Tom Griffiths

When someone hosts a party, when governments choose an aid program, or when assistive robots decide what meal to serve to a family, decision‐makers must determine how to help even when their recipients have very different preferences. Which combination of people’s desires should a decision‐maker serve? To provide a potential answer, we turned to psychology: What do people think is best when multiple people have different utilities over options? We developed a quantitative model of what people consider desirable behavior, characterizing participants’ preferences by inferring which combination of “metrics” (maximax, maxsum, maximin, or inequality aversion [IA]) best explained participants’ decisions in a drink‐choosing task. We found that participants’ behavior was best described by the maximin metric, describing the desire to maximize the happiness of the worst‐off person, though participant behavior was also consistent with maximizing group utility (the maxsum metric) and the IA metric to a lesser extent. Participant behavior was consistent across variation in the agents involved and tended to become more maxsum‐oriented when participants were told they were players in the task (Experiment 1). In later experiments, participants maintained maximin behavior across multi‐step tasks rather than shortsightedly focusing on the individual steps therein (Experiment 2, Experiment 3). By repeatedly asking participants what choices they would hope for in an optimal, just decision‐maker, and carefully disambiguating which quantitative metrics describe these nuanced choices, we help constrain the space of what behavior we desire in leaders, artificial intelligence systems helping decision‐makers, and the assistive robots and decision‐makers of the future.

Modeling human cognition is challenging because there are infinitely many mechanisms that can generate any given observation. Some researchers address this by constraining the hypothesis space through assumptions about what the human mind can and cannot do, while others constrain it through principles of rationality and adaptation. Recent work in economics, psychology, neuroscience, and linguistics has begun to integrate both approaches by augmenting rational models with cognitive constraints, incorporating rational principles into cognitive architectures, and applying optimality principles to understanding neural representations. We identify the rational use of limited resources as a unifying principle underlying these diverse approaches, expressing it in a new cognitive modeling paradigm called resource-rational analysis. The integration of rational principles with realistic cognitive constraints makes resource-rational analysis a promising framework for reverse-engineering cognitive mechanisms and representations. It has already shed new light on the debate about human rationality and can be leveraged to revisit classic questions of cognitive psychology within a principled computational framework. We demonstrate that resource-rational models can reconcile the mind's most impressive cognitive skills with people's ostensive irrationality. Resource-rational analysis also provides a new way to connect psychological theory more deeply with artificial intelligence, economics, neuroscience, and linguistics.

Identifying patterns in the world requires noticing not only unusual occurrences, but also unusual absences. We examined how people learn from absences, manipulating the extent to which an absence is expected. People can make two types of inferences from the absence of an event: either the event is possible but has not yet occurred, or the event never occurs. A rational analysis using Bayesian inference predicts that inferences from absent data should depend on how much the absence is expected to occur, with less probable absences being more salient. We tested this prediction in two experiments in which we elicited people's judgments about patterns in the data as a function of absence salience. We found that people were able to decide that absences either were mere coincidences or were indicative of a significant pattern in the data in a manner that was consistent with predictions of a simple Bayesian model.

Information changes as it is passed from person to person, with this process of cultural transmission allowing the minds of individuals to shape the information that they transmit. We present mathematical models of cultural transmission which predict that the amount of information passed from person to person should affect the rate at which that information changes. We tested this prediction using a function-learning task, in which people learn a functional relationship between two variables by observing the values of those variables. We varied the total number of observations and the number of those observations that take unique values. We found an effect of the number of observations, with functions transmitted using fewer observations changing form more quickly. We did not find an effect of the number of unique observations, suggesting that noise in perception or memory may have affected learning

If Bayesian Fundamentalism existed, Jones & Love's (J&L's) arguments would provide a necessary corrective. But it does not. Bayesian cognitive science is deeply concerned with characterizing algorithms and representations, and, ultimately, implementations in neural circuits; it pays close attention to environmental structure and the constraints of behavioral data, when available; and it rigorously compares multiple models, both within and across papers. J&L's recommendation of Bayesian Enlightenment corresponds to past, present, and, we hope, future practice in Bayesian cognitive science.

The tendency to test outcomes that are predicted by our current theory (the confirmation bias) is one of the best-known biases of human decision making. We prove that the confirmation bias is an optimal strategy for testing hypotheses when those hypotheses are deterministic, each making a single prediction about the next event in a sequence. Our proof applies for two normative standards commonly used for evaluating hypothesis testing: maximizing expected information gain and maximizing the probability of falsifying the current hypothesis. This analysis rests on two assumptions: (a) that people predict the next event in a sequence in a way that is consistent with Bayesian inference; and (b) when testing hypotheses, people test the hypothesis to which they assign highest posterior probability. We present four behavioral experiments that support these assumptions, showing that a simple Bayesian model can capture people's predictions about numerical sequences (Experiments 1 and 2), and that we can alter the hypotheses that people choose to test by manipulating the prior probability of those hypotheses (Experiments 3 and 4).

Humans can learn complex functional relationships between variables from small amounts of data. In doing so, they draw on prior expectations about the form of these relationships. In three experiments, we show that people learn to adjust these expectations through experience, learning about the likely forms of the functions they will encounter. Previous work has used Gaussian processes—a statistical framework that extends Bayesian nonparametric approaches to regression—to model human function learning. We build on this work, modeling the process of learning to learn functions as a form of hierarchical Bayesian inference about the Gaussian process hyperparameters.

Convolutional neural networks (CNNs) are increasingly widely used in psychology and neuroscience to predict how human minds and brains respond to visual images. Typically, CNNs represent these images using thousands of features that are learned through extensive training on image datasets. This raises a question: How many of these features are really needed to model human behavior? Here, we attempt to estimate the number of dimensions in CNN representations that are required to capture human psychological representations in two ways: (1) directly, using human similarity judgments and (2) indirectly, in the context of categorization. In both cases, we find that low-dimensional projections of CNN representations are sufficient to predict human behavior. We show that these low-dimensional representations can be easily interpreted, providing further insight into how people represent visual information. A series of control studies indicate that these findings are not due to the size of the dataset we used and may be due to a high level of redundancy in the features appearing in CNN representations.

Language research has come to rely heavily on large‐scale, web‐based datasets. These datasets can present significant methodological challenges, requiring researchers to make a number of decisions about how they are collected, represented, and analyzed. These decisions often concern long‐standing challenges in corpus‐based language research, including determining what counts as a word, deciding which words should be analyzed, and matching sets of words across languages. We illustrate these challenges by revisiting “Word lengths are optimized for efficient communication” (Piantadosi, Tily, & Gibson, 2011), which found that word lengths in 11 languages are more strongly correlated with their average predictability (or average information content) than their frequency. Using what we argue to be best practices for large‐scale corpus analyses, we find significantly attenuated support for this result and demonstrate that a stronger relationship obtains between word frequency and length for a majority of the languages in the sample. We consider the implications of the results for language research more broadly and provide several recommendations to researchers regarding best practices.