## Heuristics or Algorithms: Confused?

The Editor of the New Scientist (Vol. 3176, 5 May 2018, Letters, p54) opined in response to Adrian Bowyer ‘swish to distinguish between ‘heuristics’ and ‘algorithms’ in AI that:

This distinction is no longer widely made by practitioners of the craft, and we have to follow language as it is used, even when it loses precision.

Sadly, I have to accept that AI folk tend to consistently fail to respect a widely held distinction, but it seems odd that their failure has led to an obligation on the New Scientist – which has a much broader readership than just AI folk. I would agree that in addressing audiences that include significant sectors that fail to make some distinction, we need to be aware of the fact, but if the distinction is relevant – as Bowyer argues, surely we should explain it.

According to the freedictionary:

Heuristic: adj 1. Of or relating to a usually speculative formulation serving as a guide in the investigation or solution of a problem.

Algorithm: n: A finite set of unambiguous instructions that, given some set of initial conditions, can be performed in a prescribed sequence to achieve a certain goal and that has a recognizable set of end conditions.

It even also this quote:

heuristic: of or relating to or using a general formulation that serves to guide investigation  algorithmic – of or relating to or having the characteristics of an algorithm.

But perhaps this is not clear?

AI practitioners routinely apply algorithms as heuristics in the same way that a bridge designer may routinely use a computer program. We might reasonably regard a bridge-designing app as good if it correctly implements best practice in  bridge-building, but this is not to say that a bridge designed using it would necessarily be safe, particularly if it is has significant novelties (as in London’s wobbly bridge).

Thus any app (or other process) has two sides: as an algorithm and as a heuristic. As an algorithm we ask if it meets its concrete goals. As a heuristic we ask if it solves a real-world problem. Thus a process for identifying some kind of undesirable would be regarded as good algorithmically if it conformed to our idea of the undesirables, but may still be poor heuristically. In particular, good AI would seem to depend on someone understand at least the factors involved in the problem. This may not always be the case, no matter how ‘mathematically sophisticated’ the algorithms involved.

Perhaps you could improve on this attempted explanation?

Dave Marsay

## Artificial Intelligence?

The subject of ‘Artificial Intelligence’ (AI) has long provided ample scope for long and inconclusive debates. Wikipedia seems to have settled on a view, that we may take as straw-man:

Every aspect of learning or any other feature of intelligence can be so precisely described that a machine can be made to simulate it. [Dartmouth Conference, 1956] The appropriately programmed computer with the right inputs and outputs would thereby have a mind in exactly the same sense human beings have minds. [John Searle’s straw-man hypothesis]

Readers of my blog will realise that I agree with Searle that his hypothesis is wrong, but for different reasons. It seems to me that mainstream AI (mAI) is about being able to take instruction. This is a part of learning, but by no means all. Thus – I claim – mAI is about a sub-set of intelligence. In many organisational settings it may be that sub-set which the organisation values. It may even be that an AI that ‘thought for itself’ would be a danger. For example, in old discussions about whether or not some type of AI could ever act as a G.P. (General Practitioner – first line doctor) the underlying issue has been whether G.P.s ‘should’ think for themselves, or just apply their trained responses. My own experience is that sometimes G.P.s doubt the applicability of what they have been taught, and that sometimes this is ‘a good thing’. In effect, we sometimes want to train people, or otherwise arrange for them to react in predictable ways, as if they were machines. mAI can create better machines, and thus has many key roles to play. But between mAI and ‘superhuman intelligence’  there seems to be an important gap: the kind of intelligence that makes us human. Can machines display such intelligence? (Can people, in organisations that treat them like machines?)

One successful mainstream approach to AI is to work with probabilities, such a P(A|B) (‘the probability of A given B’), making extensive use of Bayes’ rule, and such an approach is sometimes thought to be ‘logical’, ‘mathematical, ‘statistical’ and ‘scientific’. But, mathematically, we can generalise the approach by taking account of some context, C, using Jack Good’s notation P(A|B:C) (‘the probability of A given B, in the context C’). AI that is explicitly or implicitly statistical is more successful when it operates within a definite fixed context, C, for which the appropriate probabilities are (at least approximately) well-defined and stable. For example, training within an organisation will typically seek to enable staff (or machines) to characterise their job sufficiently well for it to become routine. In practice ‘AI’-based machines often show a little intelligence beyond that described above: they will monitor the situation and ‘raise an exception’ when the situation is too far outside what it ‘expects’. But this just points to the need for a superior intelligence to resolve the situation. Here I present some thoughts.

When we state ‘P(A|B)=p’ we are often not just asserting the probability relationship: it is usually implicit that ‘B’ is the appropriate condition to consider if we are interested in ‘A’. Contemporary mAI usually takes the conditions a given, and computes ‘target’ probabilities from given probabilities. Whilst this requires a kind of intelligence, it seems to me that humans will sometimes also revise the conditions being considered, and this requires a different type of intelligence (not just the ability to apply Bayes’ rule). For example, astronomers who refine the value of relevant parameters are displaying some intelligence and are ‘doing science’, but those first in the field, who determined which parameters are relevant employed a different kind of intelligence and were doing a different kind of science. What we need, at least, is an appropriate way of interpreting and computing ‘probability’ to support this enhanced intelligence.

The notions of Whitehead, Keynes, Russell, Turing and Good seem to me a good start, albeit they need explaining better – hence this blog. Maybe an example is economics. The notion of probability routinely used would be appropriate if we were certain about some fundamental assumptions. But are we? At least we should realise that it is not logical to attempt to justify those assumptions by reasoning using concepts that implicitly rely on them.

Dave Marsay

## The money forecast

A review of The Money forecast A Haldane New Scientist 10 Dec. 2011. On-line version is To Navigate economic storms we need better forecasting.

## Summary

Andrew Haldane, ‘Andy’, is one of the more insightful and – hopefully – influential members of the UK economic community, recognising that new ways of thinking are needed and taking a lead in their development.

He refers to a previous article ‘Revealed – the Capitalist network that runs the world’, which inspires him to attempt to map the world of finance.

“… Making sense of the financial system is more an act of archaeology than futurology.”

Of the pre-crisis approach it says:

“… The mistake came in thinking the behaviour of the system was just an aggregated version of the behaviour of the individual. …

”    Interactions between agents are what matters. And the key to that is to explore the underlying architecture of the network, not the behaviour of any one node. To make an analogy, you cannot understand the brain by focusing on a neuron – and then simply multiplying by 100 billion. …

… When parts started to malfunction … no one had much idea what critical faculties would be impaired.

That uncertainty, coupled with dense financial wiring, turned small failures into systemic collapse. …

Those experiences are now seared onto the conscience of regulators. Systemic risk has entered their lexicon, and to understand that risk, they readily acknowledge the need to join the dots across the network. So far, so good. Still lacking are the data and models necessary to turn this good intent into action.

… Other disciplines have cut a dash in their complex network mapping over the past generation, assisted by increases in data-capture and modelling capability made possible by technology. One such is weather forecasting … .

Success stories can also be told about utility grids and transport networks, the web, social networks, global supply chains and perhaps the most complex web of all, the brain.

…  imagine the scene a generation hence. There is a single nerve centre for global finance. Inside, a map of financial flows is being drawn in real time. The world’s regulatory forecasters sit monitoring the financial world, perhaps even broadcasting it to the world’s media.

National regulators may only be interested in a quite narrow subset of the data for the institutions for which they have responsibility. These data could be part of, or distinct from, the global architecture.

…  it would enable “what-if?” simulations to be run – if UK bank Northern Rock is the first domino, what will be the next?”

I am unconvinced that archeology, weather forecasting or the other examples are really as complex as economic forecasting, which can be reflexive: if all the media forecast a crash there probably will be one, irrespective of the ‘objective’ financial and economic conditions. Similarly, prior to the crisis most people seemed to believe in ‘the great moderation’, and the good times rolled on, seemingly.

Prior to the crisis I was aware that a minority of British economists were concerned about the resilience of the global financial system and that the ‘great moderation’ was a cross between a house of cards and a pyramid selling scheme. In their view, a global financial crisis precipitated by a US crisis was the greatest threat to our security. In so far as I could understand their concerns, Keynes’ mathematical work on uncertainty together with his later work on economics seemed to be key.

Events in 2007 were worrying. I was advised that the Chinese were thinking more sensibly about these issues, and I took to opportunity to visit China in Easter 2008, hosted by the Chinese Young Persons Tourist Group, presumably not noted for their financial and economic acumen. It was very apparent from a coach ride from Beijing to the Great Wall that their program of building new towns and moving peasants in was on hold. The reason given by the Tour Guide was that the US financial system was expected to crash after their Olympics, leading to a slow-down in their economic growth, which needed to be above 8% or else they faced civil unrest. Once tipped off, similar measures to mitigate a crisis were apparent almost everywhere. I also talked to a financier, and had some great discussions about Keynes and his colleagues, and the implications for the crash. In the event the crisis seems to have been triggered by other causes, but Keynes conceptual framework still seemed relevant.

The above only went to reinforce my prejudice:

• Not only is uncertainty important, but one needs to understand its ramifications as least as well as Keynes did (e.g. in his Treatise and ‘Economic Consequences of the Peace’).
• Building on this, concepts such as risk need to be understood to their fullest extent, not reduced to numbers.
• The quotes above are indicative of the need for a holistic approach. Whatever variety one prefers, I do think that this cannot be avoided.
• The quote about national regulators only having a narrow interest seems remarkably reductionist. I would think that they would all need a broad interest and to be exchanging data and views, albeit they may only have narrow responsibilities. Financial storms can spread around the world quicker than meteorological ones.
• The – perhaps implicit – notion of only monitoring financial ‘flows’ seems ludicrous. I knew that the US was bound to fail eventually, but it was only by observing changes in migration that I realised it was imminent. Actually, I might have drawn the same conclusion from observing changes in financial regulation in China, but that still was not a ‘financial flow’. I did previously draw similar conclusions talking to people who were speculating on ‘buy to let’, thinking it a sure-thing.
• Interactions between agents and architectures are important, but if Keynes was right then what really matters are changes to ‘the rules of the games’. The end of the Olympics was not just a change in ‘flows’ but a potential game-changer.
• Often it is difficult to predict what will trigger a crisis, but one can observe when the situation is ripe for one. To draw an analogy with forest fires, one can’t predict when someone will drop a bottle or a lit cigarette, but one can observe when the tinder has built up and is dry.

It thus seems to me that while Andy Haldane is insightful, the actual article is not that enlightening, and invites a much too prosaic view of forecasting. Even if we think that Keynes was wrong I am fairly sure that we need to develop language and concepts in which we can have a discussion of the issues, even if only ‘Knightian uncertainty’. The big problem that I had prior to the crisis was the lack of a possibility of such a discussion. If we are to learn anything from the crisis it is surely that such discussions are essential. The article could be a good start.

The short long. On the trend to short-termism.

Control rights (and wrongs). On the imbalance between incentives and risks in banking.

Risk Off. A behaviorist’ view of risk. It notes that prior to the crash ‘risk was under-priced’.

## Background

This is a personal view of GLS Shackle’s uncertainty. Having previously used Keynes’ approach to identify possible failure modes in systems, including financial systems (in the run-up to the collapse of the tech bubble), I became concerned  in 2007 that there was another bubble with a potential for a Keynes-type  25% drop in equities, constituting a ‘crisis’. In discussions with government advisers I first came across Shackle. The differences between him and Keynes were emphasised. I tried, but failed to make sense of Shackle, so that I could form my own view, but failed. Unfinished business.

Since the crash of 2008 there have been various attempts to compare and contrast Shackle and Keynes, and others. Here I imagine a solution to the conundrum which I deem possible: unless you know different?

## Imagined Shackle

Technically, Shackle seems to focus on the wickeder aspects of uncertainty, to seek to explain them and their significance to economists and politicians, and to advise on how to deal with them. Keynes provides a more academic view, covering all kinds of uncertainty, contrasting tame probabilities with wicked uncertainties, helping us to understand both in a language that is better placed to survive the passage of time and the interpretation by a wider – if more technically aware – audience.

Politically, Shackle lacks the baggage of Lord Keynes, whose image has been tarnished by the misuse of the term ‘Keynesian’. (Like Keynes, I am not a Keynesian.)

Conventional probability theory would make sense if the world was a complicated randomizing machine, so that one has ‘the law of large numbers’: that in the long run particular events will tend to occur with some characteristic, stable, frequency. Thus in principle it would be possible to learn the frequency of events, such that reasonably rare events would be about as rare as we expect them to be. Taleb has pointed out that we can never learn the frequencies of very rare events, and that this is a technical flaw in many accounts of probability theory, which fail to point this out. But Keynes and Shackle have more radical concerns.

If we think of the world as a complicated randomizing machine, then as in Whitehead, it is one which can suddenly change. Shackle’s approach, in so far as I understand it, is to be open to the possibility of a change, recognize when the evidence of a change is overwhelming, and to react to it. This is an important difference for the conventional approach, in which all inference is done on the assumptions that the machine is known. Any evidence that it may have change is simply normalised away. Shackle’s approach is clearly superior in all those situations where substantive change can occur.

Shackle terms decisions about a possibly changing world ‘critical’. He makes the point that the application of a predetermined strategy or habit is not a decision proper: all ‘real’ decisions are critical in that they make a lasting difference to the situation. Thus one has strategies for situations that one expects to repeat, and makes decisions about situations that one is trying to ‘move on’. This seems a useful distinction.

Shackle’s approach to critical decisions is to imagine potential changes to new behaviours, to assess them and then to choose between those deemed possible. This is based on preference not expected utility, because ‘probability’ does not make sense. He gives an example of  a French guard at the time of the revolution who can either give access to a key prisoner or not. He expects to lose his life if he makes the wrong decision, depending on whether the revolution succeeds or not. A conventional approach would be based on the realisation that most attempted revolutions fail. But his choice may have a big impact on whether or not the revolution succeeds. So Shackle advocates imagining the two possible outcomes and their impact on him, and then making a choice. This seems reasonable. The situation is one of choice, not probability.

Keynes can support Shackle’s reasoning. But he also supports other types of wicked uncertainty. Firstly, it is not always the case that a change is ‘out of the blue’. One may not be able to predict when the change will come, but it is sometimes possible to see that there is an economic bubble, and the French guard probably had some indications that he was living in extraordinary times. Thus Keynes goes beyond Shackle’s pragmatism.

In reality, there is no strict dualism between probabilistic behaviour and chaos, between probability and Shackle’s complete ignorance. There are regions in-between that Keynes helps explore. For example, the French guard is not faced with a strictly probabilistic situation, but could usefully think in terms of probabilities conditioned on his actions. In economics, one might usefully think of outcomes as conditioned on the survival of conventions and institutions (October 2011).

I also have a clearer view why consideration of Shackle led to the rise in behavioural economics: if one is ‘being open’ and ‘imagining’ then psychology is clearly important. On the other hand, much of behavioral economics seems to use conventional rationality as some form of ‘gold standard’ for reasoning under uncertainty, and to consider departures from it as a ‘bias’.  But then I don’t understand that either!

I have often noticed that decision-takers and their advisers have different views about how to tackle uncertainty, with decision-takers focusing on the non-probabilistic aspects while their advisers (e.g. scientists or at least scientifically trained) tend to, and may even insist on, treating the problem probabilistically, and hence have radically different approaches to problem-solving. Perhaps the situation is crucial for the decision-taker, but routine for the adviser? (‘The agency problem.’) (Econophysics seems to suffer from this.)

I can see how Shackle had much that was potentially helpful in the run-up to the financial crash. But it seems to me no surprise that the neoclassical mainstream was unmoved by it. They didn’t regard the situation as crucial, and didn’t imagine or deem possible a crash. Unless anyone knows different, there seems to be nothing in Shackle’s key ideas that provide as explicit a warning as Keynes. While Shackle was more acceptable that Keynes (lacking the ‘Keynesian’ label) he also still seems less to the point. One needs both together.

Prigogine , who provides models of systems that can suddenly change ‘become’. He also  relates to Shackle’s discussion on how making decisions relates to the notion of ‘time’.

Dave Marsay

## The Logic of Scientific Discovery

K.R. Popper The Logic of Scientific Discovery 1980 Routledge A review. (The last edition has some useful clarifications.) See new location.

Dave Marsay

## State of the art – software engineering

Composability is a system design principle that deals with the inter-relationships of components. A highly composable system provides recombinant components that can be selected and assembled in various combinations … .”For information systems, from a software engineering perspective,  the essential features are regarded as modularity and statelessness. Current inhibitors include:

“Lack of clear composition semantics that describe the intention of the composition and allow to manage change propagation.”

Composability has a natural interpretation as readiness to be composed with others, and has broader applicability. For example, one suspects that if some people met their own clone, they would not be able to collaborate. Quite generally, composability would seem necessary but perhaps not sufficient to ‘good’ behaviour. Thus each culture tends to develop ways for people to work effectively together, but some sub-cultures seem parasitic, in that they couldn’t sustain themselves on their own.

Cultures tend to evolve, but technical interventions tend to be designed. How can we be sure that the resultant systems are viable under evolutionary pressure? Composability would seem to be an important element, as it allows elements to be re-used and recombined, with the aspiration of supporting change propagation.

## Analysis

Composability is particularly evident, and important, in algorithms in statistics and data fusion.  If modularity and statelessness are important for the implementation of the algorithms, it is clear that there are also characteristics of the algorithms as functions (ignoring internal details) that are also important.

If we partition a given data set, apply a function to the parts and the combine the result, we want to get the same result no matter how the data is partitioned. That is, we want the result to depend on the data, not the partitioning.

In elections for example, it is not necessarily true that a party who gets a majority of the votes overall will get the most candidates elected. This lack of composability can lead to a loss of confidence in the electoral process. Similarly, media coverage is often an editor’s precis of the precis by different reporters. One would hope that a similar story would emerge if one reporter had covered the whole.

More technically, averages over parts cannot, in general, be combined to give a true overall average, whereas counting and summing are composable. Desired functions can often be computed composably by using a preparation function, then composable function, then a projection or interpretation function. Thus an average can be computed by finding the number of terms averaged, reporting the sum and count, summing over parts to give an overall sum and count, then projecting to get the average. If a given function can be implented via two or more composable functions, then those functions must be ‘conjugate’: the same up to some change of basis. (For example, multiplication is composable, but one could prepare using logs and project using exponentiation to calculate a product using a sum.)

In any domain, then, it is natural to look for composable functions and to implement algorithms in terms of them. This seems to have been widespread practice until the late 1980s, when it became more common to implement algorithms directly and then to worry about how to distribute them.

## Iterative Composability

In some cases it is not possible to determine composable functions in advance, or perhaps at all. For example, where innovation can take place, or one is otherwise ignorant of what may be. Here one may look for a form of ‘iterative composability’ in which one hopes tha the results is normally adequate, there will be signs if it is not, and that one will be able to improve the situation. What matters is that this process should converge, so that one can get as close as one likes to the results one would get from using all the data.

Elections under FPTP (first past the post) are not composable, and one cannot tell if the party who is most voter’s first preference has failed to get in. AV (alternative vote) is also not composable, but one has more information (voters give rankings) and so can sometimes tell that there cannot have been a party who was most voters first preference who failed to get in. If there can have been, one could have a second round with only the top parties’ candidates. This is a partial step towards general iterative composability, which might often be iteratively composable for the given situation, much more so than fptp.

Parametric estimation is generally composable when one has a fixed number of entities whose parameters are being estimated. Otherwise one has an ‘association’ problem, which might be tackled differently for the different parts. If so, this needs to be detected and remedied, perhaps iteratively. This is effectively a form of hypothesis testing. Here the problem is that the testing of hypotheses using likelihood ratios is not composable. But, again, if hypotheses are compared differences can be detected and remedial action taken. It is less obvious that this process will converge, but for constrained hypothesis spaces it does.

## Innovation, transformation, freedom and rationality

It is common to suppose that people acting in their environment should characterise their situation within a context in enough detail to removes all but (numeric) probabilistic uncertainty, so that they can optimize. Acting sub-optimally, it is supposed, would not be rational. But if innovation is about transformation then a supposedly rational act may undermine the context of another, leading to a loss of performance and possibly crisis or chaos.

Simultaneous innovation could be managed by having an over-arching policy or plan, but this would clearly constrain freedom and hence genuine innovation. To much innovation and one has chaos, too little and there is too little progress.

A composable approach is to seek innovations that respect each other’s contexts, and to make clear to other’s what one’s essential context is. This supports only very timid innovation if the innovation is rational (in the above sense), since no true (Knightian) uncertainty can be accepted. A more composable approach is to seek to minimise dependencies and to innovate in a way that accepts – possibly embraces – true uncertainty. This necessitates a deep understanding of the situation and its potentialities.

## Conclusion

Composability is an important concept that can be applied quite generally. The structure of activity shouldn’t impact on the outcome of the activity (other than resource usage). This can mean developing core components that provide a sound infrastructure, and then adapting it to perform the desired tasks, rather than seeking to implement the desired functionality directly.

Dave Marsay

## Complexity Demystified: A guide for practitioners

Complexity Demystified: A guide for practitioners, Triarchy Press, 2011.

## First Impressions

• The title comes close to ‘complexity made simple’, which would be absurd. A favourable interpretation (after Einstein) would be ‘complexity made as straightforward as possible, but no more.’
• The references look good.
• The illustrations look appropriate, of suitable quality, quantity and relevance.

Skimming through I gained a good impression of who the book was for and what it had to offer them. This was born out (below).

## Summary

### Who is it for?

Complexity is here viewed from the viewpoint of a ‘coal face’ practitioner:

• Dealing with problems that are not amenable to a conventional managerial approach (e.g. set targets, monitor progress against targets, …).
• Has had some success and shown some insight and aptitude.
• Is being thwarted by stakeholders (e.g., donors, management) with conventional management view and using conventional ‘tools’, such as accountability against pre-agreed targets.

### What is complexity?

Complexity is characterised as a situation where:

• One can identify potential behaviours and value them, mostly in advance.
• Unlike simpler situations, one cannot predict what will be the priorities, when: a plan that is a program will fail.
• One can react to behaviours by suppressing negative behaviours and supporting positive ones: a plan is a valuation, activity is adaptation.

## Details

### Complexity science principles, concepts and techniques

The first two context-settings were well written and informative. This is about academic theory, which we have been warned not to expect too much of; such theory is not [yet?] ‘real-world ready’ – ready to be ‘applied to’ real complex situations – but it does supply some useful conceptual tools.

### The approach

In effect commonplace ‘pragmatism’ is not adequate. The notion of pragmatism is adapted. Instead of persisting with one’s view as long as it seems to be adequate, one seeks to use a broad range of cognitive tools to check one’s understanding and look for alternatives, particular looking out for any unanticipated changes as soon as they occur.

The book refers to a ‘community of practice’, which suggests that there is already a community that has identified and is grappling with the problems, but needing some extra hints and tips. The approach seems down to earth and ‘pragmatic’, not challenging ideologies, cultures, values or other deeply held values.

### Case Studies

These were a good range, with those where the authors had been more closely involved being the better for it. I found the one on Ludlow particular insightful, chiming with my own experiences. I am tempted to blog separately on the ‘fuel protests in the UK in 2000’ as I was engaged with some of the team involved at the time, on related issues. But some of the issues raised here seem quite generally important.

## Interesting points

• Carl Sagan is cited to the effect that the left brain deals with detail, the right with context – the ‘bigger’ picture’. In my opinion many organisations focus too readily on the short term, to the exclusion of the long-term, and if they do focus on the long-term they tend to do it ‘by the clock’ with no sense of ‘as required’. Balancing long-term and short-term needs can be the most challenging aspect of interventions.
• ECCS 09 is made much of. I can vouch for the insightful nature of the practitioners’ workshop that the authors led.
• I have worked with Patrick, so had prior sight of some of the illustrations. The account is recognizable, but all the better for the insights of ECCS 09 and – possibly – not having to fit with the prejudices of some unsympathetic stakeholders. In a sense, this is the book that we have been lacking.

## Related work

### Management

Takes a similar view about complexity and how to work with it.
• The Cynefin Framework.
Positions complexity between complicated (familiar management techniques work) and chaos (act first). Advocates ‘probe-sense-respond’, which reflects some of the same views as ‘complexity demystified. (The authors have discussed the issues.)..

## Conclusions

The book considers all types of complexity, revealing that what is required is a more thoughtful approach to pragmatism than is the norm for familiar situations, together with a range of thought-provoking tools, the practical expediency of some of which I can vouch for. As such it provides 259 pages of good guidance. If it also came to be a common source across many practitioner domains then it could also facilitate cross-domain discussions on complex topics, something that I feel would be most useful. (Currently some excellent practice is being obscured by the use of ‘silo’ languages and tools, inhibiting collaboration and cross-cultural learning.)

The book seems to me to be strongest in giving guidance to practitioners who are taking, or are constrained to take, a phenomenological approach: seeking to make sense of situations before reacting. This type of approach has been the focus of western academic research and much practice for the last few decades, and in some quarters the notion that one might act without being able to justify one’s actions would be anathema. The book gives some new tools which it is hoped will be useful to justify action, but I have a concern that some situations will be stil be novel and that to be effective practitioners may still need to act outside the currently accepted concepts, whatever they are. I would have liked to see the book be more explicit about its scope since:

• Some practitioners can actually cope quite well with such supposedly chaotic situations. Currently, observers tend not to appreciate this extreme complexity of others’ situations, and so under-value their achievements. This is unfortunate, as, for example:
• Bleeding edge practitioners might find themselves stymied by managers and other stakeholders who have too limited a concept of ‘accountability’.
• Many others could learn from such practitioners, or employ their insights.
• Without an appreciation of the complexity/chaos boundary, practitioners may take on tasks that are too difficult for them or the tools at their disposal, or where they may lose stakeholder engagement through having different notions of what is ‘appropriately pragmatic’.
• An organisation that had some appreciation of the boundary could facilitate mentoring etc.
• We could start to identify and develop tools with a broader applicability.

In fact, some of the passages in the book would, I believe, be helpful even in the ‘chaos’ situation. If we had a clearer ‘map’ the guidance on relatively straightforward complexity could be simplified and the key material for that complexity which threatens chaos could be made more of. My attempt at drawing such a distinction is at https://djmarsay.wordpress.com/notes/about-these-posts/work-in-progress/complexity/ .

In practice, novelty is more often found in long-term factors, not least because if we do not prepare for novelty sufficiently in advance, we will be unable to react effectively. While I would never wish to advocate too clean a separation between practice and policy, or between short and long-term considerations, we can perhaps adopt a leaf out of the book and venture some guidance, not to be taken too rigidly. If conventional pragmatism is appropriate at the immediate ‘coal face’ in the short run, then this book is a guide for those practitioners who are taking a step back and considering complex medium term issues, and would usefully inform policy makers in considering the long-run, but does not directly address the full complexities which they face, which are often inherently mysterious when seen from a narrow phenomenological stance. It does not provide guidance tailored for policy makers, and nor does it give practitioners a view of policy issues. But it could provide a much-needed contribution towards spanning what can be a difficult practice / policy divide.