Cognition Distributed: How cognitive technology extends our minds
Edited by Itiel E. Dror and Steven Harnad
John Benjamins Publishing Company, 2008
258 pages with references
Over the course of a decade, entire schools of science can come in and out of fashion, just like gadgets or buzzwords. Last decade, the trendy term Cognitive Science began to replace outdated terms like Behavioral Studies or Psychology or Neuroscience, and now even Cog-Sci (an obsolete buzzword) is starting to show its wear as Distributed Cognition (or DC, the trendy new buzzword) takes over. Distributed Cognition seems like an easy concept to grasp, but don’t be fooled. Cognition Distributed: How cognitive technology extends our minds presents a didactic set of essays informing us how we should think of distributed cognition in the most formalized and analytical way.
The first third of ‘Cognition Distributed’ is devoted to the task of defining what the terms ‘cognition’ and ‘distributed’ and ‘distributed cognition’ actually mean. This field is so new people actually get hung up on the semantics, but DC is just the latest term for thinking in groups. DC concedes that decision making never happens in isolation, so any system that makes decisions is confined or empowered in outcome by external factors, including people and environment and technology. So any cognitive system inherently relies on distribution, or the offloading and coordination of specialized tasks to perform a complex function; whether it be a brain, a hive of ants, a classroom, or a cockpit full of pilots. The new school of DC says the rules for analyzing all of these distributed systems are inherently the same.
Since the elements of distributed cognition include individuals, groups, and their environment, defining the boundaries of DC seems like an arbitrary task. Technically, every form of cognition is distributed across some systemic framework, and the primary source of information and feedback control for all known cognitive systems is reality, a data set which contains the entire universe in motion. Where precisely do the boundaries of DC start and end? At the cellular level? The species level? The tribal level? The family level? The galactic level? Unfortunately Cognition Distributed barely addresses this slippery issue, instead offering models, dynamics, and systems that describe likely examples of DC at work at the macro level.
To break down the fundamentals of DC, the current thinking is that any set of cognitive units that works in a groups (neurons, ants, people) will naturally fall into states of specialized cooperation that maximize the energy efficiency and output of the entire system. These states of specialized cooperation can be studied in terms of input and output to see what kind of results and errors emerge at the group level that would not exist if all parts of a system performed the same tasks in serial or parallel. The performance variables between emergent (dynamic) and aggregate (unchanging) states in distributed systems can be applied equally to paths worn into the lawns connecting the buildings of university campuses or the corporate-mandated behaviors of a team of baristas filling orders at your local coffee shop. The explicit rules that govern group behavior are not always most efficient for the individual, and thus improvised shortcuts are taken and new group behaviors emerge from the bottom up. These energy saving shortcuts are not evidence of laziness or bending the rules, they are evidence of distributed cognition at work. See, it’s easy.
In truth, ‘Cognition Distributed’ (See how they did that, switching the words to make them mean something else) is what corporate management types drool over as a means to turn employees into blind producers in a system they can never quite grasp — replaceable units of the corporate swarm intelligence manifesting spontaneous bottom-up strategies to save time and money. A favorite essay of mine, vaguely titled “Thinking in Groups,” demonstrates that too much information shared uniformly across a network of people leads to less efficient solutions than networks bound by only two or three closely related neighbors sharing overlapping networks. Lessons like “Divide and Conquer: Exploration and Exploitation in groups” and “More information isn’t always better” sum up the managerial slant of this text. Research has now shown that consensus breaks the information flow by producing too many interpretations with too many opposing viewpoints; smaller networks with limited information perform tasks more efficiently because they work quickly to maximize skills and improvise multiple solutions. Hence, all DC systems conform to standards of energy efficiency through delegation of complex tasks across specialized units.
When not belaboring the importance of human coupling with symbols or the environment, or pounding home fundamental concepts like the coordination of tasks within a system, Cognition Distributed does include a few forward-thinking gems. The appropriately brief three-page analysis of social tagging on services like Flickr and Delicious as evidence of distributed cognition, by Luc Steels, is brilliantly concise. Closer scrutiny of search engine categorization rules demonstrate empirically why Google and Amazon return the most relevant results. And analyzing the epistemic weight of behavioral decisions made when rotating a falling Tetris piece may be stretching the boundaries between DC, visualization, intuition, and reflex, but it is still a fascinating read. There is stuff for even the geekiest geek to like here.
Where precisely do the boundaries of Distributed Cognition start and end? At the cellular level? The species level? =The galactic level?
Nothing in Cognition Distributed is totally mind-blowing. If anything, the text steers towards the conservative. A conjunction of all the articles would tell me that the current state of DC is that distributed cognition means whatever you say it means, because DC applies to any system that exhibits learning or behavior at a group level, which is just about everything that moves. Is a rock a form of distributed cognition? It may be made up of many elements, and they are all performing an aggregate, non-emergent task of retaining their form over time, but are they learning anything? Are they performing any behaviors? Probably not anything tangible, but you could make a case that the rock is only behaving within the affordances allowed it by physics, and thus the rock is a prime example of raw elements put to the distributed task of maintaining solidity over time. ‘Cognition Distributed’ often mires itself in this level of granular debate. This is fine for academic clarity, but is certainly not as stimulating as examining DC in teenage peer pressure situations, or in Hollywood pitch meetings, or in Political or Wall Street or GroupThink sessions, where tasks that should be easy go horribly awry due to some X factor like Stress or Ideology or Demographics or Polling, usually the result of some other system of distributed cognition feeding into skewed output.
Making a case that DC is used in the gathering of information in a crime scene is easy. But what about DC applied to running a war, or trying to insure the collective property of a million people? Corporate and military structures seem ripe subjects for DC discussion, but no such controversial territory is covered here. In the real world, DC is messy and error prone, efficiencies emerge but not in the smooth and idealistic ways that the academics in this text suggest. Phenomena emerging from the bottom up is another word for revolt against imposed order, error correction and re-correction over time. If the stock market is evidence of DC, then the entire field of distributed cognition has yet to prove its value beyond formalizing the taxonomy for complex systems still prone to human error.
While ‘Cognition Distributed’ is a fine text for conceptualizing the boundaries and domains of DC in its many forms, I was disappointed to see so little work related to the internet. Search engines and their cognitive functions (cataloging, filtering, assessing relevance) were scrutinized, but beyond collaborative tagging, there was nothing on Wikipedia or Web 2.0 content generation. When I type a query into Wikipedia, there is a long train of DC behind whatever answers I receive, starting from the invention of written symbols and continuing right up to the thoughts and actions of Wikipedia producers and contributors, as well as the people who developed, produced, and delivered my computer to my door, and made all the hardware and software in between. Thousands of years of distributed cognition went into crafting the technology and logistics necessary for me to tap my fingers on a keyboard and offload my personal cognition onto the distributed system of experts grouped behind topics on the internet and receive my answer instantly. This goes without mentioning all the programmers, network administrators, cable-layers, wireless experts, middle-managers, investors, corporations, R&D scientists, government agencies, DB admins, and so on — all of them coordinating the infrastructure that couples my query to my answer across thousands of miles of Earth at light speed within micron precision. In a book called Cognition Distributed, you might think this emergent phenomena of globally coupled peers sharing information instantly would be addressed on the first page, but the heady topic of global interconnectedness as evidence of DC is limited to a three page analysis of collaborative Flickr tagging, and there is no mention of the rise of cluster or swarm computing that marks the beginning of new computing era. These might be subjects for next decade’s pop science trend, probably with a whole new cool buzzword.
Cognition Distributed ends with two chapters on speech recognition and computer-aided translation that hard-core AI geeks will find worthy, but ultimately this text left me unfulfilled. It is a great reference for anyone who wants a sharp overview of the DC field, but don’t get hung up on the buzzwords. DC is just the cool new term for studying group coordination of complex tasking. Driving in traffic, reading the news, walking through a mall, ordering coffee, texting… These are all evidence of distributed cognition at work. It’s so trendy, everybody’s doing it.