Distributing Situated Cognition to Mimic Generalizable Knowledge

Update on Project Natal
In my October 21, 2010 blog post on situated learning spaces, I mentioned Project Natal’s virtual boy named Milo (there’s also a virtual girl named Kate), and posted a 2009 demo launch of Milo.  Peter Molyneux, head of Microsoft’s European games division, has since appeared on TEDGlobal to talk about Project Natal (July 2010), where he showed the real Milo technology on TED’s stage:

We see the demonstrator interacting with Milo using body gestures, facial gestures, and voice – all of which needs to be “taught” to Milo.

Milo’s Mind as Metaphor of Knowledge Building
What really interests me about Milo is what Molyneux says in his TED presentation about Milo’s mind:  “Milo’s mind is in a cloud”.  In other words, the more people who use Milo, the more objects he’ll learn and recognize.  Fascinating!  To me, this implies that Milo’s artificial intelligence is distributed, and Milo’s mind is the synthesis of training combinations contributed by a community of Milo users.  Milo’s mind is a wonderful metaphor for collaborative knowledge construction, where Milo’s mind represents a knowledge community’s World 3 Rise Above knowledge artifact, and the community of Milo users is the knowledge community.

Situating Milo’s Distributed Mind
In my October 21, 2010 blog post, I had hypothesized that “the more the situated learning context approximates a realistic context of the domain of study, the deeper and more meaningful the learning will be”, implying that Milo could contribute to deeper situatedness in wider contexts.

Since then, I have re-read Carl Bereiter’s 1997 book chapter, Situated Cognition and How to Overcome It.  He mentions what he and Scardamalia (1989) have termed intentional learning, which has 3 goal levels:

  1. Task completion goals – to be completed immediately (e.g., assignment due date tomorrow); World 1; highly situated
  2. Instructional goals – to be completed summatively in the near future (e.g., end of the course); world 2; less situated; somewhat transferrable
  3. Knowledge building goals – may extend indefinitely into the far future and past; World 3; weakly situated (i.e. immediate situation); highly transferrable
In my December 7, 2010 blog post, I mentioned Bereiter’s reasoning behind why it would be prudent to overcome situated cognition – situated learning is inversely proportional to generalizability of that knowledge.  Thus, an individual’s knowledge would be high transferrable if that knowledge was learned with an intentional orientation toward knowledge building goals.
Continuing on this train of thought, when I consider the “learning” that Milo “does” as he is being used by multiple members of the Milo user community, I would say that almost all of these learnings are for task completion goals.  Thus Milo’s artificial intelligence should be highly situated and therefore highly ungeneralizable.  If we assume that Milo will have a large and active globally-distributed user community, this alone would exponentially increase Milo’s potential learning situations/contexts.  If Milo is taught by this large globally-distributed user community to recognize an object – a chair for example – in multiple and varied situations, then he should be able to recognize chairs of all types and across all situations.  In other words, the object “chair” mimics generalizability.  Have I found a way to achieve a mimicry World 3 with a low-level goal of task completion – by exponentially increasing learning situations contributed by a large knowledge community?

This brings to mind the authors of Wikinomics (2008) and Macrowikinomics (2010) Don Tapscott and Anthony Williams’ idea of Murmuration Macrowikinomics.  Here’s a narrated video metaphor of it:

Tapscott is careful to note that “…this is not a collective intelligence or collective consciousness of course, because individual birds are not intelligent or conscious…”.  He describes the thousands of starlings moving as one mass, as a “…loosely conjoined network of relationships and impulses” – much like a globally-distributed Milo user community?

Tapscott asks:  “Will we come to consider networking as the neural roots that connect human beings in a way that creates something fundamentally new?”  I do not hesitate to say “YES – but…..!”  Much like Milo’s “mind” (i.e. artificial intelligence) cloud, that grows as his distributed user community “teaches” him more, Milo’s knowledge base will only ever mimic World 3, and appear to be generalizable.  What allows Milo’s mind to mimic generalizable knowledge is the numerous situations in which he has learned the same thing.  If millions of users around the world “teach” Milo to recognize the object known as “chair”, presumably, they’re all using different kinds of chairs available in their homes.  These chairs could vary greatly in age, design, style, size, etc.  Hence, Milo acquires the ability to recognize any chairs that match any one of the millions of chairs his distributed user community has inputed into his memory stores – giving the appearance that Milo has generalized recognition of the object called “chair”.  Milo’s seeming intelligence is indeed artificial.

Back to Tapscott’s question.  I think that networking neural roots connecting people would create something fundamentally new, but not necessarily better.  Like Milo’s mind cloud, this new thing would only mimic World 3 due to the exorbitant number of task-completion-oriented contributions to its knowledge base from its massive and globally-distributed user community.  Highly transferrable World 3 knowledge can only be created with an orientation toward knowledge building goals, and must consider the far past and continue into the far future (Bereiter, 1997).

References
Bereiter, C. (1997). Situated cognition and how to overcome it. In D. Kirshner & J. A. Whitson (Eds.), Situated cognition: Social, semiotic, and psychological perspectives (pp. 281-300). Hillsdale, NJ: Erlbaum.  Available: http://www.ikit.org/fulltext/1997situated.pdf

3 thoughts on “Distributing Situated Cognition to Mimic Generalizable Knowledge

  1. C. Fong

    Hi Peter,

    Thanks so much for your thoughtful and enlightening comment! You've definitely broadened my understanding of 'metacognition' and 'intentional learning', and how they differ!

    I completely agree with you – Milo can't deal with these other elements. Though I wonder…if the distributed Milo user community is EXTREMELY large (e.g., millions of active users), then presumably, they're all teaching Milo's mind cloud things (some of which would be similar) in millions of varying contexts/situations. Could these vastly varying contexts randomly combine to approximate something that might resemble emotion? How does one explain the Data character on Star Trek The Next Generation? (Yes, I used to be a Trekkie when Jean Luc Piccard was Captain – and proud of it!!!)

    In any case, even if this approximation were possible, I still think (like you) that Milo, lacking consciousness, intention, and emotion; would never really be able to learn intentionally.

    ~Cresencia

  2. Peter Skillen

    Hi Cresencia,

    Thanks again for a very thoughtful post about stuff which I love to ponder. It also distracts me from the work I am supposed to be doing. 🙂

    Project Natal is fascinating – there is no doubt about it.

    Some thoughts, however, come to mind re learning in this context. One of these is the difference between 'metacognition' & 'intentional learning' – neither of which I think Milo could handle so well – and, which are essential to the human condition and to learning.

    Metacognition, as we know, is usually considered to consist of both knowledge about cognition and regulation of cognition. Knowledge about cognition refers to information that people have about their own cognitive processes and those of others (Flavell & Wellman, 1977). Metacognitive skills include, not only this knowledge, but an awareness and ability to engage in “planning activities prior to undertaking a problem (predicting outcomes, scheduling strategies, various forms of vicarious trial and error, etc.), monitoring activities during learning (testing, revising, rescheduling one's strategies for learning), and checking outcomes (evaluating the outcome of any strategic actions against criteria of efficiency and effectiveness)." (Brown, Bransford, Ferrara, and Campione (1983, p. 107))

    Intentional Learning theory differs from metacognitive theory in that there is an explicit recognition of other aspects of self. Intentional Learners are developing, not only well-developed metacognitive skills, but also attitudes (an affective stance), motivations, and social behaviours that are focused on, and conducive to, advancing one’s own knowledge and the knowledge of others. Bereiter & Scardamalia suggest that to generate a useful educational theory one cannot concentrate solely on the knowledge aspect of intentional cognition, but must also come to understand and include other aspects as well. These include motivation, affect, allocation-of-resources, and ecology.

    I am pretty sure that Milo can not deal with these other elements. The learning is externally directed in a non-aware state devoid of intention and emotion. So, as thrilling as it is, we sure have a lot of work to do in creating 'real learners' online. 🙂

    But, it sure gives us material with which to think.

    thx again!

    Peter

Comments are closed.