‘Robots’ vs environmental managers

Automated model-based algorithms compete against humans in conservation games


  • Computer games allow us to compare human-based decisions against model-based decisions
  • On average the ‘robots’ made better decisions than humans using intuition
  • There is real value in the greater use of quantitative methods in environmental management

In this game you will be asked to manage a hypothetical Coho salmon fishery.

In this game you will be asked to manage a hypothetical Coho salmon fishery.

Given all the real world complexities involved when managing ecosystems, do quantitative methods (which ignore most of these complications) really help decision makers achieve better environmental outcomes? How do these quantitative methods compare to the alternative: humans making decisions based on intuition, experience and their best judgement?

Unfortunately, it is difficult to answer this question in real life because experiments in management are usually not repeatable. That is, once a manager ‘acts’ based on their experience, it’s usually impossible to compare the results to how well an alternative decision, aided by a mathematical model, would have performed.

But what is difficult or impossible in real life can sometimes be achieved in the virtual world. Computer games allow us to pit human-based decisions against automated, model-based decisions (a.k.a. ‘robots’ following simple mathematical rules). For every game a human plays (using intuition), a robot plays using predetermined instructions which are optimal given some quantitative model. The model, being a simplification, is an approximation meaning it is always, to some degree, wrong.

We explored this approach using environmental science university students (Holden and Ellner, 2016) as our test humans. We had them play an online computer game where the players tried to harvest a hypothetical salmon population in order to maximise long-run, sustainable profit.

If the player harvests too few fish, then they don’t make much money. But, if they harvest too many fish early in the game, there are no fish in the ocean to harvest during the later turns. The player decides how many fish to take out of the ocean on each turn balancing the future benefits of leaving fish in the ocean against the present profits from fishing. To allow the students to gain some experience managing the fishery, all players played a practice game before playing the game for a score.

Unbeknown to the students, robots were competing against them behind the scenes, making decisions based on simple mathematical rules. Even when these rules were based on completely incorrect descriptions about how the salmon population changed through time, the automated player still on average made better decisions than humans using intuition and their past experience (from playing a practice game).

This shows there is real value in the greater use of quantitative methods in environmental management. We’re not saying that humans should be removed from the decision making process. Humans will be absolutely necessary for defining conservation goals (and revising them), engaging stakeholders, choosing appropriate models and analysing data, and deploying on the ground actions. While we would argue for increased transparency and objectivity (which modelling can help to provide), we are still a long way away from robots taking over the field of environmental management.

Can lessons from games with robots inform environmental management in the real world? (Photo by Megan Saunders)

Can lessons from games with robots inform environmental management in the real world? (Photo by Megan Saunders)

More info: Matthew Holden m.holden1@uq.edu.au


Holden MH & SP Ellner (2016). Human judgement vs. quantitative models for the management of ecological resources. Ecological Applications. doi: 10.1890/15-1295 http://onlinelibrary.wiley.com/doi/10.1890/15-1295/full

1 comment on “‘Robots’ vs environmental managers”

  1. cris brack Reply

    Yea robots! My PhD back in 1991 found “robots” made better forest management decisons for forest plantations than experienced forest managers too. My most sophisticated “robot” even managed to learn how to “cheat” (recommending something none of the humans considered) that allowed it to really dominate the answers.

Leave a Reply