This is the name of workshop I am attending today at the University of Southampton in the UK. It is a gathering of 20 or so survey methodologists, communications researchers, technology types, and just plain researchers—public and private sector. The basic theme is this: the general trends in the real world are toward “self-administered” technologies like IVR, voice recognition, IM, iChat, etc. and the survey world will follow suit. Interviewer administration is declining, whether by face-to-face or telephone, and this will accelerate as we find new ways to conduct surveys without interviewers. In some sense, that’s what the Web stampede has been all about.
One obvious direction is the use of avatars, automated interviewers, virtual humans or whatever you choose to call them. Using the same technology that is used to translate the movements of LeBron James into a video game representation researchers can now build talking head interviewers who look and behave almost like real people. Out of that comes a whole lot of interesting issues around what these guys should look like, how should they behave, and how respondents react to them.
This is pretty futuristic stuff and I frankly have been skeptical about near terms uses of the technology. But as the workshop has progressed I have come to see the problem in ways that can worked on now and implemented sooner than one might think. I now see two aspects to the problem: the interface and the functionality.
Humans being a self-absorbed species (I guess most are) we tend to think that the ideal interview is one conducted by an interviewer, a.k.a “agent” in the speak of the group convened here. And so we get distracted by the problem of representing something that looks and acts like a real human being on a computer screen. But at least in the short-run I view the interface as a secondary problem. The real challenge we can work on now is the functionality issue. In a little qual study done at UM that systemically varied the degree of natural movements in the agent and its ability to deal with respondent problems and cues of respondent difficulty (like taking too long to answer) they found that the latter characteristic had a stronger impact on the accuracy of responses than did the former. In other words, the most important characteristic of the agent was its ability to deal with respondent issues rather than the naturalness of its appearance. To put it another way, functionality trumped the interface. More later.