Anna Strasser

Standard notions in philosophy of mind have a tendency to characterize socio-cognitive abilities as if they were unique to sophisticated human beings. However, assuming that it is likely that we are soon going to share a large part of our social lives with various kinds of artificial agents, it is important to develop a conceptual framework providing notions that are able to account for various types of social agents. Recent minimal approaches to socio-cognitive abilities such as mindreading and commitment present a promising starting point from which one can expand the field of application not only to infants and non-human animals but also to artificial agents. Developing a minimal approach to the sociocognitive ability of acting jointly, I present a foundation for future discussions about the question of how our conception of sociality can be expanded to artificial agents.

Can large language models produce expert‐quality philosophical texts? To investigate this, we fine‐tuned GPT‐3 with the works of philosopher Daniel Dennett. To evaluate the model, we asked the real Dennett 10 philosophical questions and then posed the same questions to the language model, collecting four responses for each question without cherry‐picking. Experts on Dennett's work succeeded at distinguishing the Dennett‐generated and machine‐generated answers above chance but substantially short of our expectations. Philosophy blog readers performed similarly to the experts, while ordinary research participants were near chance distinguishing GPT‐3's responses from those of an “actual human philosopher”.

Although even very advanced artificial systems do not meet the demanding conditions which are required for humans to be a proper participant in a social interaction, we argue that not all human-machine interactions (HMIs) can appropriately be reduced to mere tool-use. By criticizing the far too demanding conditions of standard construals of intentional agency we suggest a minimal approach that ascribes minimal agency to some artificial systems resulting in the proposal of taking minimal joint actions as a case of a social HMI. Analyzing such HMIs, we utilize Dennett’s stance epistemology, and argue that taking either an intentional stance or design stance can be misleading for several reasons, and instead propose to introduce a new stance that is able to capture social HMIs—the AI-stance.

Can large language models be trained to produce philosophical texts that are difficult to distinguish from texts produced by human philosophers? To address this question, we fine-tuned OpenAI's GPT-3 with the works of philosopher Daniel C. Dennett as additional training data. To explore the Dennett model, we asked the real Dennett ten philosophical questions and then posed the same questions to the language model, collecting four responses for each question without cherry-picking. We recruited 425 participants to distinguish Dennett's answer from the four machine-generated answers. Experts on Dennett's work (N = 25) succeeded 51% of the time, above the chance rate of 20% but short of our hypothesized rate of 80% correct. For two of the ten questions, the language model produced at least one answer that experts selected more frequently than Dennett's own answer. Philosophy blog readers (N = 302) performed similarly to the experts, while ordinary research participants (N = 98) were near chance distinguishing GPT-3's responses from those of an "actual human philosopher".

Natural language processing based on large language models (LLMs) is a booming field of AI research. After neural networks have proven to outperform humans in games and practical domains based on pattern recognition, we might stand now at a road junction where artificial entities might eventually enter the realm of human communication. However, this comes with serious risks. Due to the inherent limitations regarding the reliability of neural networks, overreliance on LLMs can have disruptive consequences. Since it will be increasingly difficult to distinguish between human-written and machine-generated text, one is confronted with new ethical challenges. This begins with the no longer undoubtedly verifiable human authorship and continues with various types of fraud, such as a new form of plagiarism. This also concerns the violation of privacy rights, the possibility of circulating counterfeits of humans, and, last but not least, it makes a massive spread of misinformation possible.

Can we build machines with which we can have interesting conversations? Observing the new optimism of AI regarding deep learning and new language models, we set ourselves an ambitious goal: We want to find out how far we can get in creating a digital replica of a philosopher. This project has two aims; one more technical, investigating of how the best model can be built. The other one, more philosophical, explores the limits and risks which are accompanied by the creation of digital replicas. In cooperation with Daniel Dennett, we took his complete works to fine-tune a GPT-3. In this paper, we share our first results with some piloting models and discuss legal and ethical questions about creating and interacting with digital replicas.

Artificial agents have become increasingly prevalent in human social life. In light of the diversity of new human–machine interactions, we face renewed questions about the distribution of moral responsibility. Besides positions denying the mere possibility of attributing moral responsibility to artificial systems, recent approaches discuss the circumstances under which artificial agents may qualify as moral agents. This paper revisits the discussion of how responsibility might be distributed between artificial agents and human interaction partners (including producers of artificial agents) and raises the question of whether attributions of responsibility should remain entirely on the human side. While acknowledging a crucial difference between living human beings and artificial systems culminating in an asymmetric feature of human–machine interactions, this paper investigates the extent to which artificial agents may reasonably be attributed a share of moral responsibility. To elaborate on criteria that can justify a distribution of responsibility in certain human–machine interactions, the role of types of criteria (interaction-related criteria and criteria that can be deferred from socially constructed responsibility relationships) is examined. Thereby, the focus will lay on the evaluation of potential criteria referring to the fact that artificial agents surpass in some aspects the capacities of humans. This is contrasted with socially constructed responsibility relationships that do not take these criteria into account. In summary, situations are examined in which it seems plausible that moral responsibility can be distributed between artificial and human agents.

Research in social cognition aims to illuminate how agents can understand, communicate, and interact with other agents. When defining socio-cognitive abilities, standard cognitivist approaches tend to require demanding representational information processing. Thereby, they describe rather ideal cases. However, interdisciplinary research indicates multiple forms of how socio-cognitive abilities can be realized. Recent minimal approaches offer notions accommodating different kinds of cognitive processing. Nevertheless, the introduction of minimal cases of cognition raises new questions of how to account for commonalities and differences with respect to the standard concepts. It seems to be a widespread strategy to adapt ideas of a two-system approach in order to distinguish less demanding instances from more demanding cases. This paper critically explores such an interpretation of a two-system approach and argues that a dichotomous understanding fails to capture the actual diversity of cognitive processes.

Standard notions in philosophy of mind have a tendency to characterize socio-cognitive abilities as if they were unique to sophisticated human beings. However, assuming that it is likely that we are soon going to share a large part of our social lives with various kinds of artificial agents, it is important to develop a conceptual framework providing notions that are able to account for various types of social agents. Recent minimal approaches to socio-cognitive abilities such as mindreading and commitment present a promising starting point from which one can expand the field of application not only to infants and non-human animals but also to artificial agents. Developing a minimal approach to the socio-cognitive ability of acting jointly, I present a foundation for future discussions about the question of how our conception of sociality can be expanded to artificial agents.

This special issue of Grazer Philosophische Studien brings together a number of carefully selected and timely articles that explore the discussion of different facets of self-consciousness from multiple perspectives. The selected articles mainly focus on three topics of the current debate: (1) the relationship between conceptual and nonconceptual ways of self-representation; (2) the role of intersubjectivity for the development of self-consciousness; (3) the temporal structure of self-consciousness. A number of previously underexposed, yet important connections between different approaches are explored. The articles not only represent the state of the art in their respective areas of research and make new insights available, but also provide an overview of different methodologies: ranging from philosophy of language and mind to phenomenology and cognitive science. The volume is of interest for philosophers, cognitive scientists and researchers in related disciplines who are concerned with investigating the nature and origin of self-consciousness.

Human and non-human animals are social beings, both have social interactions. The ability to anticipate behavior of others is a fundamental requirement of social interactions. However, there are several ways of how agents can succeed in this. Two modes of anticipation, namely mindreading and behavior-reading, shape the animal mindreading debate. As a matter of fact, no position has yet convincingly ruled out the other. This paper suggests a strategy of how to argue for a mentalistic interpretation as opposed to a behavioral interpretation. The first step suggests considering a pluralistic approach in order to allow for shortcomings such as not having a natural language. Second, a critical examination of the principle called Morgan’s Canon will show that this principle cannot be used as a final argument to rule out mentalistic approaches. Finally, the author argues that the setting of current experiments is responsible for the indistinguishability of mindreading versus behavior-reading and she suggests alternative experimental designs.

In many cases, the ascription of self-consciousness is uncontroversial. For example, the ability to use the first person pronoun ‘I’ in the right way is obviously related to self-consciousness, although this is not true in all cases. The ascription of self-consciousness to infants, to persons with psychopathological syndromes, or to animals is controversial. In this paper, I will focus on the question of how ascribing self-consciousness to infants can be justified. There are two main subjects relevant to this debate. Firstly, the conceptual question: What notion of self-consciousness should be under discussion regarding infants? In other words: how minimal can a feasible notion of selfconsciousness be? Positions in this debate vary from presuming innate abilities such as the postulation of pre-reflexive self-consciousness (cp. Frank 1991) and the claim that language has to be developed and the right use of the personal pronoun ‘I’ has to be proven (cp. Castaneda 1966). Claiming that self-ascription abilities are an essential condition for all notions of self-consciousness, I begin by analyzing self-ascription abilities using the terminology of the sense of ownership, the sense of agency, and the sense of authorship. Equipped with this clarification, I suggest a minimal notion of self-consciousness that may serve as a foundation of a broader notion of self-consciousness. Secondly, the epistemological question: How can we know when ascribing self-consciousness is justified? This question is combined with an additional problem concerning the ascription of self-consciousness to infants. Infants show impressive behavior that seems to be tied to self-consciousness, but in some cases there are convincing alternative explanations for such behavior. Such alternatives do not refer to self-consciousness and still provide satisfying explanations. Finally, I will discuss several paradigmatic cases regarding the question of whether the ascription of self-consciousness is justified in such cases.

In dieser Arbeit wird die Frage nach der Handlungsfähigkeit künstlicher Systeme im Schnittfeld von Philosophie und KI behandelt. Eine positive Antwort auf Seiten der Philosophie hat deren anthropozentrischer Handlungsbegriff verhindert. Daher wird unterhalb des philosophischen Handlungsbegriffes der Begriff einer Quasi-Handlung entwickelt, welcher die Möglichkeit bietet, zwischen verschiedenen Kategorien des Verhaltens künstlicher Systeme zu unterscheiden. Als wesentliches Kriterium werden hierzu Unterschiede in der Flexibilität des Informationensverarbeitungsprozesses zur Differenzierung verschiedener Verhaltensklassen vorgeschlagen. Dies führt auch zu einer Auseinandersetzung mit dem Begriff der Kognition und den verschiedenen Arten des Lernens.

Eine immer wiederkehrende Frage der Analytischen Philosophie lautet: Wie kann man sich mit Sprache auf die Welt beziehen? Um eine Antwort auf diese Frage zu bekommen, muss man untersuchen, was alles eine Rolle spielt, wenn man Bedeutungen von Äußerungen versteht. Zu diesem Zwecke wird in dieser Arbeit der programmatische Vorschlag einer Bedeutungstheorie von Donald Davidson vorgestellt. Dazu ist es notwendig, sich mit der Wahrheitstheorie Tarskis zu beschäftigen. Interessant ist nun, welche Erkenntnisse unabhängig von Tarski in dieser Bedeutungstheorie enthalten sind. Hier kann man zum einen die These der Unerforschlichkeit der Referenz und zum anderen das Prinzip der Nachsicht nennen. Durch die Voraussetzung des Prinzips der Nachsicht, nämlich größtmögliche Übereinstimmung zwischen Interpret und Sprecher anzunehmen, ist es einem Interpreten möglich, in die Interdependenzbeziehung zwischen Überzeugung und Bedeutung einzudringen. So kann der radikale Interpret erste Interpretationshypothesen aufstellen, die er später mit einem bedeutungsholistischen Ansatz auf logische Konsistenz in bezug auf die faktische Menge aller seiner Interpretationshypothesen überprüfen kann.