BACKGROUND OF INTELLIGENT AGENTS
There are several factors contributing to the increased demand for "digital assistants". The sheer proliferation of information and the increased pace at which information is delivered due to greater bandwidth, the growing use of the Internet and World Wide Web, burgeoning complexity of desktop applications, as well as heightened competitive pressures within information rich industries and of course, the insistence of civilization to maintain a 24 hour day, have all contributed to the need for electronic helpers that find, filter and provide enhanced accessibility to information, in addition to automating tasks and in the case of mobile agents, manage networks. Such helpers are formally referred to as agents, which are generally defined as "self-contained, intelligent software entities that perform tasks on behalf of a user." (FTP Software Whitepaper).
Agent software employs artificial intelligence and knowledge-based or expert systems techniques which have developed gradually over the last 20 years. The exponential growth of the World Wide Web has propelled agent presence in the marketplace. The technical advantage of today's agents over expert systems are that agents are communications oriented, designed to communicate with the user, with system resources and with other agents, and are designed to be distributed, to travel through boundless bytes of data and across networks.
Pattie Maes, Pattie Maes Associate Professor at MITís Media Laboratory Media Lab , is one of the pioneers of agent research and development. She founded and continues to direct MITís Software Agents Group. Maesí group pioneered the use of machine learning to build agents (semi-intelligent computer programs to assist a user with the overload of information and the complexity of the online world) and invented new algorithms such as collaborative filtering. Her team built the first successful prototypes of agents for personalized information filtering, eager assistant agents, agents that buy and sell on behalf of a user, matchmaking agents and remembrance agents. She also oversaw the formation of Firefly Network, Inc. Firefly.com (see company brief below) in Cambridge, Massachusetts, one of the first companies to commercialize software agent technology in addition to founding Agent Society, Agent Society an international industry and professional organization established to assist in the widespread development and emergence of intelligent agent technologies and markets.
Metagroup.com study predicted that agenting software will become an innate piece of Internet and corporate intranet services by 2000. A 1994 Ovum consultancy report predicted that the market sector totals for agent software and products for USA and Europe will be worth at least $3.9 billion by the year 2000.(Computer Reseller News). A white paper produced in 1997 by Ovum claimed that total revenues earned by the four types of suppliers (agent technology developers, basic software developers, product developers and service providers) from agent-related products and services on the Web will grow from $19 million in 1996 to $4.7 billion in 2006. Projections indicate that most of this money will be retained by the service providers and product developers. Only 12% of total revenues will be obtained from the sale of agent architectures (Stark). There are presently 50 vendors currently supplying agent-enabled software and services. They are spread across virtually every application area (Intelligent Agent Whitepaper).
TOTAL
AGENT RELATED REVENUES ($MILLIONS)
There are three broad groups of customers who stand to benefit from agent-related products and services: (Stark)
AGENT CHARACTERISTICS
There are several unique factors that make agents differ from other software applications. On a general level, four characteristics of agents which are always true include: (IBM Intelligent Agent Whitepaper).
INTELLIGENT AGENTS: PRESENT
"The next wave of them has arrived, trained to race around cyberspace doing our (largely consumerist) biding while we're out watching football, making love, or doing whatever we've decided not to let our software do for us." (Douglas Rushkoff, "Second sight: The soul suckers" October 30, 1997, The Guardian)
Within the last two years, agents have been behind the major change taking place on the Internet away from "pull" technology, by which users have to find their own information, towards, "push" technology, which delivers the type of information they have requested to their screens. Moving from client-side search engine applets allowing users to seek out information on each and every occasion, agents have allowed for server-side customization and personalization, allowing marketers to capitalize on one-to-one marketing techniques and freeing the user from making all decisions. Internet marketers are using the agent software to market to groups of like minded people and to individuals. Firefly Passport represents use of the former, actively soliciting recommendations and opinions from customers on commerce driven sites (i.e., Barnes and Noble, etc.), then using that information to make purchasing or selection recommendations. Open Sesame employs the latter approach, which cumulatively collects information about an individual user's search patterns and makes recommendations on that basis alone.
Flexibility
on the part of the agent is critical in delivering accurate and usable
information. Because Firefly is targeted at a larger subset, it is
inherently more flexible in its parameters and has thus been employed more
frequently. Organizations like Pointcast, which already uses intelligent
agents to push news and information to subscribers hopes to use a combination
of both group and individual marketing to first sort out new customers,
determine their specific news preference and then assign them to a group
similar in orientation, and finally deliver aggregated group specific news.
Aside
from delivering content, intelligent agents are also presently being used
to manage information overload for people immersed in the clickstream.
With information entered by the user, an agent can learn to search the
Internet for important information that the user cares about. For
example, the agent can monitor the hundreds of Internet newsgroups in search
of anything the user might find helpful, such as information on a business
competitor. E-mail agents can examine incoming e-mail and tell the
user that an important message is waiting or begin to write replies for
some messages. Software such as AppleSearch or IBMís DataDetective
are two examples of agents which can perform these types of activities
now.
PRESENT PROBLEMS
Intelligent Agent software products (agents) offer a world of possibilities. They stand to make the flow of information on the Internet far more valuable. In fact, they might provide the long awaited push technology that would give customers personalized one-to-one marketing and a more exciting online experience. Some tout agents as being "the most deadly of the killer apps." However, there are some major problems to be overcome before they can live up to that type of billing. These problems can generally be separated into three main areas.
VENDOR CONCERNS
One series of problems relates to vendors. The biggest problem in this regard is that there just aren't enough of them online. Even if consumers are provided with effective agents, the number of vendors with online services is not even close to enough for those agents to replace the traditional commercial experience. In some cases, the best deal to be found on the Internet may not even crack the top ten of what a customer can find on their own. As a result, the usefulness of these agents is still limited to a fairly narrow set of tasks.
Additionally, even as more vendors set up shop online, they may not be too receptive to the use of customer agents. The problem from a vendor's standpoint is that these tools may tend to make commodities out of many vendor offerings. Because agents currently have no way to value qualitative factors like service quality or name brand, they simply make price comparisons. This could have the effect of eroding price differentials and eventually threaten the entire role of retailers by making it easier for the manufacturers to deal directly with customers. In the future, retailers could provide additional fields that would allow agents to make some qualitative comparisons. However, that day seems a long way off.
CUSTOMER CONCERNS
Another series of problems relates to customers themselves. The biggest concern here is the potential invasion of privacy. Agents used by vendors would potentially have the ability to collect enough information about a customer so that personalized suggestions could be made to fit the preferences of that person. However, others could use that information to do the same thing. This presents a real threat to the privacy of this information. If customers feel that they are going to open themselves up to all types of unintended parties, they may be hesitant to use a vendor's site. Thus, vendors must find a way to avoid the perception of a privacy threat.
An additional problem is that agents may limit customers to a constrained set of choices. For instance, what if a consumer who is about to go on vacation and spends a majority of their time in the prior month searching for vacation related information? The customer agent may then assume that this is the main area of customer interest and then proceed to send more information on that subject. This could eventually lead to an ever-decreasing range of information being sent to the customer. While this problem may be solvable, it is the perception of this type of problem that will have to be overcome.
TECHNOLOGICAL CONCERNS
One of the biggest technological problems with agents is the issue of "data context." The problem is that the exact same data can mean different things in different environments. For example, if the agent goes out and finds a price for something, is the price in dollars or pesos? This is just a small example of a very thorny issue. Due to the nature of the Internet, there is no way to get everyone to standardize the presentation of information. Therefore, agent programmers must find some way to translate data from different contexts into a common form. If this problem can be solved, it could provide huge benefits in areas such as financial analysis and procurement.
Another technological problem involves the interaction of different agents. One could envision a world where one specialized agent could interact with others in pursuit of its final goal. There are actually two problems with this vision. The first is that there is a lack of standards with regard to how agents are programmed. Because of this, it will be difficult to allow agents to communicate with each other. The second problem, bandwidth, comes in when this communication problem is solved. If the day comes when agents are roaming around the Internet interacting with one another, it will take up an enormous amount of space.
Overall,
agents seem to provide great promise. However, they still have a
long way to go before they will fulfill the vision that many have for them.
Some of the problems facing agents may be relatively easy to solve.
However, problems like data context and qualitative comparisons will be
far more difficult. Effective solutions to these problems will be
critical to the continuing evolution of intelligent agents.
.
THE FUTURE
In the academic world, when one jumps "three orders of magnitude," he, she or it is considered to have created a whole new science. For intelligent agents, such a jump appears imminent. As the number of applications and task-specific agents explodes in the near future, agents will shop for your favorite shirts, conduct stock transactions, maybe even make coffee in the morning. After that, the possibilities seem virtually endless.
1) Near Future
The Task. Today's push technology is considered by many to be "too pushy," often overloading users with information that they don't need. In the near future, we can expect a generation of delivery technology that uses intelligent agents to deliver the information that users want, when they want it. (PR NewsWire, 7/22/97, "Jazz Multimedia...")
The current "surf"-based web will also become a mere memory in the minds of users in the near future. "The Web will go away because it is too painful a process. I see a future in which the agents will be able to negotiate, learn user preferences, search and ferret out information for people, provide access to data that may not be on the Web and collect that information for you in a way that you want to see it." (Network Computing, 9/1/96) Industry.net chief scientist noted last year that "constant internet connectivity and networked computers, inevitable by the year 2000, combined with intelligent agents specialized to their tasks, will bring a future to homes and businesses in which agents do your information research, discover bargains, find movies, and scout recipes."
The Technology. Intelligent agents currently work, generally, in isolation and perform singular, relatively simple tasks such as filtering e-mail and retrieving information on a scheduled basis. The next step will be the widespread introduction of a system that is portable. Mobile agents are the tools to enable this. However, while mobility will appear gradually over time, as the infrastrucure for agents matures, in reality they are "not really viable right now." (Computer Reseller News, 4/17/97) Additionally, agents will soon work as a team, with each member designed to carry out a specific function. Experts also envision agents being able to find other agents in cyberspace to perform specific tasks on a contractual basis.
In the software market, with the move to E-Commerce as the model for business, capabilities such as Firefly will become standard components of purchases during the next couple of years. In following, analysts believe that either more server-side intelligent agent developers will emerge on the market soon, or the few that already exist will be swallowed up by the likes of Microsoft or Netscape and repackaged under their already-powerful brand names.
The System. Finally, the near future may also hold a standard agent building system which allows system designers to easily add intelligent agents to both existing and new applications. This would help accelerate the acceptance of agents into the marketplace.
2) Far Future
No Limits. Farther into the future, we may begin to see a new generation of intelligent agents called "Digital Pets." Not those things you cart around on your key chain, but bits of software that can carry out your stock bidding by scanning your e-mail for important messages, and further customize how you interact with your computer. The technology behind this is called "artificial life." When combined with intelligent agents, the agent element would execute the userís instructions while the "a-life" would allow it to evolve and adapt to the userís tastes. (Neil Gross, "Into the Wild Frontier," 6/23/97, Business Week)
CSTaR, Anderson Consultingís Center for Strategic Technology Research, is developing a "wearable computer" named the "Awareness Machine," which frees a user from the desktop and provides all relevant and important information to the user wherever he or she may be. The proposed device would work by incorporating intelligent agents into the server that is located at the userís home office, pre-programming them to watch the userís schedule, to know his/her long-term interest, to constantly monitor the userís mailbox, to filter the news available on the Internet. Linked by radio waves. (Wendy Lee, "Converging Technologies for ubiquitous computing," 8/14/97, New Straits Times).
At
MITís Media Lab, one student is developing a project called Yenta, which
monitors a userís incoming e-mail and begins to learn about the subjects
that the user cares about. "It does statistical analyses of how often
words show up in the documents you get and send, and forms clusters of
topics. It can also introduce the user to others who share the same
interests, or the same problems within a company to promote collaboration.
(Scott Kirsner, "Demo or Die," 9/20/97, Editor & Publisher Magazine)
INTERFACE
AGENT:
Autonomy,
Inc. was founded in March 1996 by Dr. Michael Lynch, an expert in the field
of adaptive pattern recognition. The company is headquartered in Palo Alto
but its roots lie with Dr. Lynch in Cambridge, England. Dr. Lynch
applied the arcane science of neural networks (systems which mimic
the thought patterns of the human brain) to attempt to disentangle the
huge amount of information available on both the Internet and corporate
intranets. Autnomy's intelligent agent software can look for patterns
in large numbers of documents and decide what is similar. According
to an article in the October 16th, 1997 issue of The Guardian,
"central to Autonomy's success has been the flexibility of its pattern
recognition technology (which) allows online services to personalize their
offerings." Internet users face the everyday dilemma of sorting
through countless streams of non-relevant information. Autonomy's
systems helps companies categorize the information to make it user friendly
and relevant. In addition, Autonomy has developed certain products
to screen out offensive material.
The company is privately held with a current valuation of more than $45 million but Dr. Lynch has announced plans for a partial flotation Nasdaq within one year. As of July 1997, after only 13 months of business, Autonomy reached profitability. Autonomy is an outgrowth of a company called Neurodynamics which was founded in 1991 to develop applications for pattern recognition work. The most successful product from that company, a system to match fingerprints with car license plates, became the catalyst to explore other applications for pattern recognition technology.
Autonomy
develops, markets and integrates solutions for the intelligent management
of unstructured data and personalized information delivery in online environments.
Autonomy's products provide complete end-to-end information management
solutions. According to the company literature "Agentware products get
the right information to the right people when they need it - automatically.
Autonomy's technology understands user interests and the context and concepts
within unstructured digital content the products understand user interests
via either observation or explicit natural language training." Autonomy's
"Concept Agents" use this understanding to retrieve relevant content
from all sources. Concept Agents are intelligent in that they learn and
adapt to a user's changing interests. Since the technology is language
independent, the Concept Agents do not rely on any particular language
structure, and are not dependent on grammatical or regional variations
in language.
RESIDENT
TASK-SPECIFIC AGENTS:
Firefly
takes "word of mouth" recommendations to a global level. This Boston-based
company uses intelligent agent software, developed at MIT's Media Lab,
to take advantage of the scope of the internet to communicate directly
with individuals. Firefly allows participating companies to build
extensive profiles of user preferences and then use those sets of preferences
to make recommendations for other like users. The company is initially
focused on the music, book and movie industry but has plans to expand into
mutual funds and restaurants. Information collected from users is
also applied to the advertising banners that individuals see. If
you express a preference for biographies and classical music, the
banner ad that you see will be dramatically different from the one a user
with interests in sports and hard rock views.
According to the company literature, Firefly is "the leading provider of products and services for relationship management and advanced personalization. Firefly products enable businesses to build and manage profitable relationships with their customers while offering them personalized attention and services." In other words, Firefly offers companies the ability to gather relatively inexpensive demographic data about customers and develop the ultimate one-to-one marketing relationship. According to the October 7th issue of Business Week "the software agents developed by Firefly could move (marketers) closer to their Holy Grail by providing a way to predict what customers are likely to want next -- and the means to reach them with a customized pitch that could cost a tenth of the more traditional marketing programs." Firefly also enables marketers build affinity groups around their products in a more efficient manner.
Visitors to a Firefly site register using an alias to protect their privacy and begin to step through a series of questions designed to tally their likes and dislikes. For example users of the MyLaunch music site begin by telling which radio station they listen to in their home area. Then based on that information, they are asked to rank, on a scale of one to six, over 40 different albums. This is where the intelligent agent software takes over and sorts through similar preferences of other users. The information culled from these "nearest neighbors" is used to make recommendations about other albums or music related events. Since users are required to reveal a fair amount of personal information, Firefly has taken aggressive steps to protect the privacy of users, including twice yearly audits by Coopers & Lybrand to monitor its promises.
Although
the company is not currently profitable the revenues received through advertising
have given the company a growing income base. However, the true money
most likely lies with licensing agreements to other companies. Current
rumors abound surrounding the future uses of Firefly include licensing
by Yahoo!, ZDNet and Reuters.
MOBILE
AGENTS:
General
Magic develops and markets intelligent voice-centered communications products
and services for people who need constant access to information. Integrating
voice and data, and using the company's advanced voice-user-interface technology,
General Magic's products and services give their subscribers an easy and
convenient way to keep in touch with the people and information important
to them. The handheld devices are designed to meet the needs of information-intensive
industries, in which mobile users need to access and manipulate large amounts
of data. Founded in 1990, the company is headquartered in Sunnyvale,
California. The company's target markets include: Utilities,
Healthcare and Transportation. The
newest product introduced by General Magic is Serengeti, a virtual
assistant that keeps you in touch anytime, anywhere.
General Magic offers products and services to make communication and access to information easy and convenient for people, no matter where they are. The company uses user-interface technology to provide simple ways for people to communicate, conduct business and keep track of their lives.
"A few years ago, General Magic fell on hard times when Magic Cap, the operating system it had developed for handheld computers, failed to win customers over. The company attributed Magic Cap's tepid reception to business people's perception of PDAs as unnecessary and burdensome devices. And, around the same time, the Net grew up, changing market conceptions of mobile computing." (John Marshall, The San Francisco Chronicle, 6/27/97). With the introduction of Serengeti, General Magic could be poised to make a significant comeback with a base technology that is over one hundred years old: the telephone. Workers can dial up the Serengeti system to check email, route calls, access information, change appointments on shared calendars and even get up to the minute reports on traffic conditions. What makes the Serengeti system unique is its Voice User Interface which does away with annoying menu based choices and allows the user to interrupt and redirect the commands as if he or she were talking to an assistant.
General
Magic is based in Sunnyvale California and has lost over $119 million since
its inception 1990. The company went public two years ago in the
middle of the hype concerning high technology internet companies.
The stock price raced from a high of $32 on its first day of trading to
a low of $1.38. However with a new CEO in place and a new product that
will earn revenues for the company from sales of a product rather than
through the licensing of a technology, General Magic may be on the road
to recovery.
Other Examples of Agents In Use or Development for EC
1) Jango : Recently purchased by Excite for $35 million, NetBot, Inc. created this intelligent agent software product which is geared toward online shopping. Jango searches the Web for a product the user describes and presents the results in terms of location, product reviews, and, where appropriate, pricing. It can also prepare an order form and then hands the customer off to the merchant.
2) NextPoint's "S3" : Software package coming in December which uses intelligent agents to gather information such as response time and availability data, from networks and applications, to help businesses better manage networks.
3) Netmosphere, Inc. : Providing agent software to businesses that track work flow, issue reminders about deadlines, and identifies bottlenecks.
4) Kinetoscope 's "VIA" (Versatile Intelligent Agent): Can execute assigned search and retrieval tasks across the internet and private intranets, enter databases, and dispatch e-mail, paging alerts, faxes, and more. Complies with Java Development Kit 1.1.
5) IBM's Web Browser Intelligence (Webby) : Software agent helps user remember a web site visited previously, search previously viewed sites by keyword on titles or full content of the site, provides alerts to speeds of links, advises users of changes at web sites, rank order viewed sites by frequency and how recently they have been visited, and learn user patterns and suggest shortcuts.
6) Datalink Systems Corp.'s SplitXpress : Uses intelligent agents which continuously monitor financial newswires and immediately alert users of stock splits via alphanumeric pager or PCS phone.
7) Guideware : Has embraced a mobile agent technology in its new software development kit that lets developers program components to move independently from one computer to another. Once launched by a user, the agents can travel through a network performing different tasks in a variety of hosts.
8)
Brightware :
Sales
server that includes intelligent agents to "accurately understand your
customer's free-form messages and input to Web forms, diagnose their input
as needs, match those needs to prodcut features and benefits, and lead
them down the path to a purchase." A truly bright star on the frontier
of electronic commerce-facilitating agents.