![[INTERACT97 Logo]](http://i97.syd.dit.csiro.au/i97/logob.gif)
In this address I will re-examine this argument in light of reflections on the past ten years of developments in computing and HCI. I will suggest that while interesting developments have unquestionably emerged during that time, critical issues remain. These issues have to do with the mutual intelligibility of people and machines, and the relations of mutual intelligibility to participation in a social world not reducible to discrete abilities and skills. The intractability of such problems recommends a research agenda aimed less at the design of interactive machines than at an understanding of the distinctive dynamics of working with computational artifacts, and at their artful integration with the rest of the social and material world.
Note: Scanned from paper copy of speakers notes, may be transcription errors. Items in bold are captions from slides (slides not included). Items in italics are speakers hand-written changes to notes. Spell checking of American English with Australian English dictionary may have introduced additional errors. - Tom Worthington
I'm starting off today from the observation that The growth of the internet has brought with it renewed interest in the idea of software agents, "knowbots," and other computational artefacts attributed with a capacity for intelligent, interactive behaviour. It was ten years ago that I published my first attempt to characterise what I called then "the problem of human-machine communication." My central aim in articulating the problem in the way that I did was to suggest that the challenge of interactive interface design is actually a much more subtle and interesting one than what it was assumed to be by my colleagues in HCI at the time. Basically, it seemed to me, their assumption was that computational artefacts just are interactive, in roughly the same way we are, albeit with some more or less obvious limitations. However ambitious, the problem in their view was a fairly straightforward task of overcoming the limitations of machines by encoding more and more of the cognitive abilities attributed to humans into them.
What I want to do with you today is to look again at that assumption about the problem of designing interactive machines, and my argument with it, in light of developments in the ten years since. The developments that I'm interested in are both in the project of designing interactive systems, and of theorising [within anthropology and sociology] about relations between humans and machines. What I hope is that by looking again at these projects in relation to each other, we might be able to move towards some different ways of thinking about interface design.
[AI Magazine cover]
So let me start by taking you back ten years (or actually more like 15), to the early 1980's at Xerox PARC. I'm going to proceed a bit autobiographically here, to give you a sense for how the problem of interactive machines first presented itself to me, as an unsuspecting graduate student in anthropology arriving at PARC in 1979, interested in writing a dissertation on work and technology design.
My colleagues at PARC in the early 1980's were leading proponents of what was at the time (and as far as I know still is) called "knowledge representation." Most generally, the project was to develop formalisms that could encode the corpus of "common sense knowledge" taken to be foundational to human intelligence. Closely related to this enterprise were efforts to encode "goals" and "plans" that, when executed, would lead an artificially intelligent machine imbued with the necessary knowledge to take appropriate courses of action.
Around 1981 or so, a delegation of mid-level Xerox managers travelled to PARC to report on a problem with a recent product and enlist research advice in its solution. The product was a relatively large, feature-rich photocopier that had just been "launched," mainly as a placeholder to establish the company's presence in a particular market niche that was under threat from other, competitor companies. The product had been assembled very quickly, largely through the re-use of components from already existing products. (In acknowledgement of this less than ideal design process, the product during its development was ironically code-named it chainsaw.") While the machine had features of so-called "dedicated, high-volume" machines used in print and copy shops and operated by workers whose sole responsibility was to run them, the marketing goal was to position this as a "hallway" machine for the "casual, walk-up" user. The problem as formulated by the delegation, however, was that customers were complaining about the product that it was "too complicated."
The solution these managers proposed was to attach a video display terminal to the machine that would somehow enhance its current, paper and liquid crystal display-based instruction set. My colleagues viewed this proposed solution as woefully inadequate insofar as it failed to consider the question of just what software should underlie the, display. At the same time, they were responsive to my suggestion that a first step in addressing the question should be to explore further just what experiences were actually glossed by the reported customer complaints that the machine was "too complicated." Following up on this suggestion entailed some visits to "customer sites"; that is, places in which people were actually trying to make use of the machine in the course of their everyday work. Specifically, I and a colleague spent some time hanging around the machine in several organisations where it had been placed.
From these observations we confirmed the confusing behaviour of the machine, but found ourselves just as confused as those we observed. In order to explore in detail just what troubles people encountered in attempting to make sense and use of this machine, we managed to obtain a machine ourselves and install it in our workplace. I 'then invited others of my co-workers, including some extremely eminent computer scientists, to try using the machine to copy their own papers for colleagues. To make this nostalgic tour complete, I want to show you a video that I produced for John Seely Brown, for his keynote address to CHI in 1983, titled "When User Hits Machine."
A note on the second segment, of two of my colleagues using the machine. You'll see a small inset picture in the lower left corner, which is a view of the machine's control panel. And when they seem to be gazing at the wall behind the machine, they're consulting a poster of instructions, provided by Xerox as a further aid to customers.
[VIDEO]
From looking closely at actual encounters with this machine, I began to develop the idea that its obscurity was less a function of lack of technological sophistication on the part of its users, than of their lack of familiarity with this particular machine. I argued that the machine's illegibility was tied not to any esoteric technical characteristics, but to mundane difficulties of interpretation, or interpretative flexibilities, characteristic of any unfamiliar artefact. My point was that reading a new artefact is an inherently problematic activity. This reframed the problem of user interface design from creating a self-evident machine, to writing one that is readable (with all the problematics that writing and reading imply.) Moreover, I wanted to argue that however improved the machine interface or instruction set might be, this would never obviate the need for learning on the part of prospective users. This called into question, then, the very viability of marketing the machine as self-evidently easy to use.
What such marketing typically does is to reassure purchasers (generally middle-managers) that their workforce will be able to incorporate the new machine's functionality without any accompanying drop in productivity, while leaving front- line workers to cope with somehow mastering the machine's operation while being given no resources (time, money or experienced others) with which to do so. In the case of this particular machine, that strategy foundered insofar as there was a sufficient gap between rhetoric and reality that prospective users resisted loudly enough for the company to feel the consequences. Rather than being a unique or unprecedented incident, however, I believe this case simply made visible a pervasive dilemma for office workers faced with new technology, being distanced from the sites of its production and denied resources for its appropriation into their particular circumstances of use.
My colleagues meanwhile had begun a project of designing an "intelligent, interactive" computer-based interface that would replace the machine's existing instruction set. Their strategy was to take the planning model of human action and communication then prevalent within the Artificial Intelligence research community as a basis for the design. My dissertation became a close study of a second series of videotaped encounters, by diverse people including office workers and, again, eminent computer scientists, now with their prototype interface. While the troubles that people encountered in trying to operate the machine shifted with the use of this "expert system" as an interface, the task seemed as problematic as before.
My analysis located the problem of human-machine communication, in sum, in continued and deep asymmetries between person and machine. My concern was, and still is that while the language of interactivity and the dynamics of computational artefacts obscures these asymmetries, users inevitably rediscover them in practice. This has led me to question the aptness of the metaphor of interactive technologies, and to wonder what other metaphors we might usefully employ.
[Show transcription template]
To try to understand those troubles better, I developed a simple transcription device for the videotapes. [Describe comparison of my analysis to the machine's, the development of my transcription template to locate disjunctures, and my analysis of them.]
[Microsoft Bob, Ultra Hal]
But before we get to that, we need to look briefly at what's happened in the world of interactive artefacts since the 1980's. I'm going to be concerned here not with "interaction design" broadly defined, but specifically with those projects that continue to position computational artefacts as knowledgeable, intelligent and in other ways cognisant of and responsive to their users' actions. So for example Microsoft's discovery of this (to my mind old) idea was loudly proclaimed a few years ago in the personage of Bob, and more recently of "Ultra Hal" (whose purpose, however, seems more just to engage in inane "conversation" with us.)
[Software agents e.g. Pattie Maes]
The dominant form of this project today is the promotion of computational agents that will serve as a kind of personal representative or assistant to the user. The idea of personal agents was animated perhaps most vividly in the form of "Phil," the bow-tied agent in Apple's video 'The Knowledge Navigator." But more modest implementations of this fantasy now abound in the form of "knowbots" and other software agents. One of the best known proponents of such artefacts is Pattie Maes. Here's the first of a series of web pages showing a talk that she gave at the "Doors of Perception" conference back in February of 1995. I'd love to read the whole thing to you, but in the interest of time i'll just point out a couple of things. Maes assures us that:
"the home of the future' will be half real, half virtual, and that: the virtual half of our home won't just be a passive data landscape waiting to be explored by us. There will be active entities there that can sense the environment ... and interact with us. We call these entities software agents.
Agents are personified as cartoon faces, and attributed with capacities of altertness, thinking, surprise, gratification, confusion and the like. [And she goes on:
just like real creatures, some agents will act as pets and others will be more like free agents. Some agents will belong to a user, will be maintained by a user, and will live mostly in that user's computer. Others will be free agents that don't really belong to anyone. And just like real creatures, the agents will be born, die and reproduce... I'm convinced that we need [these agents] because the digital world is too overwhelming for people to deal with, no matter how good the interfaces we design ... Basically, we're trying to change the nature of human computer interaction ... Users will not only be manipulating things personally, but will also manage some agents that work on their behalf.]
I could spend the rest of my time on a rhetorical analysis of her passage. But for our present purposes I'll simply note the kind of rediscovery of the interaction metaphor evident here, albeit now, with the internet, the computer other with whom we're to interact has proliferated into populations of others - agents, assistants, pets - whose reasons for being are to serve and comfort us, to keep us from being overwhelmed, in the future workplace/homeplace of cyberspace.
As I say, I'm in danger of spending the rest of my talk on this topic, proliferating examples, but I need to move on. For those of you who want more, I refer you to the 1994 CACM special issue on Intelligent Agents, which includes an interview with Marvin Minsky, another article by Pattie Maes, and much more. And I suspect many of you can add your own favourite examples.
Somewhat paradoxically, I think that it's at least in part the persistence of the old modernist human/machine divide that makes machine agency so compelling. Without this we seem to be worried that machines will remain "lifeless," and by implication, less. What appears constant is that the litmus test of full humanity is the artefact's capacity to be , autonomous, on the one hand, and just what we want, on the other. We want to be surprised, but not displeased.
[the view from Technoscience Studies]
Let me turn now briefly to recent developments in social studies of technology that I believe are relevant. The last decade has seen the emergence of what are called within anthropology and sociology "technoscience" studies (to emphasize the increasingly hybrid relations of technology and science.) A central agenda for technoscience studies is to displace traditional dualisms in our conceptualuzation of relations between, for example, humans and machines, the social and the technical, the natural and artificial, and so forth. Some of the best known and most provocative proposals for how to do this are historian of science Donna Haraway's discussion of human-machine hybrids or cyborgs, and Bruno Latour and colleagues' development of the idea of non-human agency as part of Actor Network Theory. Closer to home, Australian scholars like philosopher of science Helen Watson-Verran at the University of Melbourne, sociologist Judy Wacjman at Canberra, & Toni Robertson here at UNSW in Syndey have drawn on recent feminist theorising to frame questions of knowledge and technology in new and extremely productive ways.
[Humans-machines as hybrids]
There are, of course, significant differences across these scholars and their approaches. Nonetheless, I think we can summarise what are some central, and basically shared, perspectives for our present purpose. First, there's the view of humans and machines as in some real sense inseparable entities, or hybrids. Haraway's cyborg is a sign for the increasingly intimate relations of our biologies and technologies or, in Cynthia Cockburn's words, of our "technological subjectivity." As Cockbum puts it:
We do not need to have a heart pacemaker or be plugged into a kidney machine to be a cyborg. If you took away my glasses, my pen, My telephone, wordprocessor, cooker and car, let alone the medical, nutritive, transport and communications systems to which those connect me, I would not be identifiable as the person I am. The subjectivity of everyone in industrialised societies, man or woman, engineer or childminder, is a technological subjectivity (cited in Green et al 1993, p. 3-4).
[Knowledge and agency as relational]
Second, and relatedly, technoscience studies locate knowledge and agency neither in persons nor in artefacts, but in relations between them. I think some of the most beautiful illustrations of this idea have come from the work of Charles & Marjorie Goodwin on the social and material bases of skilled practice.
Let me give you an quick example from a project we worked on together several years ago. Jackie sequence "info system" as skilled assembly /integration of partial info resources.
This gives us, I think, a different way of understanding the problem of attributions of knowledge and agency to machines. That is, the problem is not so much that we attribute knowledge to artefacts, but that our language for talking about knowing, whether for persons or artefacts, is a language of disembodiment and isolation. In this respect the language of intelligent artefacts reproduces and is a kind of next logical step wi@ a modernist tradition that treats separation and autonomy rather then relatedness as the mark of intelligence and agency.
[Human-machine hybrids /= sameness]
(or projection of rational individuality onto both!)
At the same time - and this is really the central point that I want to leave you with - recognising the interrelations of humans and machines doesn't mean that there are no differences, that we now should treat everything as one homogeneous mass. The problem rather is how to understand differences differently.
[Design/use as the writing/reading of dynamic artefacts]
With that view in mind, let me conclude with the question of what a design practice based in the premise that humans and machines are interdependent but different could look like. The direction that I'd like to propose, or endorse insofar as some in the design community are already headed this way, is twofold. First, that the problem of mutual intelligibility between humans and machines recommends a research agenda aimed less at the creation of interactive machines, than at the writing of dynamic artefacts intended to be legible, or intelligible to their users. This shift brings a rich set of resources from recent reconceptualizations of what writing and reading involve, including the inevitable uncertainties in relations of writer's intentions to readers' interpretations, and the active role of the reader in giving life and meaning to the text. And this approach encourages us to explore and articulate the particular dynamics of computational artefacts, and what new possibilities those dynamics afford.
[Design/use as artful integrations]
Second, the dynamics of computational artefacts extend beyond the interface narrowly defined, to relations of people with each other and to the place of computing in their ongoing activities. System design, it follows, must include not only the design of innovative technologies, but their artful integration with the rest of the social and material world. The value of artefacts on this view lies less in their intrinsic features, than in their contribution to particular social-material landscapes. I've suggested in the past a new form of design award consistent with this view. Rather than awards being granted for free-standing artefacts floating 'M white space, awards would be given for the most artful integration of what might even be fairly banal technologies into a complex environment of use.
I'll talk more about this proposal in a paper this afternoon, but for now let me conclude with a brief view into the approach that my colleagues are taking.
[artful integrations slide]
Here's an overview slide that I've used to convey the program of my own research group to my colleagues at PARC.
Recent examples are a project in a law firm, currently, in a state civil engineering agency. If time: developing a prototype electronic file cabinet, supporting multiple search strategies including document images and text.
[What I am proposing is not some holistic approach in which subject and object reunite into some apolitical relativized whole, but a theory which insists on the importance of constructed boundaries and also the necessity of interrogating and refiguring them. K. Barad, "Meeting the Universe Halfway"]
In the end it's not that I want to prohibit uses of the figure of interaction to describe our relations with computational artefacts, so much as to heighten our degree of critical awareness. I do think we need to recover certain qualities - interactivity" intelligence - from trivialization. And that we need to restore authorship, and thereby accountability, to our relations with artefacts. As feminist physicist Karen Barad argues, what we need is a simultaneous account of the interrelations, the intra-actions in her terms, of humans and artefacts and of their asymmetries and differences. The point is not to have the price of recognizing the agency of artefacts be the denial of our own. Agency - and associated accountabilities - reside neither in us or in or in our artefacts, but in our intra-actions. The question, following Barad, is how to intra-act responsibly and productively with and through them.