HUMANS AND SMART MACHINES AS PARTNERS IN THOUGHT?
Large language models (LLMs) like LaMDA, GPT-3, and ChatGPT have been the subject of widespread discussion. This workshop focuses on an analysis of interactions with LLMs. Assuming that not all interactions can be reduced to mere tool use, we ask in what sense LLMs can be part of a group and take on the role of conversational partners. Can such disparate partners as humans and smart machines form a group that takes not only linguistic actions (a conversation) but also other actions, such as producing text or making decisions? To address these questions, both the attributions of abilities to individual group participants and the ways in which the abilities of such groups can be described will be examined. This raises new questions for the field of social ontology, namely whether there are "social kinds" that are not exclusively constituted by humans. On the other hand, debates about the constitution of groups and their agency can contribute to analyzing the interactions of humans and smart machines. We expect to promote a dialogue among philosophers dealing with social groups, linguists, and artificial intelligence, respectively.
To register, please fill in this form:
Registration deadline: 1 April 2023, 23:59 PST. Registration confirmations will be send out after the deadline.
Welcome & Café
Keith Frankish (online):
Eric Schwitzgebel & Anna Strasser:
Ned Block (online):
David Chalmers (online):
Walk & Talk
Daniel Dennett: We are all Cherry-Pickers
Large Language Models are strangely competent without comprehending. This means they provide indirect support for the idea that comprehension, (“REAL” comprehension) can be achieved by exploiting the uncomprehending competence of more mindless entities. After all, the various elements and structures of our brains don’t understand what they are doing or why, and yet their aggregate achievement is our genuine but imperfect comprehension. The key to comprehension is finding the self-monitoring tricks that cherry-pick amongst the ever-more-refined candidates for comprehension generated in our brains.
Eric Schwitzgebel & Anna Strasser: Asymmetric joint actions
What are we doing when we interact with LLMs? Are we playing with an interesting tool? Do we enjoy a strange way of talking to ourselves? Or do we, in any sense, act jointly when chatting with machines? Exploring conceptual frameworks that can characterize in-between phenomena that are neither a clear case of mere tool use nor fulfill all the conditions we tend to require for proper social interactions, we will engage in the controversy about the classification of interactions with LLMs. We will discuss the pros and cons of ascribing some form of agency to LLMs so they can at least participate in asymmetric joint actions.
Ned Block: Large Language Models are more like perceivers than thinkers
I will argue that LLM processing corresponds much better to human perceptual processing rather than human thought. Perceptual representations lack logical structure and perceptual processing is not rule governed, simulating inference without being truly inferential. LLM representation and processing share these properties. Human thought is hierarchically structured—our thinking is characterized by dendrophyllia. But perception and LLM processing is not.
David Chalmers: Do large language models extend the mind?
Keith Frankish: Playing a language game: An interpretivist perspective on LLMs
There is a strong case for the interpretivist view that mentalistic descriptions are justified by their predictive utility: a system has a set of beliefs, desires, and intentions if interpreting it as having them offers significant predictive leverage. Large language models such as GPT-3 present a challenge to this view. The models produce replies that seem intelligent, witty, and informative, and it is tempting to interpret them as possessing a rich array of beliefs, desires, and communicative intentions. Yet they produce their replies by purely statistical means, and it is not plausible to regard them as having any grasp of the meaning of what they say. Does this show that interpretivism is wrong? I shall argue that it does not, and that interpretivists are in fact well placed to explain the difference between LLMs and human language users. I shall show that, despite appearances, LLMs are not properly interpretable as possessing communicative intentions. Rather, they are best interpreted as making moves in a narrowly defined language game, and in so far as they have mental states, the states in question relate merely to the game, not to the multifarious conversational goals humans have when they play it. As well as responding to the challenge to interpretivism, the discussion will also shed light on aspects of human language use.
Paula Droege: Full of sound and fury, signifying nothing
Meaning, language, and consciousness are often taken to be inextricably linked. On this Fregean view, meaning appears before the conscious mind, and when grasped forms the content of linguistic expression. The consumer semantics proposed by Millikan breaks every link in this chain of ideas. Meaning results from a co-variation relation between a representation and what it represents, because that relation has been sufficiently successful. Consciousness is not required for meaning. More surprising, meaning is not required for language. Linguistic devices, such as words, are tools for thought, and like any tool, they can be used in ways other than originally designed. Extrapolating from this foundation, I will argue that Large Language Models produce speech in conversation with humans, because the resulting expression is meaningful to human interpreters. LLMs themselves have no mental representations, linguistic or otherwise, nor are they conscious. They nonetheless are joint actors in the production of language in Latour’s sense of technological mediation between goals and actions.
Joshua Rust: Minimal Institutional Agency
I make three, nested claims: first, while “machines with minds” (Haugeland 1989, 2) are paradigmatic artificial agents, such devices are not the only candidates for artificial agency. Like machines, our social institutions are constructed by us. And if some of these institutions are agential, then they should qualify as artificial agents. Second, if some institutions qua artificial agents also have genuine agency, that agency isn’t the full-blown intentional agency exhibited by human beings, but would fall under a more generic or minimal conception of agency. Moreover, since enactivists aim to articulate a minimal conception of agency that is applicable to all organisms, this suggests that enactivist accounts of minimal agency might be brought to bear on some institutions. The third claim concerns which enactivist notion of generic or minimal agency is relevant to the identification of corporate agents. Where some enactivists stress a protentive orientation to a persistence goal, others defend an account of minimal agency that only requires a model-free and retentive sensitivity to precedent. While enactivists have only glancingly applied these accounts of minimal agency to the question of corporate agency, I argue that those within the sociological tradition of structural functionalism have rigorously pursued the thought that corporate agents must be protentively oriented to a persistence goal. Criticisms of structural functionalism in particular and the protentive account of minimal agency in general motivate the conclusion that corporate agency is better construed in terms of the model-free and retentive account of minimal agency. I also claim that Ronald Dworkin (1986, 168) and Christian List and Philip Pettit (2011, 56) have defended versions of the idea that a sensitivity to precedent is a condition for institutional agency. Returning to the first claim, perhaps an investigation into the various modalities of artificial agency might be mutually illuminating. If a sensitivity to precedent is among the markers of genuine agency in institutional systems, perhaps this is also the case for more paradigmatic exemplars of artificial intelligence.
Ophelia Deroy: Ghosts in the machine - why we are and will continue to be ambivalent about AI
Large language models are one of several AI-systems that challenge traditional distinctions - here, between thinking and just producing strings of words, there, between actual partners and mere tools. The ambivalence, I argue, is there to stay, and what is more, it is not entirely irrational: We treat AI like ghosts in the machine because this is simple, useful, and because we are told to. The real question is: how do we regulate this inherent ambivalence?
*when you click on the pictures the website of the person opens
Daniel C. Dennett, Tufts University, US
(Austin B. Fletcher Professor of Philosophy)
Ned Block, NYU, US
(Silver Professor; Professor of Philosophy & Psychology)
David Chalmers, NYU, US
(Professor of Philosophy & Neural Science; co-director of the Center for Mind, Brain, and Consciousness)
Ophelia Deroy, LMU Munich, Germany
Paula Droege, Pennsylvania State University, US
Keith Frankish, University of Sheffield, UK
(Honorary Professor; Editor, Cambridge University Press)
Joshua Rust, Stetson University, US
(Professor of Philosophy)
Eric Schwitzgebel, UC Riverside, US
(Professor of Philosophy)
Anna Strasser, DenkWerkstatt Berlin, Germany
(Visiting Fellow UC Riverside, Founder of DenkWerkstatt Berlin; Associate researcher CVBE)