Maps and Territories

Terrain doesn’t fight wars. Machines don’t fight wars. People fight wars. It’s in the minds of men that war must be fought.
— John Boyd (1927-1997)

Developed by maverick military strategist and USAF Colonel John Boyd, the phrase “OODA Loop” refers to the decision cycle of observe, orient, decide, and act. Boyd applied the concept to the combat operations process, often at the strategic level in military operations. Boyd believed that “getting inside the decision cycle of an adversary” is crucial for winning wars. In a recent April 1 issue of the “Breaking Smart” series, Venkatesh Rao formulated the general concept of “map-territory distinction” and explained in detail how finding exploitable weaknesses in the adversary's map can be an important source of competitive advantage.

Red is operating with finger-tip feeling, and has a map of Blue's map. Blue is map-blind, and has no idea what Red is thinking.  Who do you think is going to come out ahead?  (Image Credit:  Breaking Smart ).

Red is operating with finger-tip feeling, and has a map of Blue's map. Blue is map-blind, and has no idea what Red is thinking. Who do you think is going to come out ahead? (Image Credit: Breaking Smart).

According to Rao, maps are used everywhere: geographic maps, organization charts, market evolution maps, genome maps, neural circuit maps, biome maps, sheet music, etc. In competitive situation, there are maps, maps pf maps, maps of maps of maps, etc. One can also make maps of others’ behaviors. Maps can thus be viewed as the basis of all competition. After all, a map is a simplified model of directly experienced reality, or phenomenology in the context of discourse related to the philosophy of science.

Like models, maps are efficient and useful. They reduce the cognitive load of mindful attention to phenomenology via one’s senses. Phenomenological awareness is much more expensive than listening to a model in one’s head. A good map can lower the cost of actions by orders of magnitude. But, like models, there is a hiden cost. When reality changes and catches one unaware, costly failures can occur (e.g., the spectacular failure of LTCM in 1998, or the financial crisis of 2008).

There is also a less dramatic, but more serious, cumulative cost to “map addiction”, according to Rao, i.e., an atrophy of sense-awareness. “Map blindness” turns mere known-unknowns into unknown-unknowns. Almgren and Chriss have this to say about the limitations of all model-driven strategies:

Finally, we note that any optimal execution strategy is vulnerable to unanticipated events. If such an event occurs during the course of trading and causes a material shift in the parameters of the price dynamics, then indeed a shift in the optimal trading strategy must also occur. However, if one makes the simplifying assumption that all events are either "scheduled" or "unanticipated," then one concludes that optimal execution is always a game of static trading punctuated by shifts in trading strategy that adapt to material changes in price dynamics.

The opportunity cost of not developing phenomenological awareness is quite high: one is effectively denied the use of tacit knowledge that has not been organized into maps (or models) in conscious awareness. German World War II military strategists refer to this particular sense-awareness as Fingerspitzengefühl, or “finger-tip feeling”. Unlike closed-loop feedback that signals where the model is wrong and how to adjust and compensate for the discrepancy, finger-tip feeling sensitizes one to the things the model does not even “know” about ( i.e., where the model is not even wrong!).

A pure map-based navigation strategy is what control theorists call open-loop strategy. One simply assumes the map is the territory, and navigate by it with eyes closed. This strategy is very cheap: a decision to not pay attention. Adding error feedback results in a closed-loop strategy, an incremental improvement that is quite a bit costlier. Now one must budget attention based on what the model assumes is important, and navigate by it with eyes wide shut. But a navigation strategy based on finger-tip feeling attempts to eliminate explicit maps from the loop altogether. By “instrumenting the phenomenology” directly, in a manner of speaking, one is finally navigating the territory not only with eyes open, but with an open mind.

In finger-top feeling based navigation, rather than budget attention based on assumed priorities, one deploys attention without importance judgment. This is a stage that precedes map-making and is vastly more expensive in terms of cognitive processing load. But this approach can achieve radical improvements in the long term. Incidentally, this is why recent advances in deep learning technology are widely considered to be significant. By instrumenting phenomenology rather than models, they can make sense of situations the model does not know about. But how does this work exactly?

I’d rather write programs to write programs than write programs.
— Richard Sites

A low-quality map requires a lot of expensive error feedback to just barely function. Sometimes it might even be worse than having no map. A high-quality map, on the other hand, might easily function well even with little feedback. But in competitive situations, one does not win with a better or more detailed map than the adversary. Instead, one wins by using finger-tip feeling to find exploitable weaknesses in the adversary’s map. “Fight the enemy, not the terrain”, as military strategist John Boyd once said. During a crisis, a feedback loop could be worse than an open-loop map; it is an automatic, subconscious habit that can be used against itself to cause a cascade of damage. For example, the Flash Crash of May 6, 2010 can be considered an extreme case of “feedback-amplified map-blindness” among an active subset of the market participants.

Unlike explicit map-and-model building, finger-tip feeling is not a one-time investment. Because the environment and one’s priorities can shift constantly, one has to always allocate a certain amount of attention to “finger-tip feeling” of the territory. One must also keep in mind that phenomenology is not reality; it is merely one’s experience of reality, limited by one’s senses and subconscious mental models. Therefore, it is advantageous to strive for continuous improvement in Fingerspitzengefühl through constant practice and deepening self-awareness; like it is a form of cognitive basic R&D.

Venkatesh Rao recognized the value of multiple models, an insight he gained from an earlier study of map-territory gaps in formal models. When multiple models collide, as Rao observed, they create dissonances; and phenomenology tends to win over all of them. One can thus see reality through the debris. Furthermore, by simply deciding to value phenomenology over maps, one can realize much of the benefit of Fingerspitzengefühl. This happens to be the approach that the MIT roboticist Rodney Brooks had earlier adopted for building his collection of “robotic creatures”, whose “insect-level intelligence” made possible by the underlying “subsumption architecture” was first described in a seminal paper in 1987 titled “Intelligence without Representation”. Brook’s main insight was that AI suffers from abstraction, and that a system cannot reason beyond its representation. So by reacting directly on the real world instead, representations (aka models) become unnecessary, thus greatly simplifying the construction of robots.

Mobile robots at the MIT AI Lab (from left to right): Allen, Tom and Jerry, Herbert (Image Credit:  Rodney Brooks ).

Mobile robots at the MIT AI Lab (from left to right): Allen, Tom and Jerry, Herbert (Image Credit: Rodney Brooks).

BellKor’s Pragmatic Chaos, winners of the Netflix Prize, display their million-dollar check (Image Credit:  Eliot Van Buskirk/ ).

BellKor’s Pragmatic Chaos, winners of the Netflix Prize, display their million-dollar check (Image Credit: Eliot Van Buskirk/

However, there is also value in blending multiple models together. In a surprising turn of events, the winning team of the $1 million dollar Netflix Prize, BellKor’s Pragmatic Chaos, was actually a hybrid team. BellKor (AT&T Research), which won the first Progress Prize milestone in the contest, initially combined with the Austrian team Big Chaos to improve their scores. To pass the 10 percent mark, Quebecois team Pragmatic Theory later joined up to create “BellKor’s Pragmatic Chaos.” The second-place team The Ensemble was also a composite. Arguably, the Netflix Prize’s most convincing lesson is that a disparity of approaches drawn from a diverse crowd is more effective than a smaller number of more powerful techniques. Joining forces allowed both teams to incorporate small, outlying techniques that are relatively inconsequential in the big picture, but crucial during the final stages where tweaking matters most.

“When we were approaching the first progress prize as the BellKor team, there were several other teams that joined together to make a real run at us, and that was surprising to us,” according to Chris Volinsky, originally of team BellKor. “The success of that collaboration told us that this was a real, powerful way to improve our scores. When you’re banging heads together in an office trying to come up with new ideas, you sometimes run out of ideas, and you need to bring in new people into the team, and that turned out to have a great benefit in terms of the predictive power of the models.”

Better solutions come from unorganized people who are allowed to organize organically. But something else also happened that was not entirely expected: Teams that had it basically wrong — but for a few good ideas — made the difference when combined with teams which had it basically right, but couldn’t close the deal on their own. The top two teams beat the challenge by combining teams and their algorithms into more complex algorithms incorporating everybody’s work. The more people joined, the more the resulting team’s score would increase. “One of the big lessons was developing diverse models that captured distinct effects,” commented Joe Sill of The Ensemble, “even if they’re very small effects.”

What lessons might we draw from this that would illuminate the path forward for organizing a community of traders centered on a trading platform? How do we use models? What happens when models collide? How should we blend models? What is the phenomenology of financial trading? Can intelligence emerge from phenomenology?

What form of intelligence would arise...

What form of intelligence would arise...

... when map-territory distinction disappears?

... when map-territory distinction disappears?


  1. Hammond, Grant T. (2012). On the Making of History: John Boyd and American Security. The Harmon Memorial Lecture, 2012. US Air Force Academy. Retrieved from:
  2. Manaugh, Geoff (2010, January 11). Nakatomi Space. BldgBlog. Retrieved from:
  3. Jorion, Philippe (2000, January). Risk Management Lessons from Long-Term Capital Management. European Financial Management, September 2000, pp. 277-300. Retrieved from:
  4. Almgren, Robert and Chriss, Neil (2001). Optimal Execution of Portfolio Transactions. Journal of Risk 3, no. 2. Retrieved from:
  5. Kirilenko, Andrei A. and Kyle, Albert S. and Samadi, Mehrdad and Tuzun, Tugkan (2015, December 28). The Flash Crash: The Impact of High Frequency Trading on an Electronic Market. Retrieved from:
  6. Brooks, Rodney (1991). Intelligence without Representation. Artificial Intelligence 47:139-159. Retrieved from: and
  7. Van Buskirk, Eliot (2009, September 22). How the Netflix Prize was Won. Wired. Retrieved from:
  8. Van Buskirk, Eliot (2009, September 29). BellKor’s Pragmatic Chaos Wins $1 Million Netflix Prize by Mere Minutes. Wired. Retrieved from:
  9. Koren, Yehuda (2009, August). The BellKor Solution to the Netflix Grand Prize. Retrieved from:
  10. Mlnarik, Hynek, and Ramamoorthy, Subramanian and Savani, Rahul (2009). Multi-Strategy Trading Utilizing Market Regimes. Retrieved from:  and