Gustavo Landfried

Social Anthropologist and Computer Science PhD Student

Gustavo Landfried was born in Buenos Aires, Argentina. He is a social anthropologist currently completing a PhD in computer science at the University of Buenos Aires. When he began his PhD, he was looking for tools to apply to the social sciences, and what he found was better than he expected: the Bayesian approaches answered the deep epistemological and methodological questions he developed as an anthropologist. And to top it off, he found it to be an ideal tool for the social sciences, because by expressing causal models through intuitive graphical methods puts non-academic expert knowledge into the spotlight. After his PhD, he intends to promote its use in the social sciences.

 

Anthropology

Why do we live one way and not another? This question was central to Gustavo’s motivation to pursue anthropology as an undergraduate. “Through studying other cultures, we open our mind. My culture shows me only one way to live.” Learning about other cultures and ways of life different from his own was his way of questioning his own culture.

He discovered many interesting cultures, but he came across one that did not sit well with him: the postmodern tradition, which believes that “truths” cannot be reached within the humanities. He was convinced that the presence of uncertainty did not mean that everything was of equal value. While giving equal value to all points of view may seem more democratic, he realised that its consequences were the opposite. “Not having a way to agree on what is right and what is wrong is a return to the criterion of authority: power as the foundation of truth.”

In the middle of a crisis with his career, he met a group of anthropologists who taught him to think of society and culture as a complex system, in which the properties of the systems are born from the behaviour of their individuals. The group was building computational systems to model different aspects of society, something he had little experience in. This gave him a different lens to view the field and reinspired his passion for anthropology once again. He realised that if he wanted to continue on this path, he had to learn these computational tools. For this reason, when he finished anthropology, he decided to start a computer science career from the very beginning.

 

PhD in Computer Science

Gustavo previously had little experience with computer science, but his determination and the support from the computer science community allowed him to begin a PhD in the field. He has been steadfast in picking up the skills and understanding he needs for combining his passion for anthropology with the practical applicability of these computational tools.

While studying the evolution of strategies in online games, Gustavo came across TrueSkill, a Bayesian solution to the famous chess skill estimation model developed by Microsoft for its video games. The mathematical learning curve was steep in his new field, but by implementing this model from scratch he acquired a deep understanding of the Bayesian approach to probability. Later, he discovered there was a better model, TrueSkill Through Time, that was not yet available in any of the major programming languages because of the difficulty of the mathematics. However, Gustavo was able to quickly implement the algorithm in Python and Julia.

The crossover paid off better than one might imagine. Gustavo found that causal reasoning based on the strict application of the rules of probability could be used to answer the deep epistemological and methodological questions he developed as an anthropologist. The new approach also proved to be an ideal method for correctly assessing hypotheses in terms of the empirical data and formal constraints imposed by models of nature. This discovery opened Gustavo to a new line of optimal decision-making and risk reduction.

 

Intersubjective agreements in contexts of uncertainty: life bets.

The Bayesian approach to probability showed him how to reach intersubjective agreements in the empirical sciences from physics to social science in which propositions must be validated within open systems that always contain some degree of uncertainty. In his explorations, he discovered even more, “the value of truth is not abstract, it is pragmatic.”

Gustavo likens the selection processes of hypotheses in probability theory, by a sequence of predictions, to the selection processes of evolving life forms, by a sequence of survival and reproduction rates. He observed that because the impacts of losses are stronger than those of gains, there is a clear advantage to variants, whether that be hypotheses or life forms, that reduce fluctuations by individual diversification, cooperation, cooperative specialisation, and cooperative heterogeneity, which he calls “epistemic”, “evolutionary”, “speciation”, and “ecological” properties respectively.

Gustavo points out that our own life depends on several levels of cooperation, from the organ structures of multicellular organisms to the societies and ecosystems in which we live, that are essential to survival. He asserts that the same happens with our knowledge: “hypotheses cooperating to form variables, variables cooperating to form causal models, and sets of models cooperating to form theories, or cultures for that matter.”

While it was long assumed that the evolution of cooperation was subject to a dilemma, Gustavo claims that in multiplicative selection processes there is a disadvantage to defection in cooperative groups, since defectors increase the fluctuations of the cooperators on whom they depend, consequently increasing their own fluctuations, and thereby negatively affecting their own long-run growth rate. He believes that the breakdown of the cooperative pact has greater impacts for life and knowledge and sees the problem of overfitting as a direct consequence of selecting a single hypothesis.

 

Plurinational Bayesian Congress

Since he has completed his doctoral activities, he has embarked on a new goal that will guide him for the rest of his professional life: to promote the Bayesian approach in all empirical sciences.

Although probability theory has been well established since the end of the 18th century and has since been adopted as the logic of all empirical sciences, in practice, it was not possible to apply its rules due to the high computational cost of evaluating the entire hypothesis space. Only at the end of the 20th century, thanks to the growth in computational power and the development of sampling methods and analytical approximation, was it possible to apply the rules in a generalised way. However, to Gustavo, the historical inertia is now the most important factor preventing the transition of the research workflow.

Gustavo is organizing the Plurinational Bayesian Congress (bayesdelsur.com.ar). The congress is focused on the applications of Bayesian methods in all data-driven sciences and will be attended by trained researchers from various regions and from a variety of disciplines ranging from Physics to Archaeology. Their aim is to grow the Bayesian communities in Latin America, promoting the adoption of the Bayesian approach as a general method for intersubjective agreements in contexts of uncertainty, in science, society and ecology.

 

You can find more information about Gustavo and his recent publications on his website.