Philosophers interested in the theoretical consequences of predictive processing often assume that predictive processing is an inferentialist and representationalist theory of cognition. More specifically, they assume that predictive processing revolves around approximated Bayesian inferences drawn by inverting a generative model. Generative models, in turn, are said to be structural representations: representational vehicles that represent their targets by being structurally similar to them. Here, I challenge this assumption, claiming that, at present, it lacks an adequate justification. I examine (...) the only argument offered to establish that generative models are structural representations, and argue that it does not substantiate the desired conclusion. Having so done, I consider a number of alternative arguments aimed at showing that the relevant structural similarity obtains, and argue that all these arguments are unconvincing for a variety of reasons. I then conclude the paper by briefly highlighting three themes that might be relevant for further investigation on the matter. (shrink)

The Bayesian brain hypothesis, predictive processing, and variational free energy minimisation are typically used to describe perceptual processes based on accurate generative models of the world. However, generative models need not be veridical representations of the environment. We suggest that they can be used to describe sensorimotor relationships relevant for behaviour rather than precise accounts of the world.

A probabilistic Chomsky–Schützenberger hierarchy of grammars is introduced and studied, with the aim of understanding the expressive power of generative models. We offer characterizations of the distributions definable at each level of the hierarchy, including probabilistic regular, context-free, (linear) indexed, context-sensitive, and unrestricted grammars, each corresponding to familiar probabilistic machine classes. Special attention is given to distributions on (unary notations for) positive integers. Unlike in the classical case where the "semi-linear" languages all collapse into the regular languages, using (...) analytic tools adapted from the classical setting we show there is no collapse in the probabilistic hierarchy: more distributions become definable at each level. We also address related issues such as closure under probabilistic conditioning. (shrink)

This paper presents a version of neurophenomenology based on generative modelling techniques developed in computational neuroscience and biology. Our approach can be described as _computational phenomenology_ because it applies methods originally developed in computational modelling to provide a formal model of the descriptions of lived experience in the phenomenological tradition of philosophy (e.g., the work of Edmund Husserl, Maurice Merleau-Ponty, etc.). The first section presents a brief review of the overall project to naturalize phenomenology. The second section presents and (...) evaluates philosophical objections to that project and situates our version of computational phenomenology with respect to these projects. The third section reviews the generative modelling framework. The final section presents our approach in detail. We conclude by discussing how our approach differs from previous attempts to use generative modelling to help understand consciousness. In summary, we describe a version of computational phenomenology which uses generative modelling to construct a computational model of the inferential or interpretive processes that best explain this or that kind of lived experience. (shrink)

ABSTRACT How can imagination generate knowledge when its contents are voluntarily determined? Several philosophers have recently answered this question by pointing to the constraints that underpin imagination when it plays knowledge-generating roles. Nevertheless, little has been said about the nature of these constraints. In this paper, I argue that the constraints that underpin sensory imagination come from the structure of causal probabilistic generative models, a construct that has been highly influential in recent cognitive science and machine learning. I (...) highlight several attractions of this account, and I favourably contrast it with Peter Langland-Hassan’s account of sensory imagination in terms of the forward models exploited in sensorimotor control. (shrink)

Generative models have been proposed as a new type of non-representational scientific models recently. A generative model is characterized with the capacity of producing new models on the basis of the existing one. The current accounts do not explain sufficiently the mechanism of the generative capacity of a generative model. I attempt to accomplish this task in this paper. I outline two antecedent accounts of generative models. I point out that both (...) types of generative models function to generate new homogenous models in the sense that the latter is a straightforward derivative of the former, both of which share many similar features. Unfortunately, both accounts are implicit about the generative capacity of generative models. Using a case study, I articulate that a two-staged process of abstraction and idealization in modeling may contribute to the generative capacity of a scientific model. I also demonstrate that this two-staged process may go beyond the capacity of generating new homogenous models to generating new heterogeneous models. (shrink)

Does the material basis of conscious experience extend beyond the boundaries of the brain and central nervous system? In Clark 2009 I reviewed a number of ‘enactivist’ arguments for such a view and found none of them compelling. Ward (2012) rejects my analysis on the grounds that the enactivist deploys an essentially world-involving concept of experience that transforms the argumentative landscape in a way that makes the enactivist conclusion inescapable. I present an alternative (prediction-and-generative-model-based) account that neatly accommodates all (...) the positive evidence that Ward cites on behalf of this enactivist conception, and that (I argue) makes richer and more satisfying contact with the full sweep of human experience. (shrink)

A successful vision system must solve the problem of deriving geometrical information about three-dimensional objects from two-dimensional photometric input. The human visual system solves this problem with remarkable efficiency, and one challenge in vision research is to understand howneural representations of objects are formed and what visual information is used to form these representations. Ideal observer analysis has demonstrated the advantages of studying vision from the perspective of explicit generative models and a specified visual task, which divides the (...) causes of image variations into the separate categories of signal and noise. Classification image techniques estimate the visual information used in a task from the properties of “noise” images that interact most strongly with the task. Both ideal observer analysis and classification image techniques rely on the assumption of a generative model. We show here how the ability of the classification image approach to understand how an observer uses visual information can be improved by matching the type and dimensionality of the model to that of the neural representation or internal template being studied. Because image variation in real world object tasks can arise from both geometrical shape and photometric (illumination or material) changes, a realistic image generation process should model geometry as well as intensity. A simple example is used to demonstrate what we refer to as a “classification object” approach to studying three-dimensional object representations. (shrink)

Although discriminative deep neural networks are currently dominant in cognitive modeling, we suggest that capacity-limited, generative models are a promising avenue for future work. Generative models tend to learn both local and global features of stimuli and, when properly constrained, can learn componential representations and response biases found in people's behaviors.

ABSTRACT How can imagination generate knowledge when its contents are voluntarily determined? Several philosophers have recently answered this question by pointing to the constraints that underpin imagination when it plays knowledge-generating roles. Nevertheless, little has been said about the nature of these constraints. In this paper, I argue that the constraints that underpin sensory imagination come from the structure of causal probabilistic generative models, a construct that has been highly influential in recent cognitive science and machine learning. I (...) highlight several attractions of this account, and I favourably contrast it with Peter Langland-Hassan’s account of sensory imagination in terms of the forward models exploited in sensorimotor control. (shrink)

Determining the semantic role of sentence constituents is a key task in determining sentence meanings lying behind a veneer of variant syntactic expression. We present a model of natural language generation from semantics using the FrameNet semantic role and frame ontology. We train the model using the FrameNet corpus and apply it to the task of automatic semantic role and frame identification, producing results competitive with previous work (about 70% role labeling accuracy). Unlike previous models used for this task, (...) our model does not assume that the frame of a sentence is known, and is able to identify null- instantiated roles, which commonly occur in our corpus and whose identification is crucial to natural language interpretation. (shrink)

Determining the semantic role of sentence constituents is a key task in determining sentence meanings lying behind a veneer of variant syntactic expression. We present a model of natural language generation from semantics using the FrameNet semantic role and frame ontology. We train the model using the FrameNet corpus and apply it to the task of automatic semantic role and frame identification, producing results competitive with previous work (about 70% role labeling accuracy). Unlike previous models used for this task, (...) our model does not assume that the frame of a sentence is known, and is able to identify null- instantiated roles, which commonly occur in our corpus and whose identification is crucial to natural language interpretation. (shrink)

We propose, following Clark, that generative models also play a central role in the perception and interpretation of linguistic signals. The data explanation approach provides a rationale for the role of prediction in language processing and unifies a number of phenomena, including multiple-cue integration, adaptation effects, and cortical responses to violations of linguistic expectations.

There are various philosophical interpretations of the account of consciousness associated with the temporal depth of generative models under the Free Energy Principle. This paper strives to develop a new philosophical interpretation of the free energy account of consciousness along the lines of intentionalism.

In this paper, we consider how certain longstanding philosophical questions about mental representation may be answered on the assumption that cognitive and perceptual systems implement hierarchical generative models, such as those discussed within the prediction error minimization framework. We build on existing treatments of representation via structural resemblance, such as those in Gładziejewski :559–582, 2016) and Gładziejewski and Miłkowski, to argue for a representationalist interpretation of the PEM framework. We further motivate the proposed approach to content by arguing (...) that it is consistent with approaches implicit in theories of unsupervised learning in neural networks. In the course of this discussion, we argue that the structural representation proposal, properly understood, has more in common with functional-role than with causal/informational or teleosemantic theories. In the remainder of the paper, we describe the PEM framework for approximate Bayesian inference in some detail, and discuss how structural representations might arise within the proposed Bayesian hierarchies. After explicating the notion of variational inference, we define a subjectively accessible measure of misrepresentation for hierarchical Bayesian networks by appeal to the Kullbach–Leibler divergence between posterior generative and approximate recognition densities, and discuss a related measure of objective misrepresentation in terms of correspondence with the facts. (shrink)

We describe a deep generative model in which the lowest layer represents the word-count vector of a document and the top layer represents a learned binary code for that document. The top two layers of the generative model form an undirected associative memory and the remaining layers form a belief net with directed, top-down connections. We present efficient learning and inference procedures for this type of generative model and show that it allows more accurate and much faster (...) retrieval than latent semantic analysis. By using our method as a filter for a much slower method called TF-IDF we achieve higher accuracy than TF-IDF alone and save several orders of magnitude in retrieval time. By using short binary codes as addresses, we can perform retrieval on very large document sets in a time that is independent of the size of the document set using only one word of memory to describe each document. (shrink)

Long-range correlation, a property of time series exhibiting long-term memory, is mainly studied in the statistical physics domain and has been reported to exist in natural language. Using a state-of-the-art method for such analysis, long-range correlation is first shown to occur in long CHILDES data sets. To understand why, Bayesian generative models of language, originally proposed in the cognitive scientific domain, are investigated. Among representative models, the Simon model was found to exhibit surprisingly good long-range correlation, but (...) {\em not} the Pitman-Yor model. Since the Simon model is known not to correctly reflect the vocabulary growth of natural language, a simple new model is devised as a conjunct of the Simon and Pitman-Yor models, such that long-range correlation holds with a correct vocabulary growth rate. The investigation overall suggests that uniform sampling is one cause of long-range correlation and could thus have a relation with actual linguistic processes. (shrink)

In this paper, I address the issue of scientific modelling in contemporary linguistics, focusing on the generative tradition. In so doing, I identify two common varieties of linguistic idealisation, which I call determination and isolation respectively. I argue that these distinct types of idealisation can both be described within the remit of Weisberg’s :639–659, 2007) minimalist idealisation strategy in the sciences. Following a line set by Blutner :27–35, 2011), I propose this minimalist idealisation analysis for a broad construal of (...) the generative linguistic programme and thus cite examples from a wide range of linguistic frameworks including early generative syntax, Minimalism, the parallel architecture and optimality theory. Lastly, I claim that from a modelling perspective, the dynamic turn in syntax can be explained as a continuation, as opposed to a marked shift, of the generative modelling paradigm. Seen in this light, my proposal is an even broader construal of the generative tradition, along scientific modelling lines. Thus, I offer a lens through which to appreciate the scientific contribution of generative grammar, amid an increased resistance to some of its core theoretical posits, in terms of a brand of structural realism in the philosophy of science and specifically scientific modelling. (shrink)

In recent decades, scientists and philosophers have developed several naturalistic theories of consciousness, in which they try to work out some theoretical foundations for a satisfactory solution to the problem of voluntary acts, in particular the genesis of voluntary bodily movements. From the author’s point of view, depending on which concept of consciousness scientists rely on in their empirical studies of voluntary movements, volition can be understood either as a generative act or as a perceptual act. The first part (...) of the article shows that nowadays there are two competing philosophical and scientific models of volition: the generative model (dualistic and casualistic types) and the perceptual model (strong and weak types). The second part of the article deals with some experimental data from cognitive psychology and neuroscience related to the study of voluntary movements and concludes that they are in favor of the perceptive model of volition. (shrink)

The new wave of ‘foundation models’—general-purpose generative AI models, for production of text (e.g., ChatGPT) or images (e.g., MidJourney)—represent a dramatic advance in the state of the art for AI. But their use also introduces a range of new risks, which has prompted an ongoing conversation about possible regulatory mechanisms. Here we propose a specific principle that should be incorporated into legislation: that any organization developing a foundation model intended for public use must demonstrate a reliable detection (...) mechanism for the content it generates, as a condition of its public release. The detection mechanism should be made publicly available in a tool that allows users to query, for an arbitrary item of content, whether the item was generated (wholly or partly) by the model. In this paper, we argue that this requirement is technically feasible and would play an important role in reducing certain risks from new AI models in many domains. We also outline a number of options for the tool’s design, and summarize a number of points where further input from policymakers and researchers would be required. (shrink)

We present a generative distributional model for the unsupervised induction of natural language syntax which explicitly models constituent yields and contexts. Parameter search with EM produces higher quality analyses than previously exhibited by unsupervised systems, giving the best published unsupervised parsing results on the ATIS corpus. Experiments on Penn treebank sentences of comparable length show an even higher F1 of 71% on nontrivial brackets. We compare distributionally induced and actual part-of-speech tags as input data, and examine extensions to (...) the basic model. We discuss errors made by the system, compare the system to previous models, and discuss upper bounds, lower bounds, and stability for this task. (shrink)

Background:Policemen all over the world are tasked with the heavy work of maintaining social security. With the imbalance in mentality brought about by high population density and social transformation, the work of the Chinese police is particularly hard. As the window of demographic dividend is closing and the number of newborns is insufficient, China has started to adjust its established fertility policy to encourage a family to have two children. However, the results have not met the expectations of the policy (...) adjustment. It is generally believed that factors such as high work pressure, high parenting costs, and low levels of happiness may be the main reasons for low fertility intentions. Studying this typical population of police officers may explore the relationship between work stress, happiness, and reproductive concerns, and provide evidence of Chinese sample.Objectives:To explore the relations between job burnout, subjective well-being, and generativity concern in Chinese police officers.Methods:The study used a cross-sectional survey to collect data from 494 police officers from H city in China. The participants completed the Maslach Burnout Inventory (MBI), the Family Adaptability and Cohesion Scale (FACESII), the Loyola Generativity Scale (LGS), and the Satisfaction with life scale (SWLS). Moderated mediation effect models assessed the association between job burnout, subjective well-being, and generativity concern.Results:Job burnout had a significant negative predictive effect on both subjective well-being and generativity concern, and subjective well-being played a mediating role between job burnout and generativity concern. In addition, family intimacy and adaptability had a significant negative moderating effect between subjective well-being and generativity concern. In a conclusion, there is a moderated mediating effect between job burnout and generativity concern.Conclusion:Subjective well-being played a mediating role between job burnout and generativity concern. (shrink)

We introduce a set of biologically and computationally motivated design choices for modeling the learning of language, or of other types of sequential, hierarchically structured experience and behavior, and describe an implemented system that conforms to these choices and is capable of unsupervised learning from raw natural-language corpora. Given a stream of linguistic input, our model incrementally learns a grammar that captures its statistical patterns, which can then be used to parse or generate new data. The grammar constructed in this (...) manner takes the form of a directed weighted graph, whose nodes are recursively (hierarchically) defined patterns over the elements of the input stream. We evaluated the model in seventeen experiments, grouped into five studies, which examined, respectively, (a) the generative ability of grammar learned from a corpus of natural language, (b) the characteristics of the learned representation, (c) sequence segmentation and chunking, (d) artificial grammar learning, and (e) certain types of structure dependence. The model's performance largely vindicates our design choices, suggesting that progress in modeling language acquisition can be made on a broad front—ranging from issues of generativity to the replication of human experimental findings—by bringing biological and computational considerations, as well as lessons from prior efforts, to bear on the modeling approach. (shrink)

We introduce a set of biologically and computationally motivated design choices for modeling the learning of language, or of other types of sequential, hierarchically structured experience and behavior, and describe an implemented system that conforms to these choices and is capable of unsupervised learning from raw natural‐language corpora. Given a stream of linguistic input, our model incrementally learns a grammar that captures its statistical patterns, which can then be used to parse or generate new data. The grammar constructed in this (...) manner takes the form of a directed weighted graph, whose nodes are recursively (hierarchically) defined patterns over the elements of the input stream. We evaluated the model in seventeen experiments, grouped into five studies, which examined, respectively, (a) the generative ability of grammar learned from a corpus of natural language, (b) the characteristics of the learned representation, (c) sequence segmentation and chunking, (d) artificial grammar learning, and (e) certain types of structure dependence. The model's performance largely vindicates our design choices, suggesting that progress in modeling language acquisition can be made on a broad front—ranging from issues of generativity to the replication of human experimental findings—by bringing biological and computational considerations, as well as lessons from prior efforts, to bear on the modeling approach. (shrink)

Mechanisms consist of component parts and processes organized in a specific way to produce changes that may give rise to one or more phenomena. I aim to examine the generative mechanism of generative material models in the production of new material models. A generative material model in biology is a living material model that is capable of generating new material models. I contend that generative mechanisms of a generative material model are not (...) to be conflated with biological mechanisms: the former is a mechanism manipulated and altered by the modeler while the latter is a naturally occurring mechanism. I examine the ways in which idealization may contribute to the generative mechanism in generating new model organisms. I articulate that the component parts and activities in a generative mechanism are to a large extent shaped by idealization in generative material modeling. Although biological mechanisms do have a role to play in the generation of model organisms, I propose that we shall view the generative mechanism as a hybrid of biological mechanisms and modeling activities. I shall argue that the notion of generative mechanisms as hybrid mechanisms is different from the extant conceptions of mechanisms in various aspects. (shrink)

Generative AI programs can produce high-quality written and visual content that may be used for good or ill. We argue that a credit–blame asymmetry arises for assigning responsibility for these outputs and discuss urgent ethical and policy implications focused on large-scale language models.

The advent of Generative AI, particularly through Large Language Models (LLMs) like ChatGPT and its successors, marks a paradigm shift in the AI landscape. Advanced LLMs exhibit multimodality, handling diverse data formats, thereby broadening their application scope. However, the complexity and emergent autonomy of these models introduce challenges in predictability and legal compliance. This paper analyses the legal and regulatory implications of Generative AI and LLMs in the European Union context, focusing on liability, privacy, intellectual property, (...) and cybersecurity. It examines the adequacy of the existing and proposed EU legislation, including the Artificial Intelligence Act (AIA), in addressing the challenges posed by Generative AI in general and LLMs in particular. The paper identifies potential gaps and shortcomings in the EU legislative framework and proposes recommendations to ensure the safe and compliant deployment of generative models. (shrink)

Learning the structure of event sequences is a ubiquitous problem in cognition and particularly in language. One possible solution is to learn a probabilistic generative model of sequences that allows making predictions about upcoming events. Though appealing from a neurobiological standpoint, this approach is typically not pursued in connectionist modeling. Here, we investigated a sequential version of the restricted Boltzmann machine, a stochastic recurrent neural network that extracts high-order structure from sensory data through unsupervised generative learning and can (...) encode contextual information in the form of internal, distributed representations. We assessed whether this type of network can extract the orthographic structure of English monosyllables by learning a generative model of the letter sequences forming a word training corpus. We show that the network learned an accurate probabilistic model of English graphotactics, which can be used to make predictions about the letter following a given context as well as to autonomously generate high-quality pseudowords. The model was compared to an extended version of simple recurrent networks, augmented with a stochastic process that allows autonomous generation of sequences, and to non-connectionist probabilistic models. We conclude that sequential RBMs and stochastic simple recurrent networks are promising candidates for modeling cognition in the temporal domain. (shrink)