Knowledge Matters™

Larry Prusak pointed me last week to this new book produced by the Asian Productivity Organisation and edited by Serafin Talisayon on KM in Asia: Experience and Lessons. It’s a combination of case book and survey of the state of KM across the APO’s member countries, with some extremely insightful cases and very astute commentary. A really valuable addition especially for KM practitioners working in Asia. Serafin Talisayon told me the hard copies sold out within days, but the soft copy is available for download (warning: 6MB!).

Introduction

In my two previous posts I’ve talked about the OODA loop framework and its relationships to the Decision Execution Cycle (DEC), Single- and Double-loop learning, and the Knowledge Life Cycle (KLC) frameworks. Here I want to discuss the relationship of Recognition Primed Decision Making (RPD), a primary type of Naturalistic Decision Making (NDM) to OODA, the DEC, Single- and Double-loop learning and the KLC.

Recognition Primed Decision Making and Rational Decision Making

The basic notion of RPD is that humans prefer to “first-pattern-match” in decision making, and then proceed by what is, essentially, sequential trial and error, if the first pattern doesn’t match either their mental simulation of the likely consequences of their decision, or the actual consequences perceived in their post-decision experience. This is a bit different than animal decision making, since humans mentally simulate the results of their contemplated decisions in much more complex and detailed ways than animals who appear to be limited to relatively simple expectations about consequences.

In Rational Decision Making (RDM), humans look for a number of plausible decision alternatives, and then comparatively evaluate them, and select the best option, or according to some notions “the optimal decision.” In the past 25 years, much research has shown that decision makers rarely use RDM, but prefer RPD, and sometimes other forms of NDM. The most well-known research of this kind has been performed by Gary Klein and his collaborators at Gary Klein Associates, and it is fair to say that this research has shown that RPD is functional in situations where RDM is either not, or is impractical to carry out, and also, raises the possibility that RPD is the kind of decision making we ought to employ in most situations, restricting RDM to relatively rare cases where the time, resources, and possible high benefit/cost ratio from an RDM procedure outweigh its far greater costs to implement.

Relationships: RPD, RDM, OODA, DEC and the KLC

In earlier posts, I pointed out the distinctions between routine learning and creative problem solving, and between routine DECs and Problem Solving DECs. I also pointed out that creative problem solving is performed in organizations through KLCs and that they are comprised of DECs motivated by a learning or problem solving incentive system, rather than, primarily, by an incentive to close an instrumental behavior gap. I also related OODA to the DEC and the KLC by identifying simple or routine OODA loops with the DEC, and by making the case that at the organizational level, routine DECs or OODA loops create activities and are organized into goal-directed processes organized around the need to close instrumental behavior gaps. Mismatches between expectations and our experience show the existence of knowledge gaps and trigger KLCs whose purpose is to make and integrate new knowledge. Such KLCs are comprised of multiple DECs or OODAs. But these are different from routine DECs or OODAs, in that they are motivated by the incentive to learn and to solve a specific problem. Once problems are solved by new knowledge and the knowledge is integrated into an organization’s DOKB, it is available for routine decision making and business processing.

Now here is where RPD and RDM fit into this picture. First, in routine learning/decision making DECs/OODAs, RPD is the dominant, if not the sole pattern of decision making, since in such DECs we always act according to our expectations about what the results of our actions will be, and this, further, implies that we are acting according to a recognized pattern coupling our contemplated actions and those expectations. Second, when, however, RPD doesn’t work and our expectations are not fulfilled, then we must recognize that our knowledge about the routine decision making situation didn’t work, that we have a knowledge gap, a problem, and that we must seek new knowledge that will work. We acquire this new knowledge through performing double-loop learning (DLL) through KLCs.

Third, it is in performing KLCs, that we encounter the choice between RPD and RDM, in an acute and complex way. I’ll develop this choice in the context of the KLC framework. In that framework, and assuming a clear formulation of the problem to be solved we distinguish information acquisition, individual and group learning, knowledge claim formulation, and knowledge claim evaluation as sub-processes within knowledge production; knowledge and information broadcasting, searching and retrieving, teaching, and sharing within knowledge integration; and finally, use of the new knowledge in post- KLC decision making. The first thing to notice about this is that in the KLC context, unlike the routine decision making context, there are multiple DECs (or simple OODA loops), and hence multiple decisions. The second thing to notice is that the distinguishing mark of RDM is its focus on multiple decision alternatives and then its evaluation of these and selection of one of them as the preferred alternative for action. This implies that RDM, unlike RPD, is a multiple decision loop process, like the KLC. I’m not sure that this point comes through very clearly in the literature when RPD is compared with RDM. There it’s made clear that RDM is far more complex than RPD, and that it requires more time and resources. However, the focus is on the operational decision coming out of RDM, and whether getting to the decision involved posing and evaluating alternatives, and selecting among them, and the idea that RDM is a process involving a pattern of decision loops, while RPD involves only one loop, seems to be overlooked.

Once we see that RDM is a complex pattern of decision loops then it becomes relevant to ask whether or not these are made up of loops that use RPD? And since, RDM is a multiple decision loop process like the KLC, it also suggests the idea that RDM may be at least a particular kind of KLC? Considering the second question first, clearly RDM does seem to be a KLC since it involves formulating alternative knowledge claims in the form of decision alternatives, and then evaluating and selecting among them. But then we arrive at yet another question, closely related to the first just above, namely whether every KLC must use RDM, or whether we can have multiple decision loop KLCs that use RPD in every one of their decisions?

The answer is that KLCs need not be instances of RDM, but can use RPD to quickly arrive at a single knowledge claim about one’s decision, which is then evaluated quickly by using mental simulation. In brief, KLCs can use RPD or RDM. In fact, things are more complex than that since, in the multiple decision loops of any KLC, the RPD, or RDM option always exists, so that various combinations of RPD and RDM are possible in any KLC.

What’s Rational in Decision Making?

Usually, research studies in NDM and/or RPD make much of the contrast with the RDM approach and in doing so manage sooner or later to imply that RDM is idealistic, excessively normative, and unrealistic to apply in most human decision making situations. They do this often with the clear implication that man is just not rational as the enlightenment and classical economics assumed, and that one of the great gifts of modern social science in general and decision making research over the past 25 years, in particular, is to unmask the fantasy of rationality that we have all been laboring under.

Now, I’m as happy as the next person, to call attention to the simplicity of enlightenment assumptions and the conception of rationality found in classical economics, and also in classical democratic theory, however, I also think it’s a bit unfair to just assume that a particular decision making model is “the” rational model of decision making, while all others are somehow non-rational. From my point of view the RDM model of decision making is not characterized by rationality. Nor is the RPD model non-rational or irrational. The use of these labels is not descriptive of the central features of these models, and I also don’t believe that either gets at the central features of rationality in a modern context.

We can begin to see this more clearly if we consider the distinguishing features of RPD and RDM and the context in which both types of models are used. The first context we discussed above for RPD is that of routine action and learning, where the right thing to do is already known. In that context, provided that previous behavior and its results have been assessed with an open and critical mind and that the first pattern is consistent with these results, it is always rational to apply RPD, simply because we have no reason to believe that our previous knowledge is mistaken. On the other hand, if, because we have failed to see problems due to our allowing what we expect to see to color our experience, or because we haven’t been diligent enough to look for problems that are indicated by weak signals, we apply RPD, then we certainly have departed from “rationality,” in a very meaningful sense of the term. That is, we have applied RPD in a situation where it doesn’t apply because we have refused to open our minds to reality, and that is one of plainest indicators of irrationality there are. So, in the context of routine action, learning, and decision making, RPD may be either rational or irrational to apply depending on the context. And the most important here is that RPD can embody Rational Decision Making in this context, so insofar as anyone claims a monopoly on rationality for the old RDM model I think they are simply mis characterizing rationality.

Moving to the KLC context here too, there are contexts where RPD is either rational or irrational, as the case may be, and where RDM, also is either rational, or irrational. For both RPD and RDM, there are external and internal aspects of rationality. The external aspects relate to whether RPD or RDM should be used in the context of a particular KLC? This question highlights a meta-decision about which type of KLC to use in a given context. This decision may be a routine one which we don’t need a KLC to figure out. That is, it can be obvious that there’s no point in using the RDM because an operational decision needs to be made before a KLC using RDM, that is, one posing and evaluating alternatives, could be completed. Or alternatively it could be equally obvious that there is plenty of time and resources available, and that the operational decision giving rise to the KLC is important enough to warrant application of the RDM. In either case, it’s perfectly rational to select one or the other alternative approach to the KLC, and equally, it would be irrational to deny our previous experience and to decide in favor of RDM, when there is no time to complete the process, or to decide in favor of the RPD when it’s really important to avoid error and we have the time and resources to go through an RDM. In short, when the decision about whether to use RPD or RDM is routine, the rational choice is clear and that meta-decision would not require application of the RDM.

Of course, the meta-decision about whether to apply the RPD or the RDM may not be a routine one. It may not be clear which of these should be applied. If that’s the case, one would have to go through another meta-KLC and to decide whether an RPD or an RDM was appropriate for this new KLC, and then the pattern of choice would repeat. Of course, my intention is not so much to point out that the choice between RPD and RDM can be complex, but rather to point out it’s sometimes not clear or obvious whether RPD or RDM is the appropriate choice for working through a KLC in any context. RDM is not necessarily the rational choice, nor is it necessarily the irrational choice. From an external viewpoint, the assignment of “rationality” to applying one model, and non-rationality or irrationality to another is not always very clear.

Next, once a choice is made about whether a particular KLC will use only RPD, or at least some RDM, then even if the initial choice was rational, that won’t guarantee that the application of either RPD or RDM will be irrational or rational in the specific context. Can an application of RPD in the KLC context be irrational? I think the answer to this is yes, since once a new decision pattern is arrived at by an individual their mental simulation and evaluation of the likely consequences of the new decision may be incoherent or inconsistent. Also, in cases where the contemplated decision is very risky, there may be no attempt at safe-fail experiments, even when there is time for them. Can it be rational? Again, I think the answer is yes, provided that the mental simulation is coherent and consistent, and safe-fail experiments are used where is both time and resources.

The possibility of irrationality is just as great in applying RDM. That is, even assuming that an application of RDM develops alternative decision possibilities, that’s no indication that the RDM will be executed rationally. In particular, the evaluation perspectives used in the RDM process can be quite irrational. They may not require consistency, or coherence, or fair comparison of alternatives. They may rely on authority as the dominant evaluation criterion. They may not employ safe-fail experiments in risky situations to pilot test decisions.

There is yet another difficulty in associating the RDM model with rationality, and that is that there is no longer agreement on the very foundation of the idea of rationality. Modern philosophy has shown that the classical conception of rationality requiring justification of one’s view in terms of some foundation that itself cannot be questioned, is no longer viable. Since there are no certain foundations for our knowledge, this concept of rationality turns out to be limited in the sense that it is always relative to foundations that themselves can’t be justified. So, according to this view, RDM models that implement a process of attempting to justify one alternative decision relative to others have only limited rationality. There’s an alternative concept of rationality available that says that rationality in the RDM requires fair critical comparison of competing decision alternatives and acceptance of the decision that best survives that fair critical comparison. That’s the view I favor, but I think I can safely say that current applications of the RDM don’t embody it. So, in my view they are not rational.

Finally, in addition to showing how RPD and RDM decision making patterns relate to OODA, the DEC, the KLC, I’ve also tried to show that the association of RPD with non-rational, or even irrational decision making, and the contrasting association of RDM with rationality, are both invalid associations. Intuition is not the same as irrationality, and it may frequently be rational to rely on it in decision making. Also, the RDM is not the same as rationality, even when there is ample time and resources to apply it, since RDM can be, and currently is, used in many irrational ways.

… without scaring the wrong people too much?

That’s a question that hovers over the discussions in our avian flu project, both in Ghana and in Ethiopia (and, if I listen to my colleagues’ experience, as well in Indonesia and Nigeria). When mapping out value networks, risk communication and response, we hear the concrete examples: In Ghana, there have been outbreaks that were rather localized and quickly stamped out by concerted government action. In Ethiopia one suspected case turned out to be a different disease, after all the “avian flu action” had been taken.

But when we ask about the economic damage done by avian flu, our partners from government, big and small poultry farms and different trading associations don’t talk about the 5 000, 10 000 or 100 000 birds that were culled on specific farms to prevent a spreading of the disease.

All they talk about is

THE SCARE!

Even though in Ethiopia it comes close to a religious duty to eat chicken on certain holidays, and the Ethiopians seem to be the most religious people I have ever met, in the middle of the avian flu scare, no one wanted to eat their Easter chicken. Farmers who couldn’t sell their fowls, just left them at the market, some farmers even killed and burned all their chicken as a preventive measure, wifes sent their husbands away who had brought Easter chicken from the village, the whole market collapsed. One lady, who owns a medium scale commercial farm (1000+ chicken for egg production) told us: “You would watch TV and get confused: First they showed us all this about always wearing plastic gloves when preparing chicken and reporting any suspicious death etc. and just afterwards they tried to tell us not to panic and to please eat more chicken.”

In the land of taxonomy there are only two species: the Lumpers who gather related things together and look for their commonalities (these are the categorisers), and the Splitters who look for the distinctions between things and separate them based on those distinctions (these are the classifiers).

After a fascinating, robust and sometimes sharp discussion with Larry Prusak and Dave Snowden a couple of days ago in Kuala Lumpur on the topic “Is Knowledge Management Dead?” I have come to the conclusion that I am a Lumper and they are Splitters, when it comes to the utility of the over-arching label of knowledge management. Plus they think that KM as a field has been irredeemably corrupted by the many false plays and hijacks it has been subjected to, while I still have hope. Watch the podcast for the full story, and a million thanks to Larry and Dave for a great conversation. (I’m probably going to get a lot of flak from librarians about a comment I make near the end).

You can download the full video file here.

At the same iKMS session as Gabriel Szulanski (previous post) we aired a pre-recorded podcast with Prof Eric Tsui from the Hong Kong Polytechnic University’s KM Research Centre. He was sharing some interesting work he’s been doing with the KMRC’s clients, around identifying appropriate knowledge sharing tools. The content is great, though the quality of the audio is a bit iffy. Eric’s group is maintaining a Google spreadsheet evaluating a whole range of knowledge sharing and collaboration tools, so if you’re interested in collaborating on this, drop him an email at eric.tsui-at-polyu.edu.hk.

You can download the video file here.

You can download a full set of Prof Tsui’s slides here

Back in April iKMS hosted a talk by Professor Gabriel Szulanski of INSEAD on his fascinating research into knowledge “stickiness” – ie the factors that inhibit knowledge transfer. It was a great talk and here it is in podcast form.

PART ONE: SUCCESS AND FAILURE FACTORS IN KNOWLEDGE TRANSFER

Download the video file here

PART TWO: BARRIERS AND SOME SOLUTIONS

Download the video file here

Professor Szulanski’s slides for the talk can be downloaded here. And the book is here!

That bloody DIK pyramid gets my goat every time. It was used far more than is good for my blood pressure at a conference in Kuala Lumpur this week. It is pervasive and nasty. Martin Fricke has published a rigourous theoretical paper (from the perspective of information science, not explicitly KM) showing exactly why the DIKW pyramid is not only wrong, but also misleading. It generates poor understandings of the relationships between data, information and knowledge, and this in turn leads to poor methodological practices. This observation in particular struck a chord:

“The DIKW theory also seems to encourage uninspired methodology. The view is that data, existing data that has been collected, is promoted to information and that information answers questions. This encourages the mindless and meaningless collection of data in the hope that one day it will be ascend to information—pre-emptive acquisition.”

Fricke cites a hilarious example of the kind of things we can learn from such practices – did you know that Librans tend to break their hips?

While every assault on the pyramid is welcome in my book, it’s a shame that the issue wasn’t addressed as well from the knowledge perspective (ie data is a designed knowledge artefact, not a primitive of information), and that the ludicrousness of the framework will only become obvious to the relatively few people in KM (my guess) who read the Journal of Information Science. Thanks to Margaret Gross via the TaxoCop community for this link.