A Framework for Putting AI in its Place

In his 2007 Harvard Business Review article, a former IBMer named Dave Snowden published his framework for managing intellectual capital, which he called the Cynefin (pronounced kin-eff-in) Framework, borrowing the Welsh word for habitat, or place.

The framework describes five domains of knowledge:

Cynefin_as_of_1st_June_2014.png
  • Obvious
  • Complicated
  • Complex
  • Chaotic
  • Disordered (middle area)

Although his graphic representation resembles a typical four-square diagram, lacking x-y axes it is not, strictly speaking.

Moving counter-clockwise from the obvious realm:

  • The obvious domain is the realm of known knowns. It is governed by rules and the relationship between cause and effect is clear; if x, then Y. In this realm, the problem solver is required to sense-categorize-respond. That is, it must establish the facts, categorize them, then respond in accordance with established rule(s) or best practice(s). 
  • The complicated domain is the realm of known unknowns. Determining the relationship between cause and effect requires expertise, as there is a range of right answers. In this case, the bot must sense-analyze-respond. While rules still apply, the problem solver must analyze the facts to determine which rules to apply.
  • The complex domain is the realm of unknown unknowns. Rules are insufficient for understanding and the relationship between cause and effect can only be deduced in retrospect; there are no right answers. To function in this realm, the problem solver must probe-sense-respond, meaning research or experimentation is required up front to establish relevant facts.
  • In the chaotic domain events are too confusing to allow for knowledge-based response. Here, the appropriate pattern in act-sense-respond, meaning first act to establish order, then sense the impact of the action(s) and continue responding to transform the situation from chaos to complex.
  • Finally, the disordered domain is the realm of confusion. The best a problem solver can do in this domain is break the situation in to constituent parts to determine which of the other realms apply and work from there.

Clearly, bots can play a role in the obvious realm, but what is less clear is that they can play a role in the complicated realm as well. In fact, IBM Watson advertising is squarely focused on Watson's ability to solve problems in the complicated domain.

On the other hand, AI as we know it remains ill suited for the complex, chaotic and disordered realms because computers, no matter how powerful or sophisticated, must have rules to follow. In my last post I referenced the AI assistant HAL from the film 2001: A Space Odyssey and the fact that the computer itself made the decision to kill the crew in as a way of resolving a conflict in programmed instructions. This is an example of an AI assistant attempting to solve a problem in a realm where it didn't belong.

While one message is that AI is inherently limited in realm, another message is that AI is potentially suited for realms beyond our present thinking. The primary factor in determining AI's usefulness is the availability and viability of knowledge, in other words, data. AI needs a lot of accurate data to perform effectively. 

The obvious place to start with AI is in the realm of known knowns. Here knowledge is the greatest and uncertainty the least. But don't stop there. Allow yourself to consider the known unknowns, those problems that can be solved through the sense-analyze-respond process provided sufficient data exists to perform the necessary analysis. Medical practitioners readily and confidently use AI to assist in their diagnoses and treatment plans, thanks to the ability of supercomputers to process millions of patient records in a matter of seconds, a feat the human brain cannot equal.

Lastly, when thinking about how AI might fit into your HR service delivery model, resist the temptation of focusing on hard-dollar benefits. AI may indeed pay for itself through headcount reductions, but remember that those reductions will be at least partly offset by new work associated with building, maintaining and generally supporting new AI tools. Instead, focus on ways in which bots can improve the employee experience by doing appropriate tasks better, faster and more conveniently. In particular, think about how AI can not only expand but improve the self-service experience for employees and managers. The payoff will come.