≡ Menu

Some thoughts on the levels of automation of a decision


A recent post on the IBM Good Decisions BlogWhen Should You Automate a Business Decision? contains a list of ten levels of automation from a paper by Parasuraman, Sheridan and Wickens (available to IEEE members here). This set of 10 levels runs from 1 (computer offers no assistance) to 10 (computer acts autonomously).

The best thing about this list is that it makes it clear how important it is to identify the list of possible decision alternatives before deciding on the amount of automation. The number of people who get in trouble with data mining/predictive analytics and business rules because they don’t fully understand the decisions involved and the possible decision actions is high. As I describe in Chapter 5 of the new book (Decision Management Systems) you need to understand and decompose your decisions into their component parts and make sure you understand the possible actions out of each component decision during the decision discovery phase and before you start writing rules or building analytic models.

I also liked that the list shows that automation can be increased from presentation of options to narrowing of options to taking action on your behalf. Too many projects think this is a black and white choice – manual or automated? In fact the use of Decision Management Systems to narrow the available options to a reasonable or allowed set is a common use case. Particularly when combined with some intelligent presentation of information to the user (so that they can make a good selection from the remaining alternatives) these kinds of systems can be very powerful.

I do have two comments / criticisms.

  • First there was no variation based on the role of the computer in supporting the decision-making of a human.
    The focus of the levels was on how the computer restricts the list of alternatives, how it selects from that list and how it interacts with the human around that selection. While this is a key element of a decision-making system the way in which the computer supports the human decision-maker is also important. Between level 1 (computer does nothing) and level 2 (computer offers complete set of decision alternatives) one could make a compelling case for a system that makes no attempt to make the decision but presents the information the human will need to make a decision. Given how widespread this kind of decision support system is I was struck by its absence. At other levels too it seemed to me that there was a need to differentiate between systems based on how they present information to help the human actor.
  • Second the automation and number of selections were too tied
    There is no differentiation between the behavior of the system in terms of acting automatically, asking for permission or giving an option to veto until the decision alternatives have been reduced to one. I come across many systems where the system has identified multiple alternatives, with one favorite. These systems vary from all options being presented with the systems preference identified, to those where the alternatives are presented but it is clear the system will act on its preferred unless stopped to those where the preferred option is taken but the user has the option to override it after the fact (for a while at least) by selecting an alternative from the list.

What did you think of the list? Would love to hear some other opinions.


Comments on this entry are closed.

  • emmanuel ferret December 2, 2011, 12:14 am

    There is a case where partial automatization is useful. It is when you are not fully able to formalize all your logic in a way interpretable by the machine or when the experts are not really experts.
    In this case you ask the machine to be able to determine the level of difficulty of the asked question. If the asked question is a too difficult one the machine is able to say “I give up”, if it is an easy one the machine tries to answer it, partially or not.
    For example, when the color is really black it is easy to say that the color is black, when the colour is really white it is easy to decide that the color is white, when the color is gray it is not easy to choose between black and white.
    The ability to say “I give up” is useful because you can start to get value from an automated decision without fully mastering the logic ot this decision. You can automate easy answers and leave the touchy ones to the most challenging experts.With the help of this pattern, you can have time to improve continuously your systems. It gives you time to formalize complex logic or, more reasonably, time to gain some more expertise.

    • James Taylor December 2, 2011, 8:37 am

      You are quite correct – another useful use case
      Thanks for the comment