian Handles a Typical Information Need
Librarians as Cooperative, Distributed Agents
Similarities of Librarians and Software Agents
Differences Between Librarians and Software Agents
The Future: Collaboration
Conclusion
Introduction
Intelligent software agents promise to traverse and organize information spaces for us, alert us, remind us, call for a refrigerator repair-person, communicate with each other ... to fundamentally alter how we accomplish many of our daily tasks. These red-hot and revolutionary software critters have a lot to learn from their closest human peers: librarians. As I read and think about how intelligent systems reason, search, classify, and filter information, I'm struck repeatedly with how librarians do exactly these same tasks. Both act as information mediators for the end user: both negotiate information spaces and retrieve information relevant to a particular user or goal. Librarians have been efficiently accomplishing many of the tasks at which the artificial intelligence community is now working to make software agents competent. Therefore, the development of software agents can be informed by a look at how human information agents do their work.
This paper will examine the characteristics of agency, the work of librarians as information mediators, the differences between human and software agents, the possible tasks for software agents in libraries, and speculate on the future of human and software agency. Bonnie Nardi and Vicki O'Day, from whom many of the formative ideas for my thinking in this paper came, said this:
"We ask librarians to consider and speak out about their critical role in the design of a diverse information ecology." (Nardi and O'Day, 1996. "What We Learned at the Library," p. 86).
This paper is one attempt to do just that.
Definitions
The definition of an intelligent agent is much disagreed upon, but this one serves well:
"An intelligent agent is software that assists people and acts on their behalf. Intelligent agents work by allowing people to delegate work that they could have done, to the software agent. Agents can, just as assistants can, automate repetitive tasks, remember things you forgot, intelligently summarize complex data, learn from you, and even make recommendations to you." (Gilbert, 1998).
Note the striking similarity to this definition of the work of a librarian:
"The user works with an oracle (i.e., expert librarian) who: helps elaborate the problem (or anomalous state of knowledge), describes the user and user background, understands the topic or subject, constructs queries or other requests for desired information, knows how to access and present the relevant information, and provides explanations as appropriate." (Fox)
Both librarians and software agents do work delegated to them by a user; both apply some expertise to the user's problem or need; and both work in the background, completing tasks for the user's information needs.
How a Librarian Handles a Typical Information Need
"Librarians are more than technicians. They are, it seems, information therapists who analyze problems as well as find answers." (Nardi, O'Day, and Valauskas, 1997)
To demonstrate the comparison of the work done by librarians and the work done by intelligent software agents, let's follow a typical medical library user request from its initial state to its particular goal state.
Librarians use what is called a reference interview to learn about and assess a client's needs. The reference interview is often an iterative process with much back and forth between librarian and client as they work together to refine the information need. This process can be quite delicate: many clients are unclear about their needs and a skilled librarian can help them specify and communicate their need. The goal of the interview is the disambiguation of the client's need. The librarian works for relevance feedback during the interview, asking the client "Is this what you mean?" and "More like this?" The librarian must learn about the user's context: knowledge of client's situation, history, and preferences, providing a context for the inquiry.
In this example, a physician identifies an information need and approaches one of the medical librarians in his organization. The physician has a patient who is taking an herb the physician does not know much about. Here's a typical telephone dialogue between the physician and a medical librarian:
Physician: Would you do a search on a specific herb taken for depression?
The physician is well-aware of the availability of this service (and is able to search for the information himself, but prefers - due to lack of familiarity with the subject - that the professional librarian do it), having used the service before, and requests a search of the journal literature. The physician phrases his information need vaguely.
Librarian: Are you looking for information for you or for information to give a patient?
The librarian begins the process of defining the information need: if the information is for the physician, she knows that he will want articles from peer-reviewed clinical journals; if the information is to be handed to the patient, it will need to be in lay language and come from quality consumer health information sources. The librarian is eliminating uncertainty, defining the context of the information need, and establishing a profile of the information consumer.
Physician: I have a patient who is taking some herb or other; she thinks that it will help with her depression. It's called St. something ...
Librarian: So you want information for you? So that you can advise her on taking this herb?
The librarian continues to clarify and confirm the context of the information need.
Physician: Yes. What do you have? When can I get that?
Librarian: OK, I will identify the herb and find general articles on its applicability to depression. How many years do you want me to cover: one, five, twenty?
The librarian confirms the topic and begins to obtain specific limits on the need.
Physician: Just the last five years. Can you fax that to me?
The librarian obtains the physician's name, phone number, fax number, and e-mail address and the conversation ends. As she obtained the information from the physician, she filled in the blanks of an existing literature search request form.
Creating a Representation of the Problem
The next step the librarian takes is to use this information to form the question in her mind:
"The librarian contributes to the client's activity and to do so effectively, creates a representation of the activity that guides and focuses the search. This representation goes beyond understanding the client's task, simplistically conceived, to a broader contextual sketch of the client, including the client's preferences, constraints, and environment." (Nardi and O'Day, 1996. "What We Learned at the Library," p. 75).
The librarian may sketch the search steps on paper, choosing databases, search terms, and operators or she may just form this representation mentally. She considers the cost of the search, the quality of the databases at her disposal, the customer profile, and the ultimate use of the information requested.
Classifying Information
The librarian's search depends upon the work of other librarians: librarians who classify information by indexing and abstracting the articles included in the MEDLINE database, the pre-eminent peer-reviewed clinical database, produced by the U.S. National Library of Medicine. The articles in MEDLINE have been classified by extracting key concepts and assigning subject terms or matching subject terms to an existing controlled vocabulary, called MeSH (Medical Subject Headings). Many MEDLINE search engines automatically map a user's input term to the appropriate MeSH term (e.g., the user inputs "ear infection" and the system maps to "otitis media"). This expert classification scheme organizes the clinical information into a database, so that a librarian skilled at searching can find the most relevant articles quickly and efficiently.
If the data were not so carefully organized, the librarian's work would be even more complicated. The World Wide Web, for example, is not well-organized: even the best indexing programs and search engines are inefficient when compared to the careful, standardized classification scheme applied to the MEDLINE data set. Indeed, the Web has been compared to a library with all the books pulled off the shelves and thrown in a heap on the floor. The classification of information that has taken place behind the scene is critical to the efficient work of the librarian in our example.
Finding Information
Once the representation of the problem is established, the librarian chooses which database(s) to search. Since this information will be for the physician's consumption, she chooses the MEDLINE database. Since the physician asked for the most recent articles, the librarian selects the segment of MEDLINE that covers 1996 to the present. Her next step is to choose search terms. Knowing that MEDLINE is based upon a controlled vocabulary, she enters the term "herbs" and then searches separately on "depression." She combines these two separate search sets with the Boolean operator and, resulting in one citation. She looks at this citation and judges it to be irrelevant; her search strategy must be reworked. Searching the broader term "alternative medicine" results in 5,847 hits; this set is anded together with the herbs set, resulting in ten hits. The librarian takes a look at these ten and finds a couple of them mention St. John's wort. One of the articles gives the botanical name of St. John's wort; she searches on that term; this search results in eight hits. She takes a look at these eight and sees that they map to the MeSH heading antidepressive agents. She searches on this term, ands it with the alternative medicine set, resulting in nineteen hits. She limits these nineteen to the English language and to studies on humans; this narrows the set to fifteen. She takes a close look at the fifteen and weeds out three that are not applicable. She downloads these twelve to her hard drive. Knowing that the physician practices Evidence-Based Medicine (EBM, defined as "an effort to bring the principles of continuous quality improvement to bear on the patient care process instead of viewing practice as an art form, where each patient is treated as an experiment with an n of one" - Stead, 1998; p. 26), she reruns the same search in a database of EBM-evaluated articles (Cochrane Database of Systematic Reviews). This turns up one article, which she looks at and then downloads. The librarian now has a total of thirteen relevant citations on the subject.
Once into the search, if she had felt that she needed more guidance from the physician, she would have contacted him and asked in which specific direction he would like her to proceed. For example, if her search had resulted in extremely high recall, she might have asked the physician how he would like the results narrowed; if she knew him well, she might know that limiting the set to review articles would be acceptable to him and therefore, she would not need to call him. Past experience with the client and knowledge of the client's context for each particular information request guides the librarian in her search: if this had been a first request from this particular physician, the librarian might have spent a little more time learning his specialties and preferences.
The best searchers feel their way through a search, using past searching experience and knowledge of the particular client, the subject domain, and the idiosyncrasies of the resource to balance recall (the percentage of hits found of relevant hits available) and precision (the percentage of hits retrieved that are pertinent) and to ensure relevance (the hits' fit to the need). This balance is delicate and dynamic and illustrates the real art of effective searching. The librarian also knows, from formal education, past experience, and a familiarly with the databases she is using, when to cut off the search. A search could go on indefinitely, database after database, if the librarian did not possess the expertise to know when to be satisfied with her results. This is a big problem in a medical setting, especially, when the searcher is tempted to think that if he or she keeps searching, they may just find that one article that saves a life. The cut-off decision is usually an intuitive one and an important one that maximizes the temporal, computational, and financial constraints a librarian must always balance with the desire for completeness of results.
Filtering Information
"Information filtering is a process in which a filtering agent reads every document in an information stream and compares it to a set of interest profiles. If the document matches a profile, the filter sends it to the appropriate user's inbox or stores it somewhere for the user. The rest of the documents are filtered out." (Williams, 1996; p. 174)
The librarian filtered information in three key ways: first, the act of selecting a database or databases to search was a filtering activity. This choice established the constraints on the results and defined the search space, including the depth and breadth of the search. The second way the librarian filtered information occurred in the act of choosing search terms and operators (which also effects the depth and breadth of the search). Finally, when she eliminated false drops (noise) from the results, she was filtering the results the end user would see. Using her knowledge of the context of the information need and the profile she had constructed of the user, she was able to identify applicable results for the physician, relevant to the specific context of this particular information need.
Packaging Information
Librarians do more than connect people to raw information: they use their expertise to help clients make sense of information. Librarians usually arrange search results and other information products into customized sets for their clients. The librarian in our example used her selected citations to create an aggregated report for the physician. She probably added a cover sheet with the contact information for her library, cleaned up the citations so that the fields the physician needed were present and deleted any he would not need. She may have sorted the results by relevance, and she may have indicated on the report which articles were available in her library.
Librarians as Cooperative, Distributed Agents
Librarians can be seen as distributed, cooperative agents. Librarians inform and negotiate with each other; cooperation is built into the fabric of library work. It is not too far-fetched to say that there is no such thing as a librarian who works truly alone: one-person libraries certainly exist, but that lone librarian works within a library community. Indeed, there is a real international community of librarians. For example, a librarian can connect to a library system called WorldCat and tell a client, within a couple of minutes, what libraries in the world own a certain book. Further, a librarian can, with a few key strokes, send a request to any one of those libraries asking to borrow that book for the client.
Librarians rely on each other in other activities: many librarians practice cooperative collection development. No one medical library, for instance, can afford to subscribe to all published medical journals. When considering a purchase, acquisitions staff check to see if that title is owned by a nearby library with whom they have a cooperative arrangement.
Libraries communicate and share information to an amazing degree. Librarians are great users of Internet mailing lists; some of the highest volume lists are those run by and for librarians. Librarians are communicating all the time: via e-mail, phone, fax, and in person. For example, I recently attended the annual meeting of the Indiana Health Sciences Librarians Association. After a meeting dinner, this group of thirty or so librarians sat until almost eleven at night talking, sharing strategies and stories. I gave a short presentation the next morning. The following day, I e-mailed copies of my library's extensive marketing plan to librarians all over Indiana who had requested it. We have few secrets and we depend upon each other implicitly.
By the very nature of our work, librarians are collaborative creatures who communicate extensively and constantly. We distribute information, share documents, strategies, and innovations with each other. Those of us with expertise in the domain of medicine query librarians with expertise in the domain of law, for example. A single library can even be seen as a distributed system of agents: the reference librarian fields this inquiry, passes that one to the acquisitions staff; the systems librarian consults the clinical librarian ... Each librarian acts autonomously with the frequently used option of consulting another librarian-agent. Within the library field, librarians efficiently share specialized knowledge and skills.
Similarities of Librarians and Software Agents
Agents - human and machine - possess certain characteristics. Agents act on behalf of a client by performing tasks the user does not want or know how to do themselves. Both human and software agents work in the background; much of the work of human library agents is invisible by design. "Human agents make a point of making their jobs look easy. In a sense, that is part of the service they provide" (Nardi and O'Day, 1996. "What We Learned at the Library," p. 79).
Both human and software agents begin to achieve a goal or solve a problem by building a representation of the problem state. Software agents create this representation using knowledge representation techniques such as first order logic and predicate logic. Agents are goal-driven: they seek to solve a specific problem, fill a specific need. Context is all-important to both agents; objectives and rewards are situation-dependent. Each information act is meaningless if not carried out in relation to some specific goal.
Both human and software agents possess certain domain information (i.e., neither is an expert in all information realms). Agents accomplish their work using some variation of a profile of the user. Agents seek to minimize work by load balancing and the efficient allocation of temporal, financial, and computational resources. The work of both human and software agents is of an iterative nature. Much back and forth takes place in checking results. Both have as their goals efficiency and optimality. The mechanical agent uses artificial intelligence techniques to aim at optimally achieving its goal. Librarians aim for optimality by striving for relevant results obtained speedily and cost-efficiently (such as choosing a low-cost database over a high-cost one). Both agents work in uncertain worlds. Their environments are dynamic and the agents themselves affect the information state. Human and software agents face many of the same challenges, concerns, and goals.
Differences Between Librarians and Software Agents
For all these similarities between librarians and software agents, there are marked differences.
Brittleness
Intelligent software programs all face the brittleness problem: how well do they function at a task outside their limited domain? They may excel at chess but flail when queried for obvious information. Librarians are rarely brittle: in fact we are so agile at handling and finding our way through information spaces that we have become far too modest. Our flexibility from client request to client request is often staggering. As a medical librarian, I have a foot in both worlds: librarianship and medicine (indeed, I could add marketing and education and other domains to that claim). Librarians are very flexible; in fact, flexibility is one key to our future relevance and usefulness. Library clients are often startled by the information a good librarian has in his or her head (e.g., "Knowing the project you're working on, I saw this article and thought you might be interested.")
User Attitudes
User acceptance of an agent is a very individual thing based on personal preferences, skill levels, and previous experience with an agent. Users have varying skill levels. Many prefer to delegate most of their information search tasks to a librarian; some prefer to search themselves. Some users will never feel comfortable with the results a software agent provides; other users prefer to deal with computers over people. Some users overestimate technological abilities and some users are Luddites who miss the wooden card catalog with its paper entries. Some users need the human touch, preferring to brainstorm with a librarian than to deal with a literal computer program. Once a person has a good experience with a librarian agent or with a software agent, that user is more likely to use the agent again in the future.
Competence
A good librarian is more competent than a software agent at this point in time, hands-down. A really good librarian has to develop by experience and by training, just as software agents do, but the development time for a human agent is quicker than for a software agent. Librarians are cheaper to install, so to speak. Hire one, give him or her a year or so to get up to speed, and you've got a professional agent who will provide clients with relevant information. Once installed, though, librarians are also much more expensive to keep and are slower in their work than software agents. The artificial intelligence community is working to make software agents more agile and more competent. It is just a matter of time before intelligence software agents are turn-key installations able to competently perform lower-level reference (and other) library tasks.
The Future: Collaboration
Those of us who work in libraries have heard both theories: the human librarian is obsolete - referred to as the "Redundancy Theory" by Hathorn (1997) - and the other extreme: this is the librarian's moment to grab the brass (Internet) ring and do our organizing magic on the wild range of information overload. Hathorn names this the "Masters of the Universe Theory." In my opinion, this will go neither extreme way, although the threat of obsolescence is real enough if librarians don't start marketing and managing our images. A reasonable information future will include human librarians and intelligent software agents. Since a real general artificial intelligence is not yet within sight, a collaborative future makes the most sense. Software agents can automate some of the more tedious and repetitive library tasks and librarians can be freed to do what they do best; guide clients through the maze to the best information for that client's particular need.
"Rather than seeing human agents and software agents as in competition, as vying for the same place in our world, the wiser course is to leverage the strengths of each, deliberately designing work practices and institutional arrangements that reflect and exploit the possibility of collaboration between human and software agents." (Nardi and O'Day, 1996, "What We Learned at the Library," p. 83)
Examples of such collaboration include the following scenarios:
Mediation between searcher and information:
- assistance with database and Internet site selection;
- automated searching and retrieval of information - based on users' queries - from the Internet, online databases, and/or corporate data warehouses;
- assistance with query structuring (e.g., matching users' terms to controlled vocabularies, subheadings, and thesaurus terms, suggesting similar concepts/terms);
- assistance with strategy modification; and,
- assistance with the interpretation and sorting of search results.
Virtual/automated reference:
- good for all-hours reference service;
- useful for serving geographically distant library customers;
- helpful for users who may be unable physically or unwilling to approach a librarian in person;
- provide automated, context-sensitive creation and presentation of frequently asked reference questions;
- guided tutorials (e.g., database searching, learning and using applications); and,
- indexing and abstracting: assigning index terms (automated matching to thesaurus/controlled vocabulary) and abstracting (identifying key sentences).
Automating serials processing:
- recommendation of titles for acquisition;
- updating databases when titles change or cease;
- identifying missing serials issues and generating claims; and,
- processing statistics and generating reports.
Interlibrary loan processing:
- identifying the best source(s) for requested documents;
- routing document request to source;
- updating and notifying document delivery staff of the status of open requests;
- archiving closed requests; and,
- generating reports (e.g., document delivery statistics).
Acquisitions:
- recommending acquisitions based on budget, reviews, size and type of institution, history of purchases and rejections, etc.;
- assignment of vendors to items selected for acquisition;
- automation of copy cataloging (and able to handle diverse formats); and,
- advisory function for human cataloging (e.g., ensuring rule compliance,
eliminating duplicates, automatic identification of main entry terms, cataloging tutorials).
Circulation:
- checking items out to users;
- generating overdue notices;
- processing renewals;
- generating circulation statistics and reports; and,
- generating invoices for overdue/lost materials.
Digital libraries:
- Indexing and classifying documents; and,
searching and retrieving documents.
Personal Information Gathering and Sorting:
- customized current awareness services;
- automatic, customized searching (sometimes in cooperation with other subject-expert agents);
- generation of personal newspapers and Web sites (e.g., MyYahoo!);
- customized search engines;
- e-mail filtering and sorting programs;
- custom recommendations of sites/items of interest; and,
- translation of documents from other languages. (Zick, 1999)
Conclusion
Librarians are thinking hard about the future of librarianship. Are we redundant? Can we be replaced by intelligent systems? What do we have to offer that machines cannot replicate? One article on the subject suggest the following as "Critical Characteristics of the New Information Professional" -
- Guide in the face of an uncertain future
- Collaborate
- Prioritize and maintain agility and flexibility in the face of changing goals
- Empower
- Understand the core capabilities of one's organization, work group, and colleagues (Griffiths,1998, p.8)
These suggestions are almost identical to the goals of the development of intelligent software agents: disambiguation, collaboration among agents, the empowerment of the user, and the customization of information for a particular user. Librarians and intelligent software agents must work together to efficiently balance workloads in the most optimal way for a particular user and in that particular user's specific context. Librarians need to resist feeling defensive about our jobs and our value and keep our focus on which technologies and which methods make the most sense for our clients. Once we decide upon methods, we need to do the best job possible to obtain those technologies, instruct our users as need be, distribute the highest quality content via our methods. If some Luddite librarians need to upgrade their skills, so be it.
Overly optimistic views of a technological future must also be avoided. The promise of easy solutions based on new technologies must be fact-based and proven, not just based on the rhapsodizing of a technophile. Who does the work is not as important as the fact that the work gets done in the best way for the client.
"Librarians are becoming more important in this information-centric universe. Intelligent agents will act as catalysts to enhance the role of librarians as the next century dawns, heralding a renaissance in information science and librarianship. Librarians are already fulfilling new roles as content providers, search strategists, digital catalogers, and information mechanics. These roles will only grow, and new opportunities will flourish, with the development of truly 'intelligent agents' that build upon the experience of librarians and information professionals." (Valauskas, 1997)
Our users are more overloaded every day and only a connectionist approach of intelligent work practices will best serve them.
About the Author
Laura Zick is a medical informatics specialist/librarian in the Medical Library & Informatics Training Program at Clarian Health Partners, Inc. in Indianapolis, Indiana. She received her Master of Information Science from Indiana University and is a member of the Academy of Health Information Professionals, the Indiana University School of Medicine faculty, and many professional organizations. Her research interests include such medical informatics issues as evidence-based medicine, artificial intelligence in medicine, the information seeking behaviors of health care professionals ... all with the end goal of intelligently, efficiently, and holistically bringing the highest quality knowledge-based information to the clinicians' point of need.
E-mail: laura@dochzi.com
Web: http://www.dochzi.com
Bibliography/Further Reading
Anonymous, 1998. "The Role of the Information Professional in the "Knowledge Economy," Electronic Library, volume 16, number 6 (December), pp. 373-376. http://dx.doi.org/10.1108/eb045664
Phil Bradley, 1999. "Intelligent Agents on the Web," Managing Information, volume 6, number 1 (January-February), pp. 35-41.
Vannevar Bush, 1945. "As We May Think," Atlantic Monthly, volume 176 (July), pp. 101-108, and at http://www.theatlantic.com/unbound/flashbks/computer/bushf.htm
Fah-Chun Cheong, 1996. Internet Agents: Spiders, Wanderers, Brokers, and 'Bots. Indianapolis: New Riders, 1996.
Graham Coult, 1999. "Intelligent Agents," Managing Information, volume 6, number 1 (January-February), pp. 33-34.
Edward A. Fox, n.d. "Electronic Librarians, Intelligent Network Agents, and Information Catalogues," draft at http://www.uky.edu/~kiernan/DL/fox.html, retrieved 11 April 1999.
Don Gilbert, 1997. "Intelligent Agents: The Right Information at the Right Time," http://www.networking.ibm.com/iag/iagwp1.html, retrieved 14 October 1998; no longer available.
Jose-Marie Griffiths, 1998. "The New Information Professional," Bulletin of the American Society for Information Science, volume 24, number 3 (February-March), pp. 8-11. http://dx.doi.org/10.1002/bult.83
Jean-Claude Guedon, 1998. "The Virtual Library: An Oxymoron?" 1998 Joseph Leiter Lecture for the Medical Library Association, at http://www.mlanet.org/publications/bmla/leiter98.html, retrieved 29 January 1999.
Clay Hathorn, 1997. "The Librarian Is Dead, Long Live the Librarian," PreText (October), at http://www.pretext.com/oct97/features/story4.htm, retrieved 28 October 1997.
Björn Hermans, 1997. "Intelligent Software Agents on the Internet: An Inventory of Currently Offered Functionality in the Information Society and a Prediction of (Near) Future Developments," First Monday, volume 2, number 3 (March), at http://firstmonday.org/issues/issue2_3/ch_123/, retrieved 11 November 1998.
Intelligent Systems Laboratory, n.d. "The Agent Based Digital Library Infrastructure Project, at http://www.sics.se/isl/diglib/, retrieved 29 January 1999.
James Jansen, 1997. Using an Intelligent Agent to Enhance Search Engine Performance," First Monday, volume 2, number 3 (March), at http://www.firstmonday.dk/issues/issue2_3/jansen/, retrieved 21 January 1999.
Deborah Jones, 1997. "Artificial Intelligence Applications in Libraries for Information Professionals," at http://www.bf.rmit.edu.au/Dimals/rguides/ai-library.htm, retrieved 24 March 1999.
F.W. Lancaster and Linda C. Smith (editors), 1992. Artificial Intelligence and Expert Systems: Will They Change the Library? Urbana-Champaign, Ill.: Graduate School of Library and Information Science, University of Illinois at Urbana-Champaign.
F.W. Lancaster and Linda C. Smith, forthcoming. Intelligent Technologies in Library and Information Services. Medford, N.J.: Information Today.
Andrew Leonard, 1997. Bots: The Origin of a New Species. San Francisco: Hardwired.
J.C.R. Licklider, 1965. Libraries of the Future. Cambridge, Mass.: MIT Press.
Patti Maes, 1994. "Agents that Reduce Work and Information Overload," Communications of the ACM, volume 37, number 7 (July), pp. 31-40, 146, and at http://pattie.www.media.mit.edu/people/pattie/CACM-94/CACM-94.p1.html http://dx.doi.org/10.1145/176789.176792
Patti Maes, 1995. "Intelligent Software: Programs That Can Act Independently Will Ease the Burdens That Computers Put on People," at http://pattie.www.media.mit.edu/people/pattie/SciAm-95.html, retrieved 16 February 1999.
Damien Miller, 1996. "Three Types of Intelligent Agents, Their Uses, and Their Future," NetscapeWorld, volume 1, number 9 (November), and at http://www.netscapeworld.com/netscapeworld/nw-11-1996/nw-11-agents.html
R. Bruce Miller and Milton T. Wolf (editors), 1992. Thinking Robots, an Aware Internet, and Cyberpunk Librarians. Chicago: Library and Information Technology Association.
Marvin Minsky, 1986. Society of Mind. New York: Simon and Schuster.
Bonnie A. Nardi, 1998. "Information Ecologies," Keynote Address at the Reference Service in a Digital Age: A Library of Congress Institute, at http://lcweb.loc.gov/rr/digiref/nardi.html, retrieved 8 January 1999.
Bonnie A. Nardi and Vicki O'Day, 1996. "Intelligent Agents: What We Learned at the Library," Libri, volume 46, number 2, pp. 59-88. http://dx.doi.org/10.1515/libr.1996.46.2.59
Bonnie A. Nardi, Vicki O'Day, and Edward J. Valauskas, 1996. "Put a Good Librarian, Not Software, in Driver's Seat,"Christian Science Monitor, (4 June), p. 18, and at http://www.csmonitor.com
Bonnie A. Nardi, James R. Miller, and David J. Wright, 1998. "Collaborative, Programmable Intelligent Agents," at http://www.miramontes.com/ADD-CACM/ADD-CACM.html, retrieved 1 January 1999.
Bonnie A. Nardi and Vicki L. O'Day, 1999. Information Ecologies. Cambridge, Mass.: MIT Press.
B. Nardi, and Y. Engeström, 1999. "A Web on the Wind: The Structure of Invisible Work," In: B. Nardi and Y. Engeström (guest editors).
Computer Supported Cooperative Work, volume 8, numbers 1-2 (special issue), at http://www.best.com/~nardi/InvisibleW.html, retrieved 29 March 1999.
Charles Petrie and Meredith Wiggins, 1997. "Pattie Maes on Software Agents: Humanizing the Global Computer," Internet Computing, volume 1, number 4, at http://computer.org/internet/v1n4/maes.htm, retrieved 11 April 1999.
Doug Reicken, 1994. "A Conversation with Marvin Minsky About Agents," Communications of the ACM, volume 37, number 7 (July), pp. 23-29.
Marina Roesler and Donald T. Hawkins, 1994. "Intelligent Agents: Software Servants for an Electronic Information World (And More!)," Online, volume 18, number 4 (July), pp. 18-32.
Stuart J. Russell and Peter Norvig, 1995. Artificial Intelligence: A Modern Approach. Englewood Cliffs, N.J.: Prentice Hall.
Marcos Silva, 1998. "Artificial Intelligence in Information Processing," at http://musicm.mcgill.ca/~marcos/http/ai2.html, retrieved 19 January 1999.
Barry G. Silverman, 1998. "The Role of Web Agents in Medical Knowledge Management," M.D. Computing, volume 15, number 4, pp. 221-231.
William W. Stead, 1998. "Positioning the Library at the Epicenter of the Networked Biomedical Enterprise," Bulletin of the Medical Library Association, volume 86, number 1 (January), pp. 26-30.
Edward J. Valauskas, 1997. "Who's Afraid of an Intelligent Agent?" Internet Trend Watch for Librarians, volume 2, number 3 (March).
Fred Weingarten, 1996. "Technology Will Free Librarians to Do ... What?" at http://www.oclc.org/oclc/man/9609ref/wein.htm, retrieved 19 April 1999.
Michael P. Wellman, Edmund H. Durfee, and William P. Birmingham, 1996. "The Digital Library as a Community of Information Agents," IEEE Expert, volume 11, number 3 (June), pp. 10-11. http://dx.doi.org/10.1109/MEX.1996.506747
H.G. Wells, 1938. World Brain. London: Metheuen.
Joseph Williams, 1996. BOTS and Other Internet Beasties. Indianapolis: Sams.net.
Laura Zick, 1999. "Artificial Intelligence and Libraries: A Primer for Librarians," at http://www.dochzi.com/l600/index.htm, retrieved 9 April 1999.
Editorial history
Paper received 19 April 2000; revised version received 28 April 2000; accepted 29 April 2000.
