From agents@SunLabs.Eng.Sun.COM Thu Jun 8 09:18 EST 1995 Date: Wed, 7 Jun 1995 16:17:26 -0700 Message-Id: <199505230025.RAA15027@chum.cs.washington.edu> From: etzioni@cs.washington.edu (Oren Etzioni) To: kauer@pcug.org.au Subject: Re: Agents debate at Chi '95 Ray, As I see it, your message raises three fundamental issues: 1. What's an agent? 2. What technology do we have for building agents and agent-based interfaces? 3. Are agent-based interfaces desirable? I'll address the first question below, but I'd like to point out that while we're still struggling to invent the appropriate technology for constructing competent agents, this does not in any way diminish the desirability of agent-based interfaces. Just because we haven't found a cure for cancer or an AIDS vaccine doesn't mean that we should stop looking. The desirability of agents and the likelihood we will obtain sophisticated ones within the next 3-5 years should be evaluated separately. 1. What's an agent? By "agent" we mean someone who acts on your behalf. Information agents are loosely analogous to travel agents, insurance agents, etc. In the hope of demystifying the term, here is a list of characteristics that have been proposed as desirable, agent qualities. o Autonomous: an agent is able to take initiative and exercise a non-trivial degree of control over its own actions: - Goal-oriented: an agent accepts high-level requests indicating what a human wants and is responsible for deciding how and where to satisfy the requests. - Collaborative: an agent does not blindly obey commands, but has the ability to modify requests, ask clarification questions, or even refuse to satisfy certain requests. - Flexible: the agent's actions are not "scripted"; it is able to dynamically choose which actions to invoke, and in what sequence, in response to the state of its external environment. - Self-starting: unlike standard programs which are directly invoked by the user, an agent can sense changes to its environment and decide when to act. o Temporal continuity: an agent is a continuously running process, not a "one-shot" computation that maps a single input to a single output, then terminates. o Character: an agent has a well-defined, believable "personality" and emotional state. o Communicative: the agent is able to engage in complex communication with other agents, including people, in order to obtain information or enlist their help in accomplishing its goals. o Adaptive: the agent automatically customizes itself to the preferences of its user based on previous experience. The agent also automatically adapts to changes in its environment. o Mobile: an agent is able to transport itself from one machine to another and across different system architectures and platforms. 2. Technology? See http://www.cs.washington.edu/research/softbots for my favorite.. 3A. The agent metaphor: I find the argument against the agent metaphor the most surprising. I believe that most people would want to delegate boring, tedious, or difficult tasks to an agent. We see this happening in human interaction constantly. Perhaps, the argument is that automated agents will not be up to the capabilities of human agents. However, this is not an argument against the agent metaphor; rather, this is simply a variation of the argument that the technology isn't there (yet). As anyone with a capable assistant knows, both the concept and the practice of delegating to an agent is remarkably helpful. This point was recognized by many visionaries. In an article entitled "Hospital Corners," Nicholas Negroponte uses the task of making one's bed as an illustration: "Today, notwithstanding the skill, I cherish the opportunity of delegating the task and have little interest in the 'direct manipulation' of my bedsheets... Likewise, I feel no imperatives to manage my computer files, route my telecommunications, or filter the onslaught of mail messages, news, and the like. I am fully prepared to delegate these tasks to agents I trust as I tend to other matters ..." 3B. An application for agents: let me demonstrate that they exist by way of example. I believe that in the future we will see more and more users desiring mobile access to information across low bandwidth channels (e.g., from a hand-held computer) or even from a cellular phone. For such access, direct manipulation and visual browsing (e.g. Mosaic) will not work. The agent metaphor of making a request or asking a question is much more appealing. 3C. Realistic expectations: this is an excellent point. If the past is any indication, we need to be very careful with the kind of expectations we generate. Otherwise, solid technology and good research will end up with a bad name. I think that the point regarding agents "taking control of the computation away from the user" is important. While users may not be interested in the details of the computation, presumably they want to remain in control, particularly as decisions involving money are made. I view this as a technological challenge (raised by Donald Norman and others) in how to make agents 1) that leave the user in control and 2) that are able to communicate what they are doing and 3) that ask for approval on important decisions. enough.. oren