Data as pictures?

Many people believe that data should provide a faithful representation or picture of the real world. While this is often a helpful simplification, it can sometimes mislead.

Firstly, the picture theory isn't very good at handling probability and uncertainty. When faced with alternative pictures (facts), people may try to pick the most likely or attractive one, and then act as if this were the truth. 

As I see it, the problem of knowledge and uncertainty fundamentally disrupts our conventional assumptions about representation, in much the same way that quantum physics disrupts our assumptions about reality. See previous posts on Uncertainty.

Secondly, the picture theory misrepresents judgements (whether human or algorithmic) as descriptions. When a person is classified as a poor credit risk, or as a potential criminal or terrorist, this is a speculative judgement about the future, which is often sadly self-fulfilling. For example, when a person is labelled and treated as a potential criminal, this may make it more difficult for them to live as a law-abiding citizen, and they are therefore steered towards a life of crime. Data of this kind may therefore be performative, in the sense that it creates the reality that it claims to describe.

Thirdly, the picture theory assumes that any two facts must be consistent, and simple facts can easily be combined to produce more complex facts. Failures of consistency or composition can then only be explained (and fixed) in terms of data quality and governance. See my post on Three Responses to Inconsistency (December 2003).

Furthermore, a good picture is one that can be verified. Nothing wrong with verification, of course, but the picture theory can sometimes lead to a narrow-minded approach to validation and verification. There may also be an assumption of completeness, treating a dataset as if it provided a complete picture of some clearly delineated domain. (The world is determined by the facts, and by their being all the facts.)

However, although there are some serious limitations with the picture theory, it may sometimes be an acceptable simplification, or even an enabling prejudice. One of the dimensions of data strategy is reach - developing a broad data culture across the organization and its ecosystem by making more data and tools available to a wider community of people. And if some form of the picture theory helps people get started on the ladder towards data mastery, that may not be a bad thing after all. (Hopefully they can throw away the ladder after they have climbed up it.)

---

 

Daniel C. Dennett, A Difference That Makes a Difference: A Conversation (Edge, 22 November 2017) 

Aaron Sloman, What Did Bateson Mean? (originally posted January 2011, revised October 2018)

See also Architecture and Reality (November 2012), From Sedimented Principles to Enabling Prejudices (March 2013), Data Strategy - Reach (December 2019), On the performativity of data (August 2021)

Architecture and Reality

There are various confusing notions of "real" and "reality" circulating in the enterprise architecture world.

1. The idea that business people are only interested in things that are real.

2. The idea that architectural models represent some form of reality.

3. The idea that data contained in a computer represents some form of reality.

4. The idea that the data contained in a computer is itself a higher form of reality.

5. The idea that one artefact can be a realization of another artefact.

6. The idea that artefacts can be arranged along a realization dimension - some artefacts are more or less real/realized than others. For example, a physical data model is supposedly more real/realized than the logical data model.

Now you could believe some of these, but I don't see how anyone could believe all of them at the same time, without fracturing the concept of reality.

Here are some posts that cover some aspects of this.

Faithful Representation, Faithful Representation 2 (Aug 2008) - on the fallacy of regarding a system as a faithful representation of the real world. Among other things, I argue that the problem of knowledge and uncertainty fundamentally disrupts our conventional assumptions about representation, in much the same way that quantum physics disrupts our assumptions about reality. This led to a further series of posts, offering several examples. Responding to Uncertainty, Responding to Uncertainty 2, Analyzing the Rusty Lawnmower (Aug-Sept 2008). 

Architecture and the Imagination (Oct 2012) - on how architects need to think about things that don't yet exist.

From AS-IS to TO-BE (Oct 2012) - three alternative ways of interpreting the notion of realization. See also Deconstructing the Grammar of Business (June 2009), on the correct meaning of "reification".

Analyzing the Rusty Lawnmower

In my previous post on Responding to Uncertainty, I imagined a garden equipment company responding to an event, based on the appearance (via satellite image) of a rusty lawnmower in John's garden. In that post I was addressing a conceptual question about the nature of the event. In this post I want to discuss a more practical question about the structure and formation of an appropriate event processing network for the garden equipment company.

My starting point is the JDL model of information fusion, via Tim Bass, which would indicate the following event flow architecture.

0. Data collection.

Obtain satellite images.

1. Situation Picture/Event Refinement.

Identify and track some objects, provisionally labelling them "garden", "grass" and "rusty lawnmower".

2. Situation Refinement.

By comparing the map references of these objects with John's address, we associate these objects with John. Furthermore, we can look at the history of these objects over an extended series of images, to see how long they've been there. We can also look at John's history, to see how long he has been living at this address and whether he has bought any gardening stuff in recent years. At this point, we may revise the object labels, because we realise that the rusty object could possibly equate to some other item John bought three years ago.

3. Opportunity/Impact Assessment.

Infer John's intentions about gardening, and the possibility of selling him a new lawnmower.

4. Process Refinement.

Monitor how many lawnmowers we have sold on this basis. Ideally, the process refinement should have some basis for monitoring false negatives (where we didn't spot the rusty lawnmower because it was half-hidden behind an overgrown bush) as well as false positives (where the rusty object was actually an expensive item of sculpture).

The original JDL model refers to these stages as "levels", but the Data Fusion website suggests that this creates confusion because "a hierarchy does not exist from a conceptual point of view". (They might possibly be "levels" in the cybernetic sense, similar to how the term was used by Bateson.)

The model is interesting for several reasons.

• The model supposedly reflects a "natural" cognitive process, including both realtime and historical situational processing.
  
• The event flows can converge at each stage: thus we might have many (possibly heterogeneous) sources of data flowing into one situation picture, or many (possibly contradictory) situation pictures flowing into one impact assessment.
  
• Each stage uses different quantities and qualities of knowledge and interpretation.
• Therefore each stage has different degrees of mission sensitivity. An organization may be comfortable (or may have no choice) with using third party services as event sources, and may be willing to share basic situation pictures with its partners, but may regard the impact assessment as highly confidential.

From a system development perspective, we now have two views of the event processing network - a component oriented view (featuring the concept of the rusty lawnmower, together with patterns and rules for recognizing and responding to instances of this concept) and the flow-oriented view (featuring the stages of the data fusion model). Opher Etzion talks about these two views in his post On Flow-Oriented and Component-Oriented Development. But we probably also need a business/mission view, which drives the requirements in the first place.

The garden equipment company probably didn't start the analysis from the concept of a rusty lawnmower. One possibility is that they started from the idea of some highly generic class of events that would prompt sales and marketing activity. This class of events is then decomposed (top-down) to identify a range of detailed events, some of which might be recognized from satellite images. Another possibility is that they worked backwards from the history of successful lawnmower sales, and then analyzed the relevant satellite images to try and identify any patterns that could be statistically correlated (bottom-up) with this sales history.

In both cases, we need to decouple the data from the explanation of the data. Let's say we find a visual pattern that correlates to lawnmower sales. Our hypothesis is that the pattern indicates a rusty lawnmower, but even if our hypothesis turns out to be incorrect the pattern still somehow works! So we don't throw away the pattern, we just have to look for a different explanation.

This is of course how science has always worked. Many important scientific advances have been based on incorrect explanations. Galileo misunderstood how the telescope worked, but that didn't stop him making some important discoveries. Business networks can be very complex, and sometimes we don't fully understand what is going on, but we still need to build systems that respond as intelligently as possible to what is going on. The point is not to construct a universal theory of lawn-mowers, the point is to sell more lawn-mowers.


Thanks to Opher and Tim for private discussion.

Responding to Uncertainty 2

Following my previous discussion on Faithful Representation and Uncertainty, here's another example. Which of the following statements are facts, straightforwardly representing something in the "real world", and which ones are more problematic?

• John has a garden behind his house. Three quarters of the garden is plain green, presumably grass.
  
• There is an object in the garden visible from the satellite picture. It looks as if it might be a rusty lawnmower. It has been there all winter.
• It is nearly Spring. John might need a new lawnmower soon.

All of these statements might be represented by events, which are fed into an event processing system. As a result, the computer sends John an attractive brochure of garden equipment, including a range of lawnmowers.

Note that many of the observations are uncertain. The grass might be artificial, and the rusty object might be an exercise bike. The garden supply company might wish to purchase higher-definition images, or invest in better image recognition software to improve the interpretation of the raw images, but this investment would be wasted unless there was a good chance of increasing the number of lawnmowers sold.

In any case, there are lots of other things that might influence John's purchasing decision. John might have another lawnmower in the shed. He might be too busy to mow his garden, so he pays Harry (who brings his own mower). Meanwhile, John might want to buy a new lawnmower as a house-warming present for his daughter.

From both a business perspective and an engineering perspective, we can construct effective and profitable systems without bothering our heads whether the rusty lawnmower really exists, or whether it is just an unreliable interpretation of pixels on a satellite image. And what about John's desire for a new lawnmower? Does this really exist, or is this just a bit of hopeful speculation by the marketing department?

So I want to build models that include things like probability, intentionality and the future. Do these things "exist" in the real world, or are they some kind of construction? Rick Murphy explores the question of Representation and Realism in the context of the Semantic Web. In a later post on Signs of the Singularity, he argues that because the semantic web follows Tarski, model theory implies realism. "The relation between a model and the world may be only one of approximation, but without realism, technological utopianism quickly precedes to simulacra and simulation."

I don't think it's as simple as all that. And if it was, would it matter?

---

See follow-up post on Analyzing the Rusty Lawnmower.

Responding to Uncertainty

How does a system respond intelligently to uncertain events?

"A person may take his umbrella, or leave it at home, without any ideas whatsoever concerning the weather, acting instead on general principles such as maximin or maximax reasoning, i.e. acting as if the worst or the best is certain to happen. He may also take or leave the umbrella because of some specific belief concerning the weather. … Someone may be totally ignorant and non-believing as regards the weather, and yet take his umbrella (acting as if he believes that it will rain) and also lower the sunshade (acting as if he believes that the sun will shine during his absence). There is no inconsistency in taking precautions against two mutually exclusive events, even if one cannot consistently believe that they will both occur." [Jon Elster, Logic and Society (Chichester, John Wiley, 1978) p 84]

Austrian physicist Erwin Schrödinger proposed a thought experiment known as Schrödinger's cat to explore the consequences of uncertainty in quantum physics. If the cat is alive, then Schrödinger needs to buy catfood. If the cat is dead, he needs to buy a spade. According to Elster's logic, he might decide to buy both.

At Schrödinger's local store, he is known as an infrequent purchaser of catfood. The storekeeper naturally infers that Schrödinger is a cat-owner, and this inference forms part of the storekeeper's model of the world. What the storekeeper doesn't know is that the cat is in mortal peril. Or perhaps Schrödinger is not buying the catfood for a real cat at all, but to procure a prop for one of his lectures.

Businesses often construct imaginary pictures of their customers, inferring their personal circumstances and preferences from their buying habits. Sometimes these pictures are useful in predicting future behaviour, and for designing products and services that the customers might like. But I think there is a problem when businesses treat these pictures as if they were faithful representations of some reality.

This is an ethical problem as well as an epistemological one. You've probably heard the story of a supermarket, which inferred that some of its female customers were pregnant and sent them a mailshot that presumed they were interested in babies. But this mailshot was experienced as intrusive and a breach of privacy, especially as some of the husbands and boyfriends hadn't even been told yet. (A popular version of the story involves the angry father of a teenaged girl.)

Instead of trying to get the most accurate picture of which customers are pregnant and which customers aren't, wouldn't it be better to construct mailshots that would be equally acceptable to both pregnant and non-pregnant customers? Instead of trying to accurately sort the citizens of an occupied country into "Friendly" and "Terrorist", wouldn't it be better to act in a way that reinforces the "Friendly" category?

Situation models are replete with indeterminate labels like these ("pregnant", "terrorist"), but I think it is a mistake to regard these labels as representing some underlying reality. Putting a probability factor onto these labels just makes things more complicated, without solving the underlying problem. These labels are part of our way of making sense of the world, they need to be coherent, but they don't necessarily need to correspond to anything.


Minor update 10 Feb 2019

Faithful representation 2

In my previous post, Faithful Representation, I discussed the view that a situation model represented some reality, and attributed this view to both Tim Bass and Opher Etzion. However I should have made clear that Tim and Opher don't see things in quite the same way.

Tim's Simple Situation Model is not as simple as Opher’s Simple Situation Model, and it contains things other than events. However, I was under the impression that Tim and Opher were nonetheless each propounding a situation model that accurately (or as accurately as possible) represented some “reality”.

Both have now clarified their respective positions. In On Faithful Representation and Other Comments, Opher points out that his model involves events (in the computer world) representing the situation (in the "real world"), and he doesn't say anything about the situation itself representing anything. Meanwhile in The Secret Sauce is the Situation Models, Tim concurs that we are interested in modelling our knowledge of the real world.

If the model represents our knowledge of the real world, is it possible to measure or analyse the gap between our knowledge and reality itself? Not without a certain amount of philosophical cunning.

Which gives us a problem with uncertainty. In his comment to my earlier post, Opher argued that this problem is orthogonal to the representation problem, but I disagree. I believe that the problem of knowledge and uncertainty fundamentally disrupts our conventional assumptions about representation, in much the same way that quantum physics disrupts our assumptions about reality.

Let's look at the implications of uncertainty for business computing. There are different strategies for dealing with an uncertain situation. One strategy is to determine the most likely situation, based on the available evidence, and then to act as if this situation was the correct one. Another strategy is to construct multiple alternative interpretations of the evidence (possible worlds), and then to find actions that produce reasonable outcomes in each of the possible worlds. The notion that a situation model must be a faithful representation of the Real World makes sense only if we are assuming the first strategy.

For example, in fraud management or security, the first approach uses complex pattern matching to divide transactions into “dodgy” or “okay”. There is then a standard system response for the “dodgy” transactions (including false positives), and a standard system response for the “okay” transactions (including false negatives). Obviously the overall success of the system depends on accurately dividing transactions into the two categories “dodgy” and “okay”. Meanwhile, the second approach might have a range of different system responses, depending on the patterns detected.

A third strategy involves creating intermediate categories: “definitely dodgy”, “possibly dodgy”, “probably okay”, “definitely okay”. In this strategy, however, we are no longer modelling the pure and unadulterated Real World, but modelling our knowledge of the real world. This shifts the question away from the accuracy of the model towards the adequacy of our knowledge.

Faithful representation

Systems people (including some SOA people and CEP people and BPM people) sometimes talk as if a system was supposed to be a faithful representation of the real world.

This mindset leads to a number of curious behaviours.

Firstly, ignoring the potential differences between the real world and its system representation, treating them as if they were one and the same thing. For example, people talking about "Customer Relationship Management" when they really mean "Management of Database Records Inaccurately and Incompletely Describing Customers". Or referring to any kind of system objects as "Business Objects". Or equating a system workflow with "The Business Process".

Secondly, asserting the primacy of some system ontology because "That's How the Real World Is Structured". For example, the real world is made up of "objects" or "processes" or "associations", therefore our system models ought to be made of the same things.

Thirdly, getting uptight about any detected differences between the real world and the system world, because there ought not to be any differences. Rigid data schemas and obsessive data cleansing, to make sure that the system always contains only a single version of the truth.

Fourthly, confusing the stability of the system world with the stability of the real world. The basic notion of "Customer" doesn't change (hum), so the basic schema of "Customer Data" shouldn't change either. (To eliminate this confusion you may need two separate information models - one of the "real world" and one of the system representation of the real world. There's an infinite regress there if you're not careful, but we won't go there right now.)

In the Complex Event world, Tim Bass and Opher Etzion have picked up on a simple situation model of complex events, in which events (including derived, composite and complex events) represent the "situation". [Correction: Tim's "simple model" differs from Opher's in some important respects. See his later post The Secret Sauce is the Situation Models, with my comment.] This is fine as a first approximation, but what neither Opher nor Tim mentions is something I regard as one of the more interesting complexities of event processing, namely that events sometimes lie, or at least fail to tell the whole truth. So our understanding of the situation is mediated through unreliable information, including unreliable events. (This is something that has troubled philosophers for centuries.)

From a system point of view, there is sometimes little to choose between unreliable information and basic uncertainty. If we are going to use complex event processing for fraud detection or anything like that, it would make sense to build a system that treated some class of incoming events with a certain amount of suspicion. You've "lost" your expensive camera have you Mr Customer? You've "found" weapons of mass destruction in Iraq have you Mr Vice-President?

One approach to unreliable input is some kind of quarantine and filtering. Dodgy events are recorded and analyzed, and then if they pass some test of plausibility and coherence they are accepted into the system. But this approach can produce some strange effects and anomalies. (This makes me think of perimeter security, as critiqued by the Jericho Forum. I guess we could call this approach "perimeter epistemology". The related phenomenon of Publication Bias refers to the distortion resulting from analysing data that pass some publication criterion while ignoring data that fail this criterion.)

In some cases, we are going to have to unpack the simple homogeneous notion of "The Situation" into a complex situation awareness, where a situation is constructed from a pile of unreliable fragments. Tim has strong roots in the Net-Centric world, and I'm sure he could say more about this than me if he chose.