The effect of inheritance on teams of problem solvers

Abstract

Inheriting solutions to everyday problems that were previously discovered by other people is foundational to cumulative cultural evolution. Cultural inheritance enables groups to discover solutions that individuals would be unlikely to discover on their own. But how does inheritance affect problem solving? Does inheritance simply give individuals a head start on a problem, or does it actually change the way the problem is solved? The purpose of this research is to investigate whether teamwork organized around inheritance–diachronic teamwork–is a unique form of collaboration available to human teams, or whether the problem solving ability of diachronic teams reduces to abilities of the individual team members. In part 1, I compare the effectiveness of diachronic teams to alternative strategies for allocating the same number of labor hours. In part 2, I investigate the real-world consequences of diachronic teamwork by studying changes in Wikipedia article quality as function of edits. In part 3, I use simulations of problem solvers navigating artificial problem landscapes to explore how complementary abilities between team members might contribute to the effectiveness of diachronic collaboration. The goal of this research is to provide a framework for assessing the impact of inheritance on problem solving, and an attempt to discover scenarios in which inheritance improves problem solving efficiency.

On average Wikipedia accumlates 6-10 edits per second. Individual articles can accumulate tens of thousands of edits over their lifetime, and are never considered complete. Incremental edits to an encyclopedia article represent the type of collaboration that is necessary for any species to exhibit cumulative cultural evolution. Cumulative cultural evolution depends on the ability to pass the cultural products of one generation on to be inherited and used by the next generation. In the case of Wikipedia, existing articles are inherited by editors and passed on to future editors. More generally, for any human invention, existing innovations are inherited by future inventors.

Cumulative cultural evolution depends on the ability to inherit existing cultural products, but it also implicitly depends on the ability of future generations to incrementally improve those products. To put it simply, the Wikipedia editing strategy has helped Wikipedia grow to become the world’s foremost source of free knowledge not because anyone can make edits, but because anyone can make improvements. No matter how small or large each improvement, an article that has been edited thousands of times is very likely better off than it was after only a handful of edits. But how efficient is this type of collaboration–where the products of one generation are passed on to be inherited and continued by the next generation? The purpose of this research is to measure the efficiency of teamwork organized around inheritance, or what I call diachronic teamwork.

In part 1, I will report the results of experiments conducted to investigate the effectiveness of diachronic teams as compared to alternative forms of allocating the same number of labor hours on the same problem solving task. Three kinds of team strategies are compared: diachronic teams working in sequence, synchronic teams working in parallel, and isolated individuals working for the same total amount of time. My hypothesis is that if diachronic teams can be shown to outperform both synchronic teams and isolated individuals on a given task, then we can safely conclude that diachronic collaboration represents a unique form of problem solving available to humans working together in teams.

In part 2, I investigate the consequences of diachronic teamwork in the context of Wikipedia. The Wikipedia editing strategy outlines a procedure for only allowing incremental improvements to articles over generations of edits. Edits that do not improve article quality are supposed to be reverted. How effective is this strategy in resulting in monotonic increases in article quality? How confidently can Wikipedia readers assume that the current or extant version of the article is higher quality than all previous or extinct versions? My propsed methods involve training machine learning algorithms to predict article and edit quality, and using the predictions of these models to assess the likelihood that article quality continually improves over generations of edits. The results of this analysis would provide a new way of measuring the success of the Wikipedia editing strategy.

In part 3, I use simulations of problem solvers navigating artificial problem landscapes to explore the conditions under which diachronic teams are likely to outperform alternative labor strategies. In the simulations, I investigate the types of problem landscapes best suited for diachronic collaboration. The advantage of these simulations is that they allow for individual skill level to be formally equated across different strategies while solving different kinds of problems. In addition, I varying the distribution of skill level in a team to assess how the effectiveness of diachronic collaboration is related to skill distribution. Are teams of heteroscadastic as opposed to homoscedastic individuals more or less benefited by diachronic collaboration? The value of the simulations lies in their ability to systematically explore any tradeoffs identified in the parts 1 and 2 regarding the efficiency of diachronic collaboration.

The remainder of this proposal describes each part of the dissertation in more detail. I briefly motivate each experiment based on the relevant literature, discuss the methods used, and present some preliminary results. I also identify the remaining work that will be included in the final dissertation.