As modern deep networks become more complex, and get closer to human‐like capabilities in certain domains, the question arises as to how the representations and decision rules they learn compare to the ones in humans. In this work, we study representations of sentences in one such artificial system for natural language processing. We first present a diagnostic test dataset to examine the degree of abstract composable structure represented. Analyzing performance on these diagnostic tests indicates a lack of systematicity in representations (...) and decision rules, and reveals a set of heuristic strategies. We then investigate the effect of training distribution on learning these heuristic strategies, and we study changes in these representations with various augmentations to the training set. Our results reveal parallels to the analogous representations in people. We find that these systems can learn abstract rules and generalize them to new contexts under certain circumstances—similar to human zero‐shot reasoning. However, we also note some shortcomings in this generalization behavior—similar to human judgment errors like belief bias. Studying these parallels suggests new ways to understand psychological phenomena in humans as well as informs best strategies for building artificial intelligence with human‐like language understanding. (shrink)

Biases in cognition are ubiquitous. Social psychologists suggested biases and stereotypes serve a multifarious set of cognitive goals, while at the same time stressing their potential harmfulness. Recently, biases and stereotypes became the purview of heated debates in the machine learning community too. Researchers and developers are becoming increasingly aware of the fact that some biases, like gender and race biases, are entrenched in the algorithms some AI applications rely upon. Here, taking into account several existing approaches that address the (...) problem of implicit biases and stereotypes, we propose that a strategy to cope with this phenomenon is to unmask those found in AI systems by understanding their cognitive dimension, rather than simply trying to correct algorithms. To this extent, we present a discussion bridging together findings from cognitive science and insights from machine learning that can be integrated in a state-of-the-art semantic network. Remarkably, this resource can be of assistance to scholars (e.g., cognitive and computer scientists) while at the same time contributing to refine AI regulations affecting social life. We show how only through a thorough understanding of the cognitive processes leading to biases, and through an interdisciplinary effort, we can make the best of AI technology. (shrink)

Meta-learned models of cognition make optimal predictions for the actual stimuli presented to participants, but investigating judgment biases by constraining neural networks will be unwieldy. We suggest combining them with cognitive process models, which are more intuitive and explain biases. Rational process models, those that can sequentially sample from the posterior distributions produced by meta-learned models, seem a natural fit.

We demonstrate that the key components of cognitive architectures (declarative and procedural memory) and their key capabilities (learning, memory retrieval, probability judgment, and utility estimation) can be implemented as algebraic operations on vectors and tensors in a high‐dimensional space using a distributional semantics model. High‐dimensional vector spaces underlie the success of modern machine learning techniques based on deep learning. However, while neural networks have an impressive ability to process data to find patterns, they do not typically model high‐level cognition, and (...) it is often unclear how they work. Symbolic cognitive architectures can capture the complexities of high‐level cognition and provide human‐readable, explainable models, but scale poorly to naturalistic, non‐symbolic, or big data. Vector‐symbolic architectures, where symbols are represented as vectors, bridge the gap between the two approaches. We posit that cognitive architectures, if implemented in a vector‐space model, represent a useful, explanatory model of the internal representations of otherwise opaque neural architectures. Our proposed model, Holographic Declarative Memory (HDM), is a vector‐space model based on distributional semantics. HDM accounts for primacy and recency effects in free recall, the fan effect in recognition, probability judgments, and human performance on an iterated decision task. HDM provides a flexible, scalable alternative to symbolic cognitive architectures at a level of description that bridges symbolic, quantum, and neural models of cognition. (shrink)

In a series of three behavioral experiments, we found a systematic distortion of probability judgments concerning elementary visual stimuli. Participants were briefly shown a set of figures that had two features (e.g., a geometric shape and a color) with two possible values each (e.g., triangle or circle and black or white). A figure was then drawn, and participants were informed about the value of one of its features (e.g., that the figure was a “circle”) and had to predict the value (...) of the other feature (e.g., whether the figure was “black” or “white”). We repeated this procedure for various sets of figures and, by varying the statistical association between features in the sets, we manipulated the probability of a feature given the evidence of another (e.g., the posterior probability of hypothesis “black” given the evidence “circle”) as well as the support provided by a feature to another (e.g., the impact, or confirmation, of evidence “circle” on the hypothesis “black”). Results indicated that participants’ judgments were deeply affected by impact, although they only should have depended on the probability distributions over the features, and that the dissociation between evidential impact and posterior probability increased the number of errors. The implications of these findings for lower and higher level cognitive models are discussed. (shrink)