Via CogSci Librarian: IM IN UR COMPUTER GIVIN U COGNTIVE DISXONANCE.

Attributed to “Brian” via the LOLCAT Builder

PS That was via CogSci Librarian. A cognitive science library! That’s new. Cool, can I be the next one?!

The great scholar Daniel Dennet recent had some serious heart trouble that required he undergo surgery wherein his heart was stopped completely. He has written about the experience and reflects on famous deathbed epiphanies that result in atheists succumbing to religiosity. He admits having an epiphany of his own but remains an atheist.

Dennett’s fans will find it a interesting quick read.

I just had to write about my recent experience of spell-checking a paper I am writing on embodiment and cognition. In my paper I argue that the embodied perspective on cognition takes findings from neuroscience seriously, but that classical cognitive science dismisses them as mere implementational issues and thus not of value in explanations of cognition. So, I’m really boosting neuroscience where it is otherwise being disrespected.

Then, I perform a spellcheck in openoffice (my word processor) and openoffice replaces all instances of “neuroscience” with “pseudoscience.” Apparently, openoffice does not have the word neuroscience in its Canadian English dictionary and the closest it could find was “pseudoscience.” Too damn funny.

Yeah… OK. I fully expect that nobody but me will find it funny.

I spent most of my week attending Access 2005, a Canadian library technology conference which was held in Edmonton, Alberta. I also spent one day at Netspeed 2005, also a library technology conference but with a different audience held immediately following Access 2005. This is my commentary. I apologize for the lack of coherent structure… I’m under time pressure but want to say what I’ve got to say. Executive summary: I observe “small parts loosely coupled” as a consistent theme of many presentation and advocate looking for theory behind the practice while optimistically pursuing all the mechanisms, solutions, and technologies that we enthusiastically presented at Access.

Access was a truely remarkable conference and I regret that I was only able to be there for about 3/4 of all the presentations. The presentations were diverse and complimentary. I will describe a set of common themes that I observed that, based on my conversation with other participants may be seen as unusual. I believe what emerged through the presentations was that library technology is striving to capture emergence. Looking at the conference through “emergence colored glasses,” the highlight of the conference was a presentation by Humanities Computing professor Stan Ruecker on content-rich browsing interfaces.

The audience was much in awe of Ruecker’s interface examples. Peter Binkley’s comments after the presentation captured the enthusiasm that I think was the concensus. He said something like, “Your interfaces are like choclate and I just want to smear them all over my own work.” (By the end of the conference I anticipated with great delite every occaision when Peter approached a microphone… he is blessed with remarkable wit). Ruecker showed us several examples of his work and the common element in all of them was presenting the user with an insane amount of data.

The first example was a pill database. The idea of the appliation is to let users get information about the medication they take by first visually identifying the pill. The interface we saw presents users with a screen with hundreds or thousands of pills. He made a strong case for how and why that interface worked. Ruecker then showed many other interfaces that had similar features. The interfaces were innovated and generated a lot of enthusiasm as Peter’s comment demonstrates.

My enthusiasm, however, is not for the specific interfaces or their domain-specific success. I was impressed by the one slide where he explained WHY these interfaces should work theoretically. He brought into the discuss J.J. Gibson’s theory of affordances. Affordances, recognize that agent(user) and environment are coupled together and that environment and the environment represent opportunities for action.

In this case the idea is that interfaces “afford” actions to their users. Why present an insane amount of information to the user? Won’t that confuse them? It will not confuse them it the interface affords some action. In this case the interface contains an insane amount of information and controls to change how it is visualized or to reduce the amount of information (both of which are domain-specific changes). This means that the interface affords opportunties to improve or refine it.

In contrast, one might argue that simple interfaces like google’s afford very little. While the impose a small cognitive load on the user the provide little or no opportunity for action. This theory is nice because it explains both why Google works and why Ruecker’s interfaces work… with one theory.

Let me say, I’m not a fan of affordances, but I am a fan of theory in science and psychology. In this case I have no objection to invoking affordances as an exploratory mechanism but I would like to go beyond Ruecker’s explanation however and bring in the concept of embodiment. Embodiment, like affordances, acknowledge’s that the user/agent is coupled with its environment. Embodied theories of behavior explicitly state that “cognition is for action.” That is, we perceive so we can act; not so that we can think. If thinking is happening, it is a separate parallel process. Embodiment also argues that we solve many symbolic problems, not by thinking about them, but by exploiting our environment. In short, a complex environment makes us smarter.

So embodiment says,

1. if we are inseparable from our environment,
2. and if our brains are hardwired so that perception result in action,
3. a user will repeated consult the environment for potential actions in loose loops,
4. thus providing opportunities for action will result in “smarter” behavior

Allow me to say, that this is of course, an extreme dumming down of this type of explanation. I recommend Andy Clark’s book “Being There: Putting Brain, Body, and World Together Again” or Nuenez and Lakoff’s “Where does Mathematics Come From?”.

This brings me to loose-coupling and emergence. Many of the presentations at Access extoled the virtues of loose-coupling. It is taken for granted now that, in software design, we want “small parts, loosely coupled.” The emphasis on web services, REST, and AJAX all result from and reinforce this idea. Such loose-coupling is seen as the right thing to do because it makes things for manageable, scalable, and flexible. If we engineer software from loosely coupled components we end up spending less time fixing things later and exploit other people’s work more easily… so there is strong reusability angle as well.

Loose-coupling is not just a popular in programming however. I see Catherine Steeves excellent (if shaky) presentation on collaboration as an effort to apply loose-coupling to social networks just as much as Gene Smith’s presentation on folksonomies was. Folksonomies exploit social networks that we didn’t know were there. Catherine advocates finding social networks (in meat-space) that we may not know are there by finding bridge-builders in our existing networks. For me, Access would not have been complete with Catherine presentation which both advocated for collaboration and warned of the dangers of over-collaboration. Essentially, the message I heard was: look for the edge of chaos and try to stay there.

The notion of loose-coupling is frequently invoked in computing science in exactly that way: the edge of choas. It helps to explain how complexity emerges from choas. And that brings me to emergence.

Gene Smith’s presentation on folksonomy was well balanced explaining how some advocate that folksonomy means that we don’t need domain experts, that we don’t need catalogers and that tagging will “just work out” in the end. Other react against that and argue that in some problem domains that cannot or will not happen. Yet, his presentation is optimistic that there is much value in folksonomy and tagging. The balanced approach he took agrees with me very much.

I see those that claim that tagging is a revolution as overly optimistic. In essence I see them as arguing that we can pick an architecture of simple components that are loosely-coupled and that we can expect the higher level complexity to emerge. They see this pattern throughout nature and are optimistic that it will happen on the web because we already have some early positive examples of interesting emergent properties (social bookmarking). I would say however, that in those examples, as in nature, we do not see consistent success of emergence from small-parts-loosely-coupled. In nature, often the “local interactions” of small part cancel each other out or result in “noise.” Similarily the results of folksonomy are not yet astonishing: I would characterize them as mildly asthonishing in some limited examples. Definately a cause for attention, optimism, and further work, but no reason to declare that we have a solution to hard problems.

It gives us reason to look as both folksonomy but also further toward ontology. The semantic web provides a tractable architecture for knowledge systems. I have read Peter recently say that he thought RDF was scary but will now look toward it. I feel that we have no choice but to look at RDF and the semantic web; if for no other reason but to explore the possibility that folksonomy may be TOO simple to build the components of loosely-coupled architecture. Historically, in part, RDF and the sematic web can trace their roots to a book by Lenat and Guha back in the 1980s called “Building Large Knowledge Based Systems.” This work describes the ideas behind the Cyc project. Probably the most interest ontology-based artificial intelligence system around. The idea is that the small parts are atomic pieces of information “stones are hard” or “math is hard.” These are loosely coupled by inference rules. So we could infer from the first two that “stones are like math in that they are both hard” and thus form the basis for metaphors used in complex ideas.

While Lenat went on to found Cycorp which applies this theory to military and big banking applications, Guha went on to work on the sematic web.

Gene Smith (I think it was him) commented at one point in his presentation that one solution to a tagging problem that came up with an application he wrote for a client might be solved by “algorithms” and someone asked what those woudl be. He said he didn’t know but it was a class of solution they might hire a programmer to invent. It might be that the class of solutions he called “algorithms” really allude to the difference between what can be handled by components that are “taxonomy/folksonomy” and those that are “ontology.” Obviously, whenever you apply a taxonomy you start to see ontology. The work of Lenat and Guhu, RDF, and the semantic web might lend some insight there.

This brings me to my last point. Synthetic psychology. If you have read Braitenberg’s vehicles you may know this terms. I know this work from the UofA’s Biological Computing Project which uses the “small parts loosely coupled” approach to in explaining cognition/psychology/behavior. The idea is similar to what we have seen but there is an emphasis on using models to discover principle crucial to explaining why a system works.

I think that is what Access 2005 impressed on me. We have good reasons to adopt some new methods and technologies. But perhaps went unsaid but that was implied is that we need to look for the reasons why these mechansims work, beyond the here-and-now. In synthetic psychology we avoid unecassarily complicated explainations by adopting this methodology:

1. Pick an architecture of simple componment and rules for there interaction
2. Design a system from that architecture
3. Look for surprising or interesting results
4. Analyse the system to discover what was critical to create the surprise/result

For programmers this is not shocking. For science it is turning things upside down. The approach is called “synthetic” because it advocated building first and analysing later. It advocates that you can learn about feature crucial to the success of a system through wise modeling choices.

Ruecker’s explanation seems to fit this method. For each problem domain a cross-discipilary group choose an architecture and designed a system. They then analyzed it to see what worked and why it worked. We can be very enthusistic about the success of the specific examples. We could say, “wow, we’ve got that pill inteface problem licked now” but we should be more enthusiastic about the method. The brilliance of this work is that is gives us a path to follow toward future success!

Oh yeah… since this has degraded into a rant… I might as well tack on my other unstructure thoughts about access. I’m sad I missed the hackfest. The food was outstanding. One fellow I had lunch with on Tuesday from Yale commented that this food was really above the norm for similar conferences. I agree. A really friendly group. Doug Poff remains the king of good questions… while the rest of us remained stunned after a good presentation he is actively seeking more info.

Robosapien quotes Andy Clark:

“The computational organization relevant to cognition is… literally spread across neural, bodily, and environmental elements.”
— Andy Clark from An embodied cognitive science?