Tuesday, September 30, 2008

AWOL-ish and thoughts on experimental design

I know posts have been sparse here (the Digital Trends people haven't heard from me in months, really) but like usual, I am being overly ambitious on large projects and having to justify it while secretly plotting even larger ones.

Really, it comes down to having left the research world for 4 years, I expected some basic experiments to be completed. They're obvious, elemental, and simple, but act as cornerstones to further development. They weren't. Honestly, when I returned to academia, I didn't know what kind of project to work on and after a few talks with my advisers, it became clear that next to nothing I perceived as the logical next steps was accomplished. I actually remember the exact moment when it hit me - during a meeting when I was asking for thesis project suggestions. I recall literally stopping, staring off across the room, and thinking that either a) it was really a verify this in greater depth type of idea or b) Wow.

It isn't 'bad' - the push has been towards taking state of the art techniques and philosophies built from monkey work and applying it to the human side. Makes perfect sense. The problem is that there should always be a parallel thread building on the basic/elemental side of the equation. Working in people is different from monkeys (obviously), but the two biggest differences, and it takes some serious hammering to get them into your skull, are that you have the ability to convey concepts (not just processes) through language and you have access to unobserved processes. That means:

a) No training is needed for task familiarization (not talking procedural memory or 'mastery')
a1) Replication of monkey experiments should be performed at breakneck speed.
(Erm, unfortunate pun, but I leave it in because it correctly conveys the idea. Just noting that I did catch that.)
a2) Papers based on replicating monkey experiments require less effort (tasks already coded, expected results and caveats already mapped) and report less significant findings (i.e. - they become "Worked in monkeys. Yup, works in people." comparative studies).
a3) Should require either a novel functional outcome ("We can therefore do X with a BCI.") or provide insight on a monkey-impossible phenomenon ("Think about grandma's apple pie while you do X.").

b) Almost any experience once relegated to the world of 'cognitive neuroscience' becomes admissible. I know what it means to "recall the scent of a new car", so do you. Most of us do. There might be significant variance in what that scent actually is, but in understanding 'the brain's response to imagined new car smell', we all know the subject/object of reference. Take out these accepted experiential elements because of the artificial distinction in neuroscience sub-fields, and I would argue that ascribing any neural activity to any unobserved process (preparation, reward, saliency, etc.) must be eliminated, too. Sorry, no ascribing unobserved forces to activity, since you have no ability to say with complete certainty that monkeys operate in a conceptual/cognitive manner similar to humans. You're just anthropomorphizing mental or neural processes based on statistically significant relationships in a constrained environment, and you're dissociating environmental cues, not internal representations. Does that colony of monkeys happen to be full of synaesthetes? How many times do you hear, "Yeah, the monkey just didn't want to work today"? Or, "Yeah, the monkey got bored."? So, the data is collected only when the monkey "wants" to work?

So, in short, human BCI experiments should incorporate 'cognitive neuroscience' techniques in order to enrich the functional/translatable significance of the outcomes. I love to use the example of crossing the street. A BCI user should be able to recall crossing streets, imagine crossing the street in front of them, tell a joke about a particular chicken crossing a street, point across the street, and want to get to the other side of the street without stepping into traffic. Otherwise, we're doomed to becoming glorious reflex arcs.

(And for extra credit, imagine a world in which Sartrian angst is always acted on. Scary.)

Anyhow, I think that the state-of-the-art and elemental streams need to always be examined in parallel (even better in concert, but tougher to design and control). And, having already spent too much time typing this, I'll leave to you to consider why that is (hint: there are practical, moral, and theoretical reasons). Enjoy the brain twisting. :D

No comments:

Post a Comment