Fascinating paper from the always thought-provoking and often inspirational Francis Heylighen, in which he draws together various models of distributed intelligence, distributed cognition, evolution and complex adaptive systems, incorporating stigmergic and networked perspectives on ways that self-organizing systems can exhibit intelligent behaviour. This is very relevant to anyone interested in connectivism, collectives, learning, intelligence, complex systems or social software.

Heylighen’s central thesis revolves around a definition of intelligence as not just problem solving but also opportunity seeking: it’s about both overcoming obstacles and seeking new possibilities. This combination is encompassed by the term ‘challenge’, which Heylighen defines as ‘a phenomenon that invites action from an agent’. Given competing positive (proactive) and negative (reactive) challenges, he sees challenge in evolutionary terms as ‘a promise of fitness gain for action relative to inaction’. All of this is framed in a context of bounded rationality and different approaches to challenge resolution, from simple look-ups to complex heuristics, and a range of factors that may motivate or demotivate different actions. This is all good stuff but it gets really interesting when he reaches the ‘challenge propagation’ referred to in the title. In essence, this applies the logic of memetics to challenges. As he puts it:

“In contrast to the standard paradigm of individual problem solving, the challenge propagation paradigm investigates processes that involve a potentially unlimited number of agents. To deal with this, our initial focus must shift from the agent to the challenge itself: what interests us is how an individual challenge is processed by a collective of agents distributed across some abstract space or network. Instead of an agent traveling (searching) across a space of challenges (problem space), we will consider a challenge traveling (propagating) across a space of agents.”

This is a brilliant idea. I love the change in perspective that this brings. There are, I think, some very large and unresolved questions about what a ‘challenge’ means in the context of a collective. This follows from the fact that it is hard to understand what fitness in such a collective might consist of, save in its utility to the agents of which it is composed, though it might shed some light on our eusociality (evolution not for the benefit of selfish genes but for the benefit of a large social collective). I find it particularly hard to map his earlier discussion of how things are valued (with ‘valences’) by an individual agent and how things might be valued by a collective. A challenge does not exist in isolation – it must have a subject. It’s not entirely clear what that subject might be here. Such fuzziness aside, as a way of understanding an otherwise massively complex intelligent system like a brain, an ant colony or human culture, it has a lot going for it.

While the foundations are very strong, I have some reservations about some of the examples Heylighen uses and some conclusions that he draws. While I can readily accept that there are some stigmergic aspects to Wikipedia, I do not believe that the act of editing a page is in any meaningful way analogous to the way that stigmergy operates in (say) termite mound building or movements of currency markets. In the first place, unlike in a true stigmergic system, there is an infinite range of possible ‘algorithms’ that might influence agents making changes to a Wikipedia article. There are path dependencies, sure, but that doesn’t make it stigmergic. Apart from some stylistic patterns that tend to replicate, there is none of the emergent self-organized behaviour that is characteristic of all stigmergic systems. A Wikipedia page is largely just the sum of its parts, not an emergent artefact. In the second place, unlike in stigmergic systems, individual agents make deliberate contributions with a clear design purpose and end-goal in mind when building a Wikipedia page: their interactions are not local but planned and focused on the whole. It is no more stigmergic than a house to which someone decides to add an extension or remodel its rooms. It’s a good model for cooperative action, but not of collective intelligence.

I’m also not entirely happy with the notion of the Internet as being a gigantic collection of forums (generalized by Heylighen as ‘meeting grounds’) to exchange challenges, though the metaphor is appealing on many levels. The same could, of course, be said about any human artifacts or ways of ‘meeting’, from buildings to tools to doorknobs to forest footpaths to books to conversations to simply passing in a street. So far so good.  He describes the propagation of challenges as involving division of labour, workflow, and aggregation – this too makes sense. He then describes how such a system becomes self-organizing and uses as an example the growth of open source software. Here I have problems, for much the same reasons as I have problems seeing Wikipedia article development as stigmergic. In real life, many large open source developments are a million miles from self-organizing. The archetypal Linux, for instance, is extremely tightly controlled by a very small number of people using very rigid processes that are in many ways more traditionally organized from the top down than most proprietary systems. While the challenges are indeed solved by individual agents acting largely independently, albeit building on what others have already built, the workflow and aggregation are firmly in a traditional designed mould and tightly controlled by a clique. This is even true of more open approaches, such as those encouraged by Github although, in this case, workflow is managed by a ‘blind’ algorithmically driven system rather than by a clique. 

My concerns are minor and they are not with the basic ideas presented here, that I find very compelling. I think this is an important paper. While it certainly needs refinement, this feels like the beginnings of a new language for discussing and describing connectivist accounts of learning. It provides some much needed solid underpinning theory and a very useful perspective on some of connectivism’s major tenets: that knowledge exists in the network, including non-human artefacts; that connections are learning; the significance of decision-making; the ways that more is different; and the value of diversity. Great stuff.

Address of the bookmark: http://cleamc11.vub.ac.be/papers/ChallengePropagation-Spanda.pdf