Overcoming organisational stupidity

A brief introduction to collective computation

“I, Pencil, simple though I appear to be … yet, not a single person on the face of this earth knows how to make me.” — Leonard E. Read


What makes an organisation stupid, even when it is full of intelligent people?

An assembly of intelligent individuals does not automatically produce intelligent outputs. The computational ability of the group is the overriding factor. Collective computation has a logic unto itself. It is a system-level phenomenon that requires a system-level approach.

Examples of collective computation

Collective computation is ubiquitous in nature. Here are just a few examples:

Even though they lack neurons, slime moulds can solve complex computation problems. They can navigate labyrinth mazes, anticipate periodic events, avoid previously exploited areas, choose between different food options, and much more1.

Termites can build intricate complexes that use passive solar energy to regulate air temperature and oxygen levels.2 In Brazil, an underground city covering the equivalent area of Great Britain has supported a single terminate nation for approximately 4,000 years. These termites navigate the vast tunnel network using pheromone 'maps' to minimise travel time.3

Pigtailed macaques use a status signaling network (visual subordination signals comprising of a sort of smile) to compute power structures, formalise consensus, and moderate conflicts.4

There is no centralised control in these examples. No project office, no HR department, and no C-Suite. Collective computation is decentralised and distributed. There are thousands of examples in nature.

Algorithms and circuits

To get a basic handle on collective computation, we need to consider two elements: algorithms and circuits.

Algorithms produce output states from input states. We are quite familiar with them from individual computational tasks. Every time we use our smartphone to find the best route we run a navigation algorithm. Every time we look in the fridge to work out what to prepare for dinner we run a choice algorithm.

In collective computation, we also need to consider how the individual agents interact with each other and process information as a system. For this, we need to consider the ‘circuitry’ that connects them. This circuitry can be physical, such as the web of nano-circuits that enable cells to orchestrate their activities5, or virtual, such as the status signaling network of the macaques discussed above. The circuits are more than simple ‘wires’ connecting agents, they also enable logical operations. It is perhaps easiest to explain via an example.

In the above schematic, two alternative circuits for a customer service unit are shown.

In version A, a customer problem that cannot be resolved on the spot by an agent is escalated to a supervisor asynchronously (via a case management system, for example). The supervisor then shunts the problem to an expert, who analyses the problem and later returns a potential solution to the supervisor, who in turn pushes it to another agent who finally informs the customer. This circuit comprises six asynchronous hops. At each hop, noise can introduce an error (e.g., a summary report might miss an important piece of information). In this example, an error was introduced in hop 3 and propagated throughout the rest of the circuit resulting in an erroneous solution being offered to the customer.

In version B, a customer problem that cannot be resolved on the spot by an agent is escalated to an expert synchronously (via a phone call, for example). An error in this example was not encountered because the synchronous conversation provided an effective error-correction mechanism (e.g., the expert asked the agent to elaborate).

In this example, version B reduces the number of potential sources of noise and error thus providing better collective computational effectiveness.

The success of circuit B rests on the agents having access to more information, namely the information of whom and how to escalate a problem (greater autonomy). A supervisor role is redundant in terms of their coordination function however the team will probably need a mechanism to self-organise access to the expert’s time. This is precisely what organisations such as ING are doing when implementing Agile customer service.

It is important to note that the diagram is not a workflow. It is a circuit diagram of information channels and logic. Computation occurs at a deeper layer than workflow. The computational components and circuitry just process information and might include AI’s and chatbots. The purpose of collective computation is to reduce the noise, error, and uncertainty that underpins work.

The above example showed the information flow and logic of just one particular form of computation: customer problem resolution. There are many different types of computation that occur inside organisations. For example, a diagram that describes the decision-making circuitry of a committee would look very different. It would include the heuristics and power dynamics that drive decision making and consensus formation: the graph of who defers to whom.

Why it matters

We’ve all experienced the phenomena of organisational stupidity:6 when smart people have collectively made dumb decisions. Some would argue that it is the norm.

Stupidity is inevitably a computation error. Once we understand what circuits and algorithms are operating in our teams and organisations, we can design potentially radical improvements.

I have only scratched the surface of collective computation in this article and will expand upon the theory and language in future transmissions.

Leave a comment

Learn More

A collection of articles and papers can be found at the Collective Computation Group @SFI

This is a very good introductory video: The Collective Computation of Reality in Nature and Society

I will be exploring this topic in more detail in the near future, please subscribe here:

Image Credits

Circuits: my training slides

Social Media banner: An Allegory of Folly (early 16th century) by Quentin Matsys

1

Subash K. Ray et al., ‘Information Transfer During Food Choice in the Slime Mold Physarum Polycephalum’, Frontiers in Ecology and Evolution 7 (2019), doi:10.3389/fevo.2019.00067.

2

Alexander Heyde, Lijie Guo, Christian Jost, Guy Theraulaz, L. Mahadevan. ‘Self-organized biotectonics of termite nests.’ Proceedings of the National Academy of Sciences, 2021; 118 (5): e2006985118 DOI: 10.1073/pnas.2006985118

3

Stephen J. Martin, Roy R. Funch, Paul R. Hanson, Eun-Hye Yoo. ‘A vast 4,000-year-old spatial pattern of termite mounds.’ Current Biology, 2018; 28 (22): R1292 DOI: 10.1016/j.cub.2018.09.061

4

J. C. Flack and D. C. Krakauer, ‘Encoding Power in Communication Networks.’, The American Naturalist 168, no. 3 (1 September 2006): E87–102, doi:10.1086/506526

5

Duan, J., Navarro-Dorado, J., Clark, J.H. et al. The cell-wide web coordinates cellular processes by directing site-specific Ca2+ flux across cytoplasmic nanocourses. Nat Commun 10, 2299 (2019). https://doi.org/10.1038/s41467-019-10055-w

6

Samaneh Karimi-Ghartemani, Naser Khani, and Ali Nasr Esfahani, ‘Developing a Framework for Organizational Stupidity’, The Journal of High Technology Management Research 31, no. 2 (1 November 2020): 100392, doi:10.1016/j.hitech.2020.100392.