Touching hands
5.3(Intelligent) Software Agents meet The World

Back to the previous page Previous page

To the Table of Contents of this chapter Chapter ToC

To the Table of Contents of this paper Paper ToC

"Recent trends have made it clear that software complexity will continue to increase dramatically in the coming decades. The dynamic and distributed nature of both data and applications require that software not merely respond to requests for information but intelligently anticipate, adapt, and actively seek ways to support users. Not only must these systems assist in coordinating tasks among humans, they must also help manage co-operation among distributed programs.

In response to these requirements, the efforts of researchers from several different fields have begun to coalesce around a common broad agenda: the development of software agents. On the one hand, researchers from the fields of human-computer interaction, intelligent and adaptive interfaces, knowledge acquisition, end-user programming, and programming-by-demonstration have concerned themselves with the implications of the agent metaphor and its concrete representations, learning and adaptivity, explanation, agent authoring, and other aspects of interaction between humans and software agents. On the other hand, researchers in the fields of distributed artificial intelligence, robotics, artificial life, and distributed object computing have contributed expertise in the areas of negotiation and planning, situated action, agent-to-agent protocols, concurrency, and component-based frameworks."

Where techniques and concepts as Ubiquitous Computing have a strong potential for usage in the real world, software agents have most potential in the virtual world. In this world, they will relieve us from mundane tasks in the short term, and may assist us in smart - and possibly truly ‘intelligent’ - ways in the long term. They will perform those tasks that are best done by computers (e.g. as the tasks are laborious or boring) and will take care of those things that humans should not need to bother with. We have discussed this topic to a large extent in chapter two and three.

Apart from the rather technical aspects, there are other issues that arise when we really (i.e. massively) start to work with agents and agent-like applications.
One such aspect is the social side of employing agents and ‘agency’. Agents will interact and collaborate with not only other agents but also with humans. This will make them part of our environment, and therefore subject (to some extent) to our social rules: the agents will become embedded in our lives, as a result of which agents will have to follow social rules because if they do not, people will have a hard time accepting them (at least to accept them as assistants or as something you can work or co-operate with productively).

"Work in artificial intelligence has never really addressed the problems of binding together its agents in human societies to the same degree as has the field of human-computer interaction. In AI agents are designed to form unrealistic social systems, or, rather, they take valid models of realistic social systems and interpret the models too literally and too strictly. The human components of conflict, morality, and responsibility, for instance, are all simplified out of existence and, therefore, agents have real problems in human societies, except in small niche contexts where people can accept these limitations. The result is that agents are not usually flexible enough to be able to work effectively in human societies."
from [WATT96]

This means that with regard to agents and agency, there are not only all kinds of technical issues that will need to be addressed, but also more ‘soft’, social ones. They will not need to be tackled all at once, as humans are able to adapt to some extent too, but "at the end of the day, we must make this shift for agents to become more than yet another temporary technological innovation".
To make this shift, technology will not be the most important key to success: people judge an agent not by what is inside of it but by the actions it initiates and performs, by the behaviour it is able to display. For instance, Eliza is really nothing more but a collection of simple if-then rules which fire as a result of a person stating its problem to it. However, if the person using it feels like Eliza is able to help him in a useful manner, then that is all that really matters (at least to this person).

Another important part of social behaviour, is communication. Which brings us to another issue: how should agents communicate? How should they ‘interface’ with each other and with humans?
In the eye of efficiency, agent-agent communication should best occur by a knowledge representation language like KQML or KIF, but these are languages that are very hard to understand for most humans. In agent-human interaction, the most preferred language would be natural language, as this is most comprehensive for humans. Technically (for both the agents, as well as for the makers of them) this is very hard to accomplish: parsing natural language is a problem that has been the subjects of science for many years, and it is unsure if it will ever be possible to make a computer fully understand natural language.1
Note that the communication issue is not only about communication through natural language: spoken or written language is just one component of the entire communication process. There is of course also the non-verbal communication that is very important. Agents can address this issue partially by using techniques such as an anthropomorphic interface. However, when they are on the receiving end of a communication, enabling agents to interpret and understand non-verbal communication will be yet another huge challange to agent designers.

Why are these social (and communication) aspects so important in the context of the employment of agents and agency?
When agents will be used as is envisioned, they will be like assistants to us, like a human being. This will mean that people - especially those who do not have any idea of what agents are really capable of - will treat and interact with them like they would (with) any other human being;

"This very kinship opens up an immense possibility for conflict when there is a dissonance between these expectations and reality - when the expectations from human-human collaboration conflict with the reality of human-computer interaction."
from [WATT96]

Sounds far fetched? Look at how people use computers and applications, and how they unconsciously attribute all kinds of human qualities and skills to computers (e.g. ‘My word processor trashed my document’ or ‘The computer did not like it when I did...’). Humans do this naturally (e.g. understanding their own and other’s mental states, goals, beliefs, and so on); it is part of the glue that holds a human society together. For agents and agency to be accepted, they must be able to display similar behaviour and be able to reason and communicate in a similar fashion.
Realising such behaviour is not something which may be expected to emerge and become available in the next few years; it will probably require decades of research to get anywhere near this goal. In the short- and medium-term, agents will therefore possess behaviour that is only a far cry from the behaviour as described above. This does not mean that such agents cannot be of good use to people, but rather that they will be perceived as a smart piece of software instead of the smart assistant they were said to be.2

1= On the Software Agents Mailing List, an intriguing future was envisioned, where agents (or rather software-based artificial life forms) had been created that used a communication language which was optimal for agents to communicate, but totally incomprehensible for most humans. Soon, the agents were able to communicate and exchange messages and knowledge at such a high rate, that they were living in a world of their own. What’s more, the agents will find it very hard to find a reason to communicate with humans, as the agents find our natural language too inexpressive and clumsy to communicate...
2= Note that this view on agents is different from that of AI in general, in a sense that the aim is not to build progressively more complex agents, but to build progressively more human agents.
On to the next page Next page
To the Hermans' Home Page Home Page
Chapter 5 - Tomorrow's Internet: an Ubiquitous and Agent-serviced Online Market Place...? "Desperately Seeking: Helping Hands and Human Touch" -
by Björn Hermans