Why Being Critical Is Essential: The Horizon Effect
There are situations where things seem to be going well and you can pretend to yourself and those around you that you are doing well for quite a bit longer even though utter failure is unavoidable. This is one aspect of “failing fast”: ask the hard questions first!
In many cases of decision making, it is impractical to plan ahead for every step along the way to ones goal. This is particularly true when there is a lot of uncertainty about the future, as for instance caused by adversarial actors. In these cases, where the space of possibilities that would need to be considered exceeds one’s ability to process them all, the common approach in AI is to use heuristic evaluations of some future situations that seem reachable — they are on the horizon—, and then decide based on those evaluations which strategy to follow in the short-term. The sub-goal is then simply to reach that point on the horizon. As one is getting closer to it, one can start planning beyond that point.
The same kind of inability to project into the future accurately, of course, happens in real-world management, too. One example is when higher-up managers evaluate their reports: Short of being able to measure whether a report is advancing the organization towards its goals, a manager might be inclined to evaluate a report based on the strategy the report has articulated for reaching those goals, and/or based on some metric of progress — we will see in a second that this progress may well be into a dead-end, but I’m getting ahead of myself.
This is where the horizon effect comes in. It describes a problem of using a horizon that is too short. Either because of one’s inability to predict the future far enough, or because one is too busy to think hard about it. Furthermore it illustrates that “too short” really depends on the situation. This figure is a textbook example of the horizon problem, described in terms of chess.
The Black bishop cannot be saved. But a computer playing with too short of a search horizon will not recognize this and will hence unnecessarily sacrifice the three pawns, because it will result in a situation where the bishop is still alive (doomed or not). The heuristic evaluation of the situation is hence much higher than one where the bishop is lost. The shortsighted player does not reason about the fact that the bishop is lost in the long-term, which would allow it to realize that it should not sacrifice the pawns as well — he’d be loosing the bishop and the pawns—but rather give up the bishop already. Clearly, the shorter the horizon, the more likely this effect will cause trouble.
In management the horizon effect offers decision makers the comfort that comes from delay of unavoidable disaster. This comfort comes at the expense of wasting time and resources, of course. Good examples of this include start-ups that try to build the perfect product, without doing any market validation up-front, instead of, e.g., following the lean approach. In that mode, you can set SMART goals, a feature road-map, and milestones, and even do some of those fashionable agile sprints so that everyone around you thinks you know what the heck you are doing. All along then you can check off items on these lists and feel like you are making measurable progress. An uneducated board member might even fall for it. However, if what you are building is useless and nobody will ever be interested to pay money for it in any shape or form, then this is not actually progress. At least not towards your goals: you are just going down a rat hole. The reason for that is that you are just not trying hard enough to fail, i.e., you are not trying hard enough to find the right questions to ask yourself — or, in this case, your prospective customers.
Another example is in C-level management, and also politics where the horizon is set by the next election. In these cases you may succeed at convincing those around you that you are doing things right as long as there is a direction that you can go in for long enough, so as to give the impression of progress.
Overcoming the horizon effect is not easy, but there are things one can do. For instance, you should always try to look at a problem from various different perspectives — in the chess example one could try a different heuristic for comparison, or try to project into the future with lower accuracy but farther. But the main thing is always the same: try to be objective and THINK.
References
[1] Stuart J. Russell, Peter Norvig: Artificial Intelligence — A Modern Approach (3. internat. ed.). Pearson Education 2010, ISBN 978-0-13-207148-2, pp. I-XVIII, 1-1132
