Modeling Human Concepts with Subspaces in Deep Vision Models

Dynamics and Hierarchical Encoding of Non-compact Acoustic Categories in Auditory and Frontal Cortex.

Diagnosticity in Category Learning by Classification and Inference

Do artifact concepts have cores?

This study examined the effect of linguistic context on the graded structure representation of common categories in adults with traumatic brain injury (TBI). Graded structure indicates that all members of a category are not equally representative with some members being better examples than others. Ten adults who had suffered traumatic brain injuries as the result of motor vehicle accidents and 10 age- and gender-matched neurologically intact adults participated in the study. The experimental task consisted of presenting each participant with 20 contextual sentences and asking them to select the best example of a category label mentioned in each sentence. Each of the 20 sentences were followed by six exemplars: four exemplars of the common category label mentioned in the sentence and two exemplars that were members of other categories. The specific exemplar of the category label was inferred by the context meaning. The neurologically intact adults were significantly more accurate than the TBI adults in choosing the best category example. However, error patterns were similar for each group with both groups choosing a significantly higher proportion of true unrelated exemplars than any other type of error. Results are discussed relative to the process of restructuring common category graded structure as a result of semantic constraints imposed by linguistic context and limitations in cognitive flexibility observed in TBI adults.

Two experiments produced evidence that category relationships contribute to similarity ratings. In Experiment 1, participants gave similarity ratings with respect to a semantic category (VEGETABLE) and a set of exemplars, some of which were members of the category (e.g., BROCCOLI) and some of which were not (e.g., BANANA). A regression analysis was used to predict the similarity ratings in terms of numbers of common and distinctive features, as reported by other participants. Perceived similarity was greater for examplars that were members of the category, independently of feature overlap. Experiment 2 examined similarity ratings with respect to pairs of exemplars. In some cases, both exemplars were members of the same category (e.g., BROCCOLI/CUCUMBER). In other cases, one exemplar was a member of the category and the other was not (e.g., BROCCOLI/BANANA). A regression analysis was used to predict the similarity ratings in terms of numbers of common and distinctive features. Perceived similarity was greater when both exemplars were members of the same category, independently of feature overlap.

As AI Systems become increasingly autonomous they are expected to engage in complex moral decision-making processes. For the purpose of guidance of such processes theoretical and empirical solutions have been sought. In this research we integrate both theoretical and empirical lines of thought to address the matters of moral reasoning in AI Systems. We reconceptualize a metanormative framework for decision-making under moral uncertainty within the Discrete Choice Analysis domain and we operationalize it through a latent class choice model. The discrete choice analysis-based formulation of the metanormative framework is theory-rooted and practical as it captures moral uncertainty through a small set of latent classes. To illustrate our approach we conceptualize a society in which AI Systems are in charge of making policy choices. In the proof of concept two AI systems make policy choices on behalf of a society but while one of the systems uses a baseline moral certain model the other uses a moral uncertain model. It was observed that there are cases in which the AI Systems disagree about the policy to be chosen which we believe is an indication about the relevance of moral uncertainty.

Humans can effortlessly categorize objects, both when they are conveyed through visual images and spoken words. To resolve the neural correlates of object categorization, studies have so far primarily focused on the visual modality. It is therefore still unclear how the brain extracts categorical information from auditory signals. In the current study, we used EEG (n = 48) and time-resolved multivariate pattern analysis to investigate 1) the time course with which object category information emerges in the auditory modality and 2) how the representational transition from individual object identification to category representation compares between the auditory modality and the visual modality. Our results show that 1) auditory object category representations can be reliably extracted from EEG signals and 2) a similar representational transition occurs in the visual and auditory modalities, where an initial representation at the individual-object level is followed by a subsequent representation of the objects' category membership. Altogether, our results suggest an analogous hierarchy of information processing across sensory channels. However, there was no convergence toward conceptual modality-independent representations, thus providing no evidence for a shared supramodal code.NEW & NOTEWORTHY Object categorization operates on inputs from different sensory modalities, such as vision and audition. This process was mainly studied in vision. Here, we explore auditory object categorization. We show that auditory object category representations can be reliably extracted from EEG signals and, similar to vision, auditory representations initially carry information about individual objects, which is followed by a subsequent representation of the objects' category membership.

AI systems have attained superhuman performance across various domains. If the hidden knowledge encoded in these highly capable systems can be leveraged, human knowledge and performance can be advanced. Yet, this internal knowledge is difficult to extract. Due to the vast space of possible internal representations, searching for meaningful new conceptual knowledge can be like finding a needle in a haystack. Here, we introduce a method that extracts new chess concepts from AlphaZero, an AI system that mastered chess via self-play without human supervision. Our method excavates vectors that represent concepts from AlphaZero's internal representations using convex optimization, and filters the concepts based on teachability (whether the concept is transferable to another AI agent) and novelty (whether the concept contains information not present in human chess games). These steps ensure that the discovered concepts are useful and meaningful. For the resulting set of concepts, prototypes (chess puzzle-solution pairs) are presented to experts for final validation. In a preliminary human study, four top chess grandmasters (all former or current world chess champions) were evaluated on their ability to solve concept prototype positions. All grandmasters showed improvement after the learning phase, suggesting that the concepts are at the frontier of human understanding. Despite the small scale, our result is a proof of concept demonstrating the possibility of leveraging knowledge from a highly capable AI system to advance the frontier of human knowledge; a development that could bear profound implications and shape how we interact with AI systems across many applications.

Here is a simple and appealing idea about the way people decide whether an object belongs to a category: The object is a member of the category if it is sufficiently similar to known category members. To put this in more cognitive terms, if you want to know whether an object is a category member, start with a representation of the object and a representation of the potential category. Then determine the similarity of the object representation to the category representation. If this similarity value is high enough, then the object belongs to the category; otherwise, it does not. For example, suppose you come across a white three-dimensional object with an elliptical profile; or suppose you read or hear a description like the one I just gave you. You can calculate a measure of the similarity between your mental representation of this object and your prior representation of categories it might fit into. Depending on the outcome of this calculation, you might decide that similarity warrants calling the object an egg, perhaps, or a turnip or a Christmas ornament.

Integration of contextual reward motivation with category representations.

The ability to take information learned about one object (e.g., a cat) and extend it to other objects (e.g., a tiger, a lion) makes human learning efficient and powerful. How are these inductive generalizations performed? Fisher, Godwin, and Matlen (2015) proposed a developmental mechanism that operates exclusively over the perceptual and semantic features of the objects involved (e.g., furry, carnivorous); this proposed mechanism does not use information concerning these objects' category memberships. In the present commentary, we argue that Fisher and colleagues' experiments cannot differentiate between their feature-based mechanism and its category-based competitors. More broadly, we suggest that any proposal that does not take into account the central role of category representations in children's mental lives is likely to mischaracterize the development of inductive generalization. The key question is not whether, but how, categories are involved in children's generalizations.

Two experiments were carried out to make clear the mechanism of inferences and the representation of semantic categories with the instantiating information. In Experiment 1, verification times of instantiating-context-dependent category membership were measured for 24 graduate and undergraduate students. Verification times in RT (related true instance) condition were faster than those in UT (unrelated true instance). But those in RF (related false instance) condition were equal to those in UF (unrelated false instance). In verifying the context-dependent category, not forward association inferences from categories with the instantiating information to instance but backward association inferences from instances to instantiating contexts occurred. The results suggested that there weren't the instantiated representations of categories. In Experiment 2, categories with the instantiating information and instances were used as the predicate term in deductive reasoning. Judgements of validity and their reaction times under time pressure were measured for 24 graduate and undergraduate students. Percentages of valid responses were high in valid figure, and reaction times in universal negative proposition were longer. It was concluded that, when the tacit premises are introduced, backward association inferences are much more used than forward association inferences.

When time shifts the boundaries: Isolating the role of forgetting in children’s changing category representations

Family resemblances: Studies in the internal structure of categories

This article studies the joint roles of similarity and frequency in determining category structure. Perceptual classification learning experiments were conducted in which presentation frequencies of individual exemplars were manipulated. The exemplars had varying degrees of similarity to members of the target and contrast categories. Classification accuracy and typicality ratings increased for exemplars presented with high frequency and for members of the target category that were similar to the high-frequency exemplars. Typicality decreased for members of the contrast category that were similar to the high-frequency exemplars. A frequency-sensitive similarity-to-exemplars model provided a good quantitative account of the classification learning and typicality data. The interactive relations among similarity, frequency, and categorization are considered in the General Discussion. Among the most well-established findings in the categorization literature is that categories have graded structures (Rips, Schoben, & Smith, 1973; Rosch, 1973, 1978; Rosch & Mervis, 1975; Smith & Medin, 1981). Rather than all instances of a category being equal, it appears that certain instances are better examples than others. For example, people reliably rate a robin as a better example of the category birds