Reflections on the DIKW Pyramid

Stan Garfield recently re-shared his 2015 post "Yet Another Myth: The DIKW Pyramid Scheme," which motivated me to share my own thoughts about DIKW here.

In case you are not familiar with the DIKW Pyramid, it usually looks something like this:

The DIKW model typically shows Data, Information, Knowledge, and Wisdom respectively. It might also include additional layers such as Facts or Understanding. The basic idea is that each layer of the pyramid is built on the layer below it, forming a hierarchy. But a hierarchy of what? That's the big question and the source of much of the criticism of the model.

Stan's article provides a wealth of links to papers, articles, and blog posts criticizing the model, which I urge you to read. And the reason there is so much criticism is because the model has somehow become widespread throughout the knowledge management literature without having much of a solid theoretical foundation behind it. Although the creation of the model is often credited to systems theorist Russell Ackoff, the DIKW Pyramid Wikipedia entry suggests that no one really knows who first came up with the concept, although most agree that the initial germ of the idea can be found in these lines in The Rock by T.S. Eliot:

Where is the wisdom we have lost in knowledge?

Where is the knowledge we have lost in information?

(But note, there's no pyramid in there anywhere!)

When it comes to theoretical models, we humans seem to really like pyramids (just do an image search on "pyramid model theory" and see what you find). Maybe it's because they are simple to comprehend and give the appearance of stability. Maybe it's because they are easy to build, both in the physical world and in PowerPoint. But the visual image of a pyramid can also come with several inherent biases:

• It implies an upward direction. A given level relies on the levels below it, but lower levels aren't affected by the levels above them.
• It implies an order of priority. Upper levels are viewed as more important or complex than lower levels because they are "higher."
• It implies a winnowing or refinement taking place from level to level because the pyramid gets narrower as it ascends.

While at first glance it might seem like a pyramid is the best way to represent a set of related concepts, even the most stable-looking pyramids can collapse over time. Here are examples of two of the most famous:

• Maslow's Hierarchy of Needs is often represented as a pyramid, but Maslow never envisioned it that way. According to this Scientific American article: "it's clear from his writings that he did not view his hierarchy of needs like a video game-- as though you reach one level and then unlock the next level, never again returning to the 'lower' levels. He made it quite clear that we are always going back and forth in the hierarchy, and we can target multiple needs at the same time."
• The U.S. Department of Agriculture's famous Food Pyramid was scrapped in 2011 in favor of a plate, which better conveys the proportions of the kinds of foods Americans should eat without implying any kind of priority.

So if we similarly scrap the DIKW Pyramid, what should we replace it with? Unfortunately, it may not be possible to represent these concepts in a neat, digestible, iconic graphic image (sorry, some things just can't be reduced to pictures). If you read the sources Stan provides in his article, you'll see two-dimensional graphs, ladders, networks, DIKW as a 4-dimensional concept space, etc. So of course not content with any of those, I'll throw another framework into the mix. I like to think of DIKW as a system with data as the fundamental "particles" in the system at one level of analysis. Then information, knowledge, and wisdom subsequently arise as emergent properties visible through the interactions, relationships, and patterns that get created at each lower level.

Let me provide a few analogies to help explain what I mean:

When you drag a file from one folder to another on your computer desktop, it certainly feels like you're dragging something. But what appear as files and folders on your screen at one level are pixels lighting up in a particular sequence at a lower level. At a level below that they are compiled instructions being executed. Below that they are ones and zeros and below that they are electrons bouncing around. At each level, the patterns of the objects at that level of analysis create the emergent properties and epiphenomena that are seen at the next level up when the scale of observation changes.

We could say that a written language system consists of letters, words, sentences, and messages or stories. We assemble letters to form words, words to form sentences, and sentences to form stories. There is certainly a hierarchy here but there is no directionality. The reader is influenced by the words and sentences chosen to tell the story, but a writer who has a particular story in mind might choose sentences and specific words in order to better convey that story. They might even change the spelling of a word to express a dialect or personality characteristic. At the bottom level, letters that happen to assemble into a particular pattern create a word, which has a meaning that is not just a property of the collection of letters. Similarly, a sentence contains a meaning that can't be derived from from the individual words in it but instead comes from the pattern created when they are grouped together. As an example, I'm reminded of the old classic:

Time flies like an arrow.

Fruit flies like a banana.

These sentences both have the words "flies" and "like," and those words are even in the same relative position in each sentence. But they clearly have vastly different meanings, meanings that don't come solely from the words themselves.

For my final example, consider these scientific disciplines:

Sociology

Psychology

Biology

Chemistry

Physics

Again there is clearly a hierarchy here, but what is studied at each level are the emergent properties created by the patterns formed by the components at the level below it. And while each level relies on the level below it to create those patterns, higher levels can also influence the kinds of patterns that are created at the lower levels.

• Hydrogen and oxygen aren't "wet," but they can combine to form something that is. But in the other direction, we have been able to use the periodic table (a pattern at the chemistry level) to identify the holes in that pattern, then search for and put together the necessary physical components to create elements that otherwise wouldn't exist in nature.
• Chemical elements come together to form an emergent property we call "life." Evolution is an emergent process that takes place at the biological level, but over time through random chance and natural selection it has created compounds at the chemistry level that otherwise wouldn't exist.
• Genes affect how we behave, but the environments we create as part of that behavior can affect and alter how those genes are in turn expressed in our offspring. Nature and nurture aren't isolated processes but continually influence each other.

So back to DIKW. It seems to me that data, information, knowledge, and wisdom combine to form a system similar to those I described above. Individual data elements will form patterns that create information, bits of information can clump together to form knowledge, etc. Systems like this have some common attributes:

• The higher the level, the more abstract the concept becomes and the harder it is to define that concept and identify the emergent properties and rules that govern it. At bottom, we know exactly how many letters there are in a written language system -- 26 in the case of the English alphabet. At the next level, we have a reasonably good idea of how many words there are and what they mean (although new ones are invented occasionally) because we have dictionaries. But although we have some rules around what constitutes a sentence, the number of possible sentences is uncountable. And there are even fewer rules as to what constitutes a story. Similarly, physics has some very strict, well-defined rules as to how its components behave, while psychology and sociology are much fuzzier. Likewise, while there are quibbles about the exact definition of data, it's a much clearer concept than knowledge. And what even is wisdom? (I admit I have been cringing at even including it in this discussion).
• The levels can interact and modify each other bi-directionally. I mentioned above how words and stories can affect each other, and how genes and environment can interact with and modify each other as well. While the traditional DIKW Pyramid has wisdom building on knowledge building on information building on data, modifications can also occur in the opposite direction. This mainly comes about because the patterns we see at any given level are often incomplete, as in the periodic table example. We might have some data forming a pattern that constitutes information, but that pattern has holes in it, which motivates us to go looking for new data to fill out or verify the pattern. As we acquire that data, we might fill in the gaps but we might alternatively discover that the pattern we thought we saw was in error and that there is actually a different pattern the data form. The same could be true of the interface between information and knowledge.
• Just because a level is "higher" doesn't necessarily mean it is more complex. Sometimes it's just the opposite. When you move your mouse to drag and drop a file, it's a simple action at that level. But the activities that occur actually become more complex as you move down the levels of the system. When molecules are heated, they move in very complex ways that are almost impossible to map out and predict at that level, but when you average them all together you get temperature, a pretty simple concept. We have a lot of fairly straightforward reliable heuristics that predict how people will behave under certain conditions but that we are still struggling to describe at the level of neural activity. So just because knowledge is a fuzzier concept than data doesn't necessarily mean that the rules we might uncover to describe behavior at that level are more complex.

Lastly, I think it might be useful to think about describing each level in terms of the kinds of relationships that cluster the elements at each underlying level to form patterns. Here's my first pass at it although I am certainly open to and welcome other interpretations:

• Information gets created when data form a pattern based on why we care about their relationship to each other. If we don't care or can't see a relationship, they are just random bits of data.
• Knowledge is often described as information in action. It is often said that "everyone complains about too much information but nobody ever complains about too much knowledge." So knowledge gets created when information forms a pattern that shows how we can use it. If we can't use it, it's by definition just useless information.
• Wisdom gets created when knowledge forms a pattern that shows when to use it. We might know many ways to achieve a goal or accomplish a task, but wisdom is knowing which knowledge to apply under what circumstances.

As I said, these are thoughts that came to mind on reading Stan's great article and the references he provided. I certainly welcome any ideas that might critique, modify, or enhance the arguments I made here. And if anyone can come up with a catchy, succinct way to show all this in a simple graphic, please do!