thanks for the long, very insightful reply – and thanks again for the long, insightful post itself .

First off, your description of how one heuristic “bundle” comes to outweigh another one has a striking resemblance to how connectionism (http://en.wikipedia.org/wiki/Connectionism) models cognition.

Next, I basically put you on the spot for something that I do myself – namely using “the 5 Why’s”/”laddering”/however you name it, and not accountig for how prone it is to post-rationalisation errors. It strikes me that this really, really gets at the root of the distinction between humanistic/phenomenological and behavioristic approaches.

For if you take post-rationalisation *really* seriously (as Skinner did), you have no choice to validate that your if-then rule is a fitting description but to run a series of experimental tests where you try to test and falsify all other alternative plausible rules that could also describe my behaviour. To stick with techcrunch: Is my rule “don’t trust anything with a ‘t’ in the domain name”? Or “don’t trust anything with a .com domain ending”? So you amass data until you can say: “This if-then rule (don’t trust anything that has ‘techcrunch’ in the URL) is the only one congruent with all data, so I stick with it”. Wich is essentially what behavioral (re)targeting does: Run with whatever fits the data best, not knowing whether you follow a correlation or causation.

Whereas if you grant that people may have at least *some* self-insight into their mental states (not, mind you, the underlying cognitive processes), and that these mental states do correlate at least to *some* degree with underlying cognitive processes, then you are allowed to ask the individual to introspect. Which is why I try – in my guidelines for user research – to instruct people to stick to *describing their lived experience*, rather than explaining what caused this experience.

Why then use the “5 Why’s”? I think we can cut ourselves some slack if we shift the interpretation of what the interviewed person does when engaging in “5 Why’s” – namely, that the person is not somehow magically lifting the mind-body veil and looking at its own cognitive wheels –, but that by and large, we are our own best oberservers in that we have amassed the largest data set of our own past behaviours from which to conjecture. The only remaining issue is that we also tend towards many self-serving biases. Hence we should take the answers to the 5 Why’s as yet one more data point, and nothing more.

Finally, on the levels of abstraction: If you assume that these heuristics ‘exist’ on different levels of abstraction, then that implies that they are not merely the emergent result of what happens in the black box, because that emergent result will by definition always be one and just one rule (if a complex one). But what I consider even more important is that thinking through design for long-term changes or change processes with mental heuristics will likely be *impractical*, because – to go back to the exercising example – you would have to map a “decision journey” of some sort with a million single decision points for every relevant decision on the way. That’s where I think the usefulness of the model breaks down, and we need another model – equally wrong, but more useful.

And yes, definitely, this debate could use another typographical space .

- About the ‘levels of abstraction’ at which these heuristics operate, I suppose the most sensible way I can conceive it working is that in a situation, we will inevitably have a number of relevant heuristics which are applicable, some of which are ‘universal’ like “if it’s going to take a long time to do, then don’t do it” and others such as “if you think it’s going to be tasty, then eat it”, “if it’s fatty, then don’t eat it” and so on. They interact and clash and work on different levels as you suggest. We may have overarching heuristics like “if something is bad for the environment, then don’t do it” yet at the coal-face, we follow heuristics like “if it’s cheaper to drive to work than get the bus, then drive to work”.

Imagine a friend has given you a recipe for a pie which sounds delicious, but will take some time and effort to cook, and is also going to be quite fatty. One or more of those (potentially conflicting) heuristics is going to ‘win’, but it’s unlikely to be simply a case of adding up columns of pros and cons for each possible action (like Bentham) or enumerating the possibilities of each (à la Darwin) – rather, something will lead one heuristic to dominate. Maybe if you’re hungry right now, the rule about time will dominate (and lead you to follow the course of action of eating something else instead), but maybe the potential tastiness will dominate. The presence of other actors might also sway you one way or the other – is it fair to inflict this fatty pie on someone you’re dating if you know she watches her weight, or will she be impressed by the time/effort/kindness/tastiness?

Why exactly one/a set of heuristics dominates, and why the dominant one(s) is/are different for different people in different contexts, is, I guess, basically what the entire field of decision research is about. Maybe all of psychology. But from the entirely pragmatic design point of view, I would venture that even the process of trying to understand the heuristics which are present at the point of decision, and what choices are really being made (e.g. is it fatty vs healthy, or time vs taste, or impress-the-girl-by-cooking vs impress-her-by-considering-her-health?) by particular users ought to give us a much better insight into what design techniques would help support / change particular decisions and behaviours.

- Getting at the heuristics. Yes, you’re right, the challenge of actually uncovering them is massive, and I did skirt round the issue in the post. There could indeed be infinite interpretations of each, and there will be post-rationalisation by both the researcher and the participant, I’m sure. I suppose my reasons for considering the heuristics as something like black boxes is exactly for this reason: compare outputs to inputs and if you can describe the results in terms of a rule – even if the reasons given behind the rule are wrong or absent – the rule may still be useful. To take your Techcrunch example, describing the heuristic as “don’t click on anything saying ‘techcrunch’” doesn’t presuppose the reasons for itself – which may be due to disliking their style, may be due to having learned that the articles will make you angry / waste your time, or it may be something else entirely (like not wanting to be exposed to YouTube-like levels of comments). But the rule “If it says Techcrunch in the URL, don’t click on it” is detailed enough to inform at least a basic model of your behaviour when presented with URLs. It’s enough to know that to get you to click on a Techcrunch link, the URL will have to be obfuscated. To the extent that I’ve thought about a process for this (which I haven’t yet in detail – hence why your comment is so helpful!) I would venture that the really basic, solely descriptive behavioural heuristics, with no motivation or reasoning assigned to them, could act as a kind of first step to investigating (and influencing) behaviour, with more detailed investigation following.

Something why always struck me about the ‘Five whys’ technique was that while it might be intended as a ‘root cause’ analysis, each level of ‘Why?’ potentially offers useful opportunities for intervention to solve a problem. If someone asks me “Dan, why don’t you do more exercise?”, my initial answer will probably be “I’m just too busy to be able to take the time to do it” (which might suggest design interventions like exercise equipment that I could use while at my desk). Further stages of ‘Why?’ might reveal “I’m too busy because I’m poor at organising my time” (which might suggest a design intervention around scheduling) or “I’m poor at organising my time because I get too easily distracted” (which might suggest some kind of focus-based design intervention) and so on. Each of them could be a valid behaviour change solution, just operating at different levels of abstraction. Of course many ‘solutions’ are basically treating the symptoms of an underlying disease rather than the disease itself, but this is not uncommon. We still find painkillers useful even though they don’t root out what the cause of the pain is in the first place. And I think design for behaviour change can still be a useful painkiller in many circumstances even as we try to understand the disease in others.

- The parcours and affordances example is interesting. There is some work in the human factors literature on ‘econiches’, e.g. Warren (1995) which is relevant here. I would say, of course, the ‘same’ heuristics may be interpreted differently by different users as they relate to themselves, but if they are derived from empirical observations of behaviour, then we just need to find a level which we are able to address through the design techniques available to us.

- You’re probably right that the design patterns don’t need to follow from the heuristics. If the heuristics are stated clearly enough, the general form of the solution will be obvious. Perhaps the value of the design patterns is really in situations where the heuristics are difficult to uncover or state, or are unclear. Maybe the two approaches could meet in the middle, working backwards from design patterns which seem like they might be relevant and forward from a set of plausible heuristics to define more fully the possibility space for the problem and possible solutions. This is something I’ll try to explore through future workshops!

I probably haven’t answered all your points, but thanks again for a really helpful comment which has certainly given me the opportunity to think this stuff through a bit more clearly!