(Following up on the furore from my previous post – somewhat tongue-in-cheek, of course, but with a serious point.)
After Dave Snowden started accusing everyone – especially me – of ‘pseudoscience’ and ‘psychobabble’ – I began to worry. What if he’s right? What if everything I do is just pseudoscience, caught up in a cult?
(Oops – another long one: better split it here with a ‘Read more…’ link)
I re-read Beyerstein’s list of characteristics of pseudoscience in Patrick’s Lambe’s post on “Is KM a pseudoscience?“, and started to worry even more. Here’s that list:
- Isolation – failure to connect with prior and parallel disciplines
- Non-falsifiability – no means to invalidate hypotheses
- Misuse of data – leveraging data out of context or beyond validity
- No self-correction, evolution of thought – often centred round a single ‘thought-leader’
- Special-pleading – the claim that this is a special-case that can’t be measured in any other terms
- Unfounded optimism – unrealistic expectations
- Impenetrability – an over-dependence on complicated ideology and obfuscation, or bluster in place of debate
- Magical-thinking – such as “the belief that good things will result from willpower alone”
- Ulterior motives – particularly ulterior motives of a commercial kind
- Lack of formal training – including certification schemes that link back to #4
- Bunker mentality – such as complaints about being ‘misunderstood’ by others, and often linked to #5 and #7
- Lack of replicability of results – especially replicability by others under controlled conditions
For example, I often work at the places where the IT-industry and consulting-industry converge, so I would need to test both of those against that list, putting a check-mark against any of the criteria that fail:
- #1? Check – not always, but way too often for comfort.
- #2? Check – ditto.
- #3? Check – often. Usually from myopia and questionable competence (“I guess we failed to take enough account of the human factors”: BPR), though occasionally from a rather more deliberate ‘sexing-up’ of the statistics to prop up the purported position (‘In Search of Excellence’ etc).
- #4? Check – often. (All those management fads…)
- #5? Check – again, too often for comfort. (Real business-case for IT-only KM or Enterprise 2.0, anyone?)
- #6? Check. (In fact rarely anything other than ‘unfounded optimism’ – at the start of a project, anyway.)
- #7? Check – lots.
- #8? Check – ditto.
- #9? Check. (Rarely anything else, perhaps? – BPR, anyone? ERP? the dot-com bubble?)
- #10? Check. (Look at most enterprise-architecture training, for example.)
- #11? Check. (It’s usually called ‘the IT/business divide’ – or worse, of course.)
- #12? Check. (Often we don’t want to replicate the results that we actually get…)
And so on, and so on. According to that review, it looks like almost the entire industry is based on little more than pseudoscience. Oops.
And we’ve already seen from Patrick Lambe that knowledge-management is perilously close to a pseudoscience too. Also ‘Oops’, I guess.
But what about Cynefin? Surely that can’t be a cult – especially given Dave’s position on pseudoscience and the like. Better go through that checklist again, just to make sure:
- #1: Isolation? Plenty of reference to cognitive-science and suchlike – but I don’t see any evidence of cognitive-science etc connecting back to Cynefin. Looks suspiciously like spurious science to me, then. Oops.
- #2: Non-falsifiability? References to ‘retrospective causality’ in the Complex domain look a bit questionable in this regard; likewise much of the definitions of the Complex and Chaotic domains, and the interactions therein. Oops.
- #3: Misuse of data? Ditto, it would seem. Oops.
- #4: No self-correction? There is a genuine community-of-practice here, but it seems often to be silenced by a single figurehead who claims to hold ‘the only real truth’ about the discipline. Oops.
- #5: Special-pleading? Tends to be very good about challenging ‘pattern-entrainment’ in others, but not so good at applying the same analysis to itself. Claims to be a ‘sense-making’ framework, but the only way to test the ‘sense’ that’s derived is in terms of the framework itself. Kinda circular, really. Oops.
- #6: Unfounded optimism? Probably. Best let that one pass as only a minor ‘oops’.
- #7: Impenetrability? Lots. There’s the ‘ganglionic cross‘ with its cryptic markings, and the insistent demand that all devotees acknowledge that there are five domains, not four; also near-religious wars as to what each of the domains ‘really means’. Oops.
- #8: Magical-thinking? All we really know is that Dave is almost obsessively against it – which by the usual psychological games probably means there’ll be lots. Complicated pathways between domains that somehow magically change things might be a good example. A bit uncertain, perhaps, but very likely to be ‘oops’.
- #9: Ulterior motives? Lots. Celebrity-status, serious-money consultancy-fees, training-fees (see #10), sales of software that can only be used by registered practitioners (see #10), and consumable-supplies that can only be purchased from the central organisation: sounds a bit like Scientology, doesn’t it? We’d have to be fair and remind ourselves that that applies to much of the IT-trade and consultancy-trade too, but even so that’s a really big ‘oops’.
- #10: Lack of formal training? Would-be practitioners generally need some serious consultancy-time under their belt, but the Cynefin training itself is defined, run, certified and validated only by the central organisation. In other words, worryingly circular and self-referential. Kinda sounds like NLP, doesn’t it? Oops.
- #11: Bunker-mentality? Probably not in most cases, but it’s notable that the figurehead has an unfortunate habit of fulminating about anything else that can’t be forced to fit within the preferred assumptions – such as denigrating Six Sigma as ‘Sick Stigma‘, and so on, regardless of where or how it’s used. So most practitioners probably okay, but the figurehead probably not. Oops.
- #12: Lack of replicability? Lots. By definition, pretty much anything in the nominal Complex or Chaotic domains is going to have limited replicability. (There’s good replicability in the Simple and Complicated domains, of course, but also no real need for Cynefin-style ‘sensemaking’ in those two domains, so we can’t really claim that one as a plus.) Just about any consulting-assignment will be in part unique, too, so again little to no replicability there, again by definition. Also, as Dave puts it, “every diagnostic is an intervention”, so the very act of enquiry changes the conditions of the experiment, impacting on any possible replicability. And if Cynefin experiments are only repeatable by Cynefin practitioners, and everything has to be assessed in Cynefin terms, it somewhat blocks the possibility of proper third-party ‘outside’ review – kinda like the worst of ‘armchair Freudianism’, for example. Another big ‘oops’.
So is Cynefin a cult? Apparently the answer is ‘Yes’, because according to Beyerstein’s criteria, it seems to fail the ‘pseudoscience’ test on just about every count. Almost the only place where it doesn’t fail, in fact, is in the two logic-based domains, Simple and Complicated, where Cynefin isn’t much use anyway. Either way, definitely ‘oops’.
[Brief note to Dave: yup, I’m well aware that that assessment above ain’t exactly rigorous and peer-reviewed and the rest, but it’s a darn sight more rigorous and honest than the cheap hatchet-job you tried to do on me over the past couple of days… yes, I am indeed still angry over that…]
But that result is kind of odd, because most of us find that Cynefin is a very useful tool in consulting practice – especially in dealing with what Cynefin describes as the Complex and Chaotic domains. Hmm. Seems like something doesn’t quite match up here, does it? And we’re left with two probable reasons for that mismatch:
- either Beyerstein’s criteria do test well for pseudoscience in areas where simple Newtonian-style logic applies, but tend to break down as soon as we hit anything closer to real-world chaos – so we’ll need something other than Beyerstein and the like to validate quality in those areas
- or the whole idea of ‘pseudoscience’ is a red-herring that can be used by superannuated academics to bully others and prop up some vain and misguided ‘mediaeval delusions’ of their own ‘superiority’, in areas where their putative expertise in formal ‘proof’ by definition can no longer apply, because by definition the ‘normal’ rules of replicability and the like are no longer reliable once we move into the Disorder, Complex or Chaotic domains
Both of these could be true, of course. But let’s be polite, and assume that it’s only the first of these: Beyerstein is probably useful in the Simple and Complicated domains, but we’ll need something else outside of that simplistic rule-based world.
But how can we tell when we’re outside of the rule-based world? And what can we use in place of Beyerstein and its ilk?
For the former, the key criterion is, once again, repeatability and replicability. In both the Simple and Complicated domains, there’s always an identifiable ‘right answer’, and if we do an experiment in the same way, we’ll always end up at the same results. (A few special-cases such as symmetries in complex-math give two or more ‘right answers’, but the set of answers in that case is still identifiable, so the basic principle remains sound.) In short, it’s repeatable, which means it’s also replicable. There are definable, straightforward (more or less!) and linear sequences of cause and effect, so if we don’t get the same right-answer under the same conditions, something’s wrong – hence falsifiability. Either true, or false: hence it makes sense to describe Cynefin’s ‘Simple’ and ‘Complicated’ as the two ‘truth’ domains.
In the other Cynefin domains, things get kinda messy. The Disorder domain is where we start, before we do any sensemaking, but it’s probably best to leave it out of this discussion for now. Yet in the other two domains – Complex and Chaotic – doing the same thing in the same way does not guarantee the same results. In the Complex domain, any apparent causality will at best become apparent only after the event (a context which Dave Snowden describes as ‘retrospective causality’); in the Chaotic domain, where everything is inherently unique in some way, even the concept of causality itself makes no sense, by definition. In effect, one way that we know we’re not in the ‘truth’ domains because it’s not repeatable. Clearly Beyerstein isn’t going to be much use to us here.
But there’s a nasty corollary that follows from this. If one test of the Complex or Chaotic domains is that it’s not repeatable, how can we tell the difference between that and plain ordinary bad-science in the ‘truth’ domains? – because that’s also not-repeatable too. Following the logic of this, we discover quite quickly that there’s no simple ‘truth’-based test that could distinguish between the two, because in both cases doing things the same nominal way may lead to different answers. Beyerstein in this instance would not only not be helpful, but could be actively misleading, always labelling the workings of the Complex domain as ‘wrong’ and therefore ‘pseudoscience’. Which it might be, or might not be, but there’s no way to tell: even the concept of ‘true’ versus ‘false’ doesn’t make much sense in that kind of context. Which is a problem.
But instead of trying to cling on to a notion of ‘true’ versus ‘false’ in a context where it won’t and can’t work, what does make sense is to use some concept of value. In other words, the test-criterion we need in the two ‘value’-domains Complex and Chaotic is usefulness, not ‘truth’.
Next question: what determines ‘usefulness’? By definition this is always going to be somewhat subjective and context-dependent – but that doesn’t mean that it’s a random free-for-all. Feyerabend’s anarchic dictum “anything goes” does indeed need to hold sway here, but it’s a disciplined ‘anything goes’ – a considered, functional form of anarchy, if you like (or even if you don’t). In turn, this brings us into the well-understood (if, by its nature, not necessarily well-defined) realm of quality-management. Which brings us to all those tools that Dave has so happily despised, such as Six Sigma and the like.
The problem – and again for some impenetrable reason Dave doesn’t seem to like this fact – is that each of these tools is context-dependent. Six Sigma, for example, is all about managing quality in terms of defects per million events: so it only makes sense to use Six Sigma if we have millions of exactly-identical events, which in practice places us in Cynefin’s Simple domain. If we’re not in the Simple domain, don’t use Six Sigma: simple, really. No need to make a song-and-dance about it and denigrate it as ‘Sick Stigma’, because it’s perfectly fine where it does work. Same with every other tool and technique: we switch between them according to context.
Within any given context, I’ve also found it useful to compare against a relatively simple yardstick of effectiveness. In my own practice, for the past decade or so, I’ve used a frame which describes effectiveness in terms of five distinct dimensions:
- efficient: makes the best use of available resources – typically the least wasteful use
- reliable: can be relied upon to deliver the required results, optimised over the required timescale
- elegant: aligns best with simplicity, clarity, ergonomics and other ‘human factors’
- appropriate: ensures that the delivered results are ‘on purpose’
- integrated: assists in bringing everything to work together as a unified whole
All of these need valid metrics: and in general, any appropriate metric will do – even if sometimes it’s just a 1-5 subjective scale, such as I use, for example, in my SEMPER organisational-capability diagnostic. Once again, in effect ‘anything goes’ here: the selection-criteria for metrics revolve around effectiveness, not ‘truth’.
And ‘truth’ approaches – such as Dave has so aggressively promoted in the comments to the previous post and elsewhere – really aren’t much help in deciding metrics and models here, because ‘truth’ only applies to part of the context, as we’ve seen above. True/false logic can’t lift itself by its own bootstraps: it can work within a set of assumptions and postulates, but it can’t be used to define or validate them. (Attempting to do so is known as ‘induction’, otherwise known as ‘cheating’ – or ‘pseudoscience’, of course.) So to make it work we have to jump up a step to a kind of ‘meta-level’, which, as I said in the previous post, might be called ‘nonrational’ or ‘arational’ or ‘metarational’, but I prefer to use the good old classic term ‘magical’. Which Dave doesn’t like, but that’s too bad, bluntly. (He also doesn’t like the alternate term ‘technology’, so he’ll just have to lump it, really.)
To help in deciding metrics and models and the like, we need to run the whole thing reflexively and recursively. (I’ve described in some depth how to do this in whole-of-enterprise architecture in my book Real Enterprise Architecture, if you’re interested.) The Cynefin frame is useful for this: we run it backwards, so to speak, to help us identify what needs to be handled in an appropriate manner for Simple, Complicated, Complex or Chaotic.
Even more useful than Cynefin for this, as mentioned in the previous post, is the frame that we developed for Disciplines of Dowsing. And as you’ll see from the two-page reference-sheet, the reason why it’s even more useful is that not only does it describes characteristic to help us identify which mode or domain we’re in, but also how to recognise when we’re losing discipline within that domain, and reasons and tactics to move from one domain to another. For example, if we’re in the ‘Scientist’ domain (i.e. Cynefin ‘Complicated’ domain), and we start getting emotional and defensive or aggressive about it, that warns us straight away that we’ve allowed ourselves to drift towards the ‘Priest’ domain (Cynefin ‘Simple’ domain), and either need to get the emotion out of it to return to the Scentist, or else intentionally switch to the Priest, or one of the other domains, as appropriate. The result is that we maintain discipline throughout the whole space – not solely in the ‘truth’ domains, as with Beyerstein and the like.
I’m well aware that dowsing and suchlike may feel a bit uncomfortable for some folks here, but unfortunately it’s the only example I have available right now. (There’s another variant in the Berg Time & Mind article, showing how to balance subjective disciplines in archaeological research with the more conventional ‘objective’ disciplines, but it’s essentially the same as in the reference-sheet.) Likewise the examples in the useful set of ‘seven sins of dubious discipline’ in the Disciplines book mostly relate to dowsing and archaeography. So if you’re working in knowledge-management or enterprise-architecture, for example, you’ll probably need to do some significant translation to make it work in your own work-context. But I assure that it is worth the effort: the result makes it a heck of a lot easier to work out what’s going on in a context – especially the kind of dysfunctional, chaotic, blame-filled business-contexts that we so often have to deal with these days – because it helps to ensure that discipline of an appropriate kind is kept in play at all times.
So, to come back to the original question, is Cynefin a pseudoscience, a cult? Short answer, as we’ve seen above, is “probably not” – but you’ll probably need a little bit of magic to help you prove it! 🙂
Constructive comments and suggestions welcomed, of course – and many thanks for sticking with me this far on this.