AI on the DSM IV – A Thought Experiment from David J. Kelley

With the recent news coverage of scientists discussing robot uprisings and the possible dangers of artificial intelligence, it’s interesting to see a direct thought experiment along these lines from Microsoft UX developer David J. Kelley. In a recent h+ Magazine article, Interview with an AI (Artificial Intelligence) – A Subtle Warning…,  Kelley provides an outline for an experiment that seeks to gain some understanding of how an AI would respond during an interview. As he explains it:

“I was thinking about ideas for an article on my train ride home from the experience lab I work in, and it came to me that it would be interesting to actually have an interview with an AI only a little bit better than us, maybe one that is one of the first kinds of true AI and for fun let’s say it has lived with us for a few decades incognito. But how can we do that?

With the help of a few people in the computer science AI field and mental health professionals it turns out that it appears we could produce a psychological profile for such an AI and from that potentially a method for extrapolating what such a AI might say or how it might respond to a given line of questions, under certain circumstances, with certain assumptions put forward to develop said profile. Before you ask, it also appears that based on the DSM IV it turns out that this AI would likely be diagnosed with potentially a number of disorders including ASD (Autistic Spectrum Disorder) and/or Asperger’s. I find it interesting that a computer intelligence would fall into that category out of the gate.”

One of the difficulties in trying to conduct and interview of this kind is obviously we have no access to “true” AI at the moment. Kelley found a novel way around that through the proposed AI’s psychological profile.  Finding someone who matched the profile, Kelley conducted the interview and adjusted the answers so that they pertained to the theoretical perspective of the AI.

“To start with, I looked at likely qualities of an AI modeled on the human mind but digitally replicated and after talking with mental health professionals in the field it became apparent that it might be likely that such ‘minds’ would have some areas in which they are better than real humans due to the nature of the hardware they are running on. It was felt that it would be easy enough to enhance such a digital mind with elements of more traditional computers, such as hyper focus, improved computational skills and that we might not get everything perfect so we might lack for example rich emotional abilities.

Given that I looked for a person (human) that matched that psychological profile as much as possible which included: above average intelligence, borderline Asperger’s and ASD (Autistic Spectrum Disorder), deep technical skill, ability to hyper focus etc. After explaining to the subject what we wanted to do I did the interview based on questions from various people and then tweaked the language as little as possible so it would be from the AI’ standpoint and not from a human standpoint but this required very little editing. Which in and of itself, seemed the most interesting fact out of the process. “

While not perfect by any means, these kinds of thought experiments can help to guide us towards more in depth investigations into the existential risks and rewards that might come from advanced technologies like artificial intelligence. Kelley plans to continue the project by using the question set to interview more people who fit the DSM IV profile for the theoretical AI, along with a control group of interviewees with different psychological profiles. To read the interview, and more about what Kelley hopes will come from the experiment, head over to h+ magazine.


, , , ,

  • alizardx

    “maybe one that is one of the first kinds of true AI and for fun let’s say it has lived with us for a few decades incognito. But how can we do that?”

    An AI that’s lived in a human community for a few decades and hasn’t managed to download a copy of DSM-IV, done extensive reading on associated literature, and hasn’t taken any of the free DSM-IV tests online, and therefore is incapable of gaming DSM-IV tests to give any test results which it believes to be advantageous to itself or which would provide itself with lulz. What an interesting notion.

    “Microsoft UX developer David J. Kelley.”

    One of the people who developed Metro for Win8? Or the version of that UI running on the new Windows phones people are also avoiding? People who consider the referenced article an example of thinking should dump their Apple devices for Win-crap.

    It figures. I suspect that we are too far away from the hardware to support sentient AIs to even speculate usefully about them.

  • BuzzCoastin

    the machines are definitely more intelligent than humans
    but that’s not saying much
    since even cockroaches tend to be more intelligent than humans

  • DeepCough

    How could you diagnose an “artificial intelligence” with a very human problem like “autism?”

    • alizardx

      The same way one would diagnose a unicorn or Bigfoot, I guess. I’ve compared current discussion of AI with discussion of the future of aviation circa 1880. Anyone who wants to know how well those projections went is invited to go to the nearest major airport and count airships.

      I think I was being entirely too generous. There are so many extrapolations and fragile inference chains in this article that I’d call it “not even wrong”. Reminder of why I don’t use H+ as a source for information. And why I won’t be buying Win 8 or RT/Surface smartphones and tablets.