The Artificial Intelligence of Intelligent Agents

The Internet's birth and subsequent rapid growth transitioned it from a place that didn't provide enough information to providing way too much. As a result, a large window of opportunity has opened for companies attempting to provide fast access to quality content. Services such as America Online and CompuServe, with their user friendly graphical user interfaces and delivery of "Cliff Notes" style content seemed to be the one-stop-shop for what you are looking for. But, the broadcast style of these content providers unavoidably filtered out information that some people wanted access to. Realizing this pitfall, search engines such as Yahoo! and WebCrawler emerged to help solve the needle in the haystack problem by making it easier to find the needle. But, as the digital haystack grows, searches are rendering more hay than needle.

Intelligent agent technology companies are hoping to succeed by offering much more focused information than their predecessors were capa­pable of providing. Born out of artificial intelligence research centers in schools such as MIT and Cambridge, intelligent agents boast the ability to autonomously roam the Internet, making calculated decisions based on user preferences. In conjunction with content providers and search engines, intelligent agent companies have realized that the challenges of data retrieval is not a supply side problem, but a demand-side solution.

Information that is not well targeted will fall upon deaf ears. By approaching the dilemma from the demand side, agenting companies are asking customers to tell them what makes the needle different from the hay. By identifing the customer's preferences, agent companies are claiming that they will be able to provide significant customized information as fast as present access technologies allow.

Categories of Agent Technologies

But, visions of intelligent agent universes interacting autonomously on their masters' behalf belie the current status of the technology. The reality is that agent companies are only just gearing up for prime time. The technology is immature, and the market is only just being identified. What is impor­tant at this early stage is the way in which compa­nies are entering the market.

We have set up a scale which helps to evaluate how intelligent agent companies will be received in the consumer market. Jupiter pro­jects that those companies with the most potential to be well-received by the consumer market will place high enough on the scale to be valuable, yet low enough to be doable; and their promise and performance will be within a reasonable distance from each other. The losers will be either too high or too low on the scale or the gap between their promise and their performance will be too great.

What all agent companies have in common is their reliance on the user's ability to identify the information they want. But, many vendors are capturing user identified information in quite different ways. In the product discussions that follow, Jupiter will assess the three types of services that are presently coming to market and their potential for future growth:

Search And Retrieval Service: These services require a good deal of interaction with the customer to aid in the identification and personalization of informa­tion.

Polling Product: These services require that a user fill out an identification form. The users input is then compared to all other user prefer­ ences already in the database, which the service uses to make a recommendation based on simi­ larities.

Profiling Service: These services compare user preferences to pre-existing databases and work to identify what other information the user is likely to find valuable. Such services are poten­ tially commercial dynamite because they could vastly improve on the effectiveness and costs of . mail order marketing.

Business Determinants

No matter what approach agent companies take in capturing and acting on customer data needs, Jupiter has identified seven business determi­nants or hurdles on the race track to success­ that they will have to clear. These are as follows:

• Target. Companies that decide to offer their agent technology as a stand-alone application on the consumer's desktop may lose their audience to easily accessible server-end companies.
• Revenue. Companies that expect customers to pay for their software will be disadvantaged in comparison to advertisement or investor driven competition.
• User Interaction. If it requires significant user interaction, a vendor's agent may not be the time saver the customer seeks, even if the agent is valuable and performs as promised.
• Reliability. Agents of companies that search for customer-identified information will have to make sure they have (and can keep) access to all the data being searched.
• Browser Integration. As a result of Microsoft's and Netscape's efforts to outdo each other, each is pushing its client into an integrat­ ed environment. Companies that sell stand-alone applications, or rely on off-line data resources, may need to rethink their strategies to allow for such changes.
• Dynamic Capabilities. Agent companies that do not tackle the information explosion, but instead rely on narrowly defined resources, will need to find a way to update and expand their data.
• Marketing. Because agent technologies are immature, the way in which companies choose to market themselves will be important.

The Players

Search and Retrieval: Both Autonomy Inc.'s Agentware and NetAngel's The Angel are two user-managed products that Jupiter thinks have a lot of potential for growth. Their prod­ucts are by no means trivial or unrealistic. With enough user interaction, a narrowed search will eventually return quality information. But, The Angle and Agentware both perform poorly on the hurdle test.

Some of the most important hurdles that these products do not clear are target (client-side only), user interaction (heavy maintenance), and integration (client only applications may be poorly positioned for an integrated desktop). What may separate NetAngels from Autonomy over time is hurdle two: revenue. The NetAngels' service is free due to heavy investor support and can be downloaded to the desktop. To be competitive with non-charging server-side companies for which consumers do not pay, NetAngels may want to consider providing a server-side compliment to its product and Autonomy may want to reconsider its target and pricing models.

Polling Product: The Firefly Network is a serv­er-side service that requires users to fill out a questionaire that evaluates their movie and music preferences. The service then makes recommendations by comparing user interests. Firefly also enables people on the network to contact each other. The service clears all of our hurdles. It does exactly what it says it is going to do and, as a result, its promise is matched by its performance.

Firefly's decision to initially provide only music and movie options could be seen as limiting in comparison to what the Internet has to offer. But, we feel that Firefly's decision to focus on these topics in the beginning stages delivers a clear, uncomplicated product. The entertainment industry, besides being perennial, is knowable ahead of time. And what is knowable ahead of time, is serviceable on time. In these preexisting markets of known behavior and pre­dictable content, the needles are all identifiable. That is a big plus. As the Firefly Network con­tinues to grow and gain consumer acceptance, it is likely that we will see Firefly expand its services into other markets.

Profiling Service: BroadVision's The Angle makes recommendations by comparing user preferences to pre-selected informational databases. We believe that hurdle four (iffy access to information) and hurdle six (dependence on static resources) are two possible obstacles in BroadVision's road to consumer acceptance. BroadVision's The Angle depends on information provided by static database companies, such as Oracle and Sybase, to provide their service. While access to database information may be dependable in the start-up stages of new information retrieval technologies, database companies may displace companies such as BroadVision with their own services.

However, we feel that as long as proactive cus­tomer profiling services position themselves well, and have appropriate non-exclusive reuse licenses, these services shouldn't depend on owning anything more than the customer's preference profiles. Companies such as BroadVision should exist in symbiosis with database companies and mail order vendors as a value-added service enabling businesses to reach potential customers.

One of the first companies developing intelligent agent technologies for the commercial marketplace was General Magic. This company is a great example of a business that placed way to high on the trivial/unlikely scale when it was starting out. Former chief executive Marc Porat lost a considerable amount of venture capital support (and eventually many top level executives) when the company's perfor­mance was not living up to its promise of an agent driven world.

A likely reason for the Silicon Valley company's troubles was the birth and growth of the Internet. General Magic's claim that its Telescript language (the foundation of its agent tools) would enable users to request and recieve information within a private network lost steam when widespread adoption of the Internet put a damper on proprietary networks.

The Shape of Things to Come

By clearing most (or all) of the hurdles that we discussed above, some intelligent agent compa­nies will be on the road toward lasting acceptance of their products. Intelligent agents are too inherently sensible not to find a market. These companies are all trying to service other companies on their customers' terms - individ­ual by individual. But, the easy sell has not translated into easy delivery, a matter complicated by the fact that the industry as a whole is guilty of gross overpromising. And to us that says "Keep it simple."