# [[meeting with enrico]] #paper/generatingphilosophy #paper/environmentalaestheticsofai #aestheticsofengineeringpaper ## Communication and File Management An email sent recently by one participant was not found by the recipient, who did not recall the subject line. The sender was asked to resend it or clarify the method of sending. A discussion occurred regarding access to videos that require a login. One individual possesses a signed contract related to these videos and intends to share a link after determining the most effective method for doing so. ## Transcription and Video Processing A procedure for obtaining transcripts from video content was outlined. This involves recording the video as it plays and subsequently uploading the audio recording to an AI (referred to as "chargpd," likely a reference to ChatGPT) for transcription. An alternative method mentioned is the use of certain meeting recording applications that can generate both an audio file and a transcript. This transcript can then be supplied to an AI for tasks such as summarisation. This work is scheduled for completion by the end of the current month, potentially within the next week, before a trip to "parties" (interpreted as Paris). ## Paper: Engineering Aesthetics and Rube Goldberg Machines A paper on engineering aesthetics is due the following month. Progress has been limited due to a focus on [[other projects]], including an "environmental" paper. A central idea for the engineering paper involves Rube Goldberg machines, inspired by a previous blog post about "gulpit machines" (understood as a mishearing of Goldberg machines). These contraptions, such as the breakfast-making machine in the film _Back to the Future_, are considered examples of engineered aesthetic satisfaction. An abstract for this paper currently exists, but further development is required. Time is to be allocated, specifically an afternoon, to further conceptualise this paper. ## Paper: [[Environmental Aesthetics]] and [[Latent Space]] Information from a conference in Venice was deemed relevant to [[the environmental aesthetics]] paper. Computer scientists at this conference discussed methods for characterising [[latent space]], referring to it as a "generative archive" or a "predictive algorithm." This characterisation of [[latent space]] is considered applicable to the environmental paper. ## Paper: [[Generating Philosophy with Artificial Intelligence]] This paper was the primary focus of the discussion, aiming to demonstrate that Large [[Language Models]] (LLMs) can perform substantive philosophical work. ### Core Aim and Conceptual Framework The central aim is to argue that LLMs are capable of engaging in significant philosophical work, primarily by simulating or performing reasoning. The discussion explored two interconnected conceptions of philosophy: 1. **Philosophy as Understanding**: This perspective draws heavily on Dellsén et al.'s account, where [[philosophical understanding]] is achieved by constructing a network of [[dependency relations]] around a particular phenomenon. [[The goal]] is to map these dependencies accurately and comprehensively. # [[meeting with enrico]] #paper/generatingphilosophy #paper/environmentalaestheticsofai #aestheticsofengineeringpaper ## Communication and File Management An email sent recently by one participant was not found by the recipient, who did not recall the subject line. The sender was asked to resend it or clarify the method of sending. A discussion occurred regarding access to videos that require a login. One individual possesses a signed contract related to these videos and intends to share a link after determining the most effective method for doing so. ## Transcription and Video Processing A procedure for obtaining transcripts from video content was outlined. This involves recording the video as it plays and subsequently uploading the audio recording to an AI (referred to as "chargpd," likely a reference to ChatGPT) for transcription. An alternative method mentioned is the use of certain meeting recording applications that can generate both an audio file and a transcript. This transcript can then be supplied to an AI for tasks such as summarisation. This work is scheduled for completion by the end of the current month, potentially within the next week, before a trip to "parties" (interpreted as Paris). ## Paper: Engineering Aesthetics and Rube Goldberg Machines A paper on engineering aesthetics is due the following month. Progress has been limited due to a focus on [[other projects]], including an "environmental" paper. A central idea for the engineering paper involves Rube Goldberg machines, inspired by a previous blog post about "gulpit machines" (understood as a mishearing of Goldberg machines). These contraptions, such as the breakfast-making machine in the film _Back to the Future_, are considered examples of engineered aesthetic satisfaction. An abstract for this paper currently exists, but further development is required. Time is to be allocated, specifically an afternoon, to further conceptualise this paper. ## Paper: [[Environmental Aesthetics]] and Latent Space Information from a conference in Venice was deemed relevant to the environmental aesthetics paper. Computer scientists at this conference discussed methods for characterising latent space, referring to it as a "generative archive" or a "predictive algorithm." This characterisation of latent space is considered applicable to the environmental paper. ## Paper: Generating Philosophy with Artificial Intelligence This paper was the primary focus of the discussion, aiming to demonstrate that Large Language Models (LLMs) can perform substantive philosophical work. ### Core Aim and Conceptual Framework The central aim is to argue that LLMs are capable of engaging in significant philosophical work, primarily by simulating or performing reasoning. The discussion explored two interconnected conceptions of philosophy: 1. **Philosophy as Understanding**: This perspective draws heavily on Dellsén et al.'s account, where philosophical understanding is achieved by constructing a network of dependency relations around a particular phenomenon. The goal is to map these dependencies accurately and comprehensively. # meeting with enrico #paper/generatingphilosophy #paper/environmentalaestheticsofai #aestheticsofengineeringpaper ## Communication and File Management An email sent recently by one participant was not found by the recipient, who did not recall the subject line. The sender was asked to resend it or clarify the method of sending. A discussion occurred regarding access to videos that require a login. One individual possesses a signed contract related to these videos and intends to share a link after determining the most effective method for doing so. ## Transcription and Video Processing A procedure for obtaining transcripts from video content was outlined. This involves recording the video as it plays and subsequently uploading the audio recording to an AI (referred to as "chargpd," likely a reference to ChatGPT) for transcription. An alternative method mentioned is the use of certain meeting recording applications that can generate both an audio file and a transcript. This transcript can then be supplied to an AI for tasks such as summarisation. This work is scheduled for completion by the end of the current month, potentially within the next week, before a trip to "parties" (interpreted as Paris). ## Paper: Engineering Aesthetics and Rube Goldberg Machines A paper on engineering aesthetics is due the following month. Progress has been limited due to a focus on other projects, including an "environmental" paper. A central idea for the engineering paper involves Rube Goldberg machines, inspired by a previous blog post about "gulpit machines" (understood as a mishearing of Goldberg machines). These contraptions, such as the breakfast-making machine in the film _Back to the Future_, are considered examples of engineered aesthetic satisfaction. An abstract for this paper currently exists, but further development is required. Time is to be allocated, specifically an afternoon, to further conceptualise this paper. ## Paper: Environmental Aesthetics and Latent Space Information from a conference in Venice was deemed relevant to the environmental aesthetics paper. Computer scientists at this conference discussed methods for characterising latent space, referring to it as a "generative archive" or a "predictive algorithm." This characterisation of latent space is considered applicable to the environmental paper. ## Paper: Generating Philosophy with Artificial Intelligence This paper was the primary focus of the discussion, aiming to demonstrate that Large Language Models (LLMs) can perform substantive philosophical work. ### Core Aim and Conceptual Framework The central aim is to argue that LLMs are capable of engaging in significant philosophical work, primarily by simulating or performing reasoning. The discussion explored two interconnected conceptions of philosophy: 1. **Philosophy as Understanding**: This perspective draws heavily on Dellsén et al.'s account, where philosophical understanding is achieved by constructing a network of dependency relations around a particular phenomenon. The goal is to map these dependencies accurately and comprehensively. 2. **Philosophy as Argument**: Within analytic philosophy, progress is often characterised by the construction of sound and comprehensive arguments. Arguments are seen as the primary means by which philosophical understanding is achieved and refined. The connection to LLMs lies in the observation that their underlying structures (e.g., vector spaces, trajectories in latent space, capabilities demonstrated through chain-of-thought prompting) can be mapped onto both the dependency networks of understanding and the inferential chains of arguments. ### Concerns and Evolution of the Paper's Focus A concern was raised that a recent abstract (approximately 1000 words) had shifted the paper's emphasis. The initial concept strongly linked the structure of LLM latent space to the network dependency relations central to Dellsén's model of understanding. The more recent version appeared to focus more heavily on LLMs' capacity for argument. While this makes the claim that LLMs can do philosophy more straightforward (philosophy involves arguments; LLMs can make arguments; therefore, LLMs can do philosophy), it was perceived as "less exciting" and "more obvious" than the original formulation, which offered a "more peculiar conception of philosophy." A key task identified is to clarify the precise relationship between understanding and argument. Arguments are posited as the tools or means by which dependency relations are established, tested, and refined. They can serve to: - Support the truth or accuracy of specific dependency claims. - Demonstrate connections between different dependencies, thereby enhancing the comprehensiveness of the overall network of understanding. ### Proposed Paper Structure A four-section structure was agreed upon for the paper: 1. **Conception of Philosophy**: This section will present the endorsed view of philosophy, integrating the Dellsén model of understanding through dependency networks with the role of argumentation. It will elaborate on arguments as the primary philosophical means to achieve understanding, focusing on maximising accuracy and comprehensiveness. This section will also distinguish philosophical methodology from that of science (which relies on experiments) and mathematics (which relies on formal proofs). 2. **Technical Profile of LLMs**: This part will describe the relevant workings of LLMs, including concepts such as latent space, vector space semantics, and prompt crafting/conditioning. The emphasis will be on those features that demonstrate a potential for understanding and argumentation. 3. **Synthesis: LLMs Can Do Philosophy**: This section will connect the first two. It will argue that, by virtue of their technical features (detailed in Section 2), LLMs are capable of performing the kinds of understanding-oriented and argument-based tasks that characterise philosophy (as defined in Section 1). The logical structure is: Premise 1: Philosophy is a matter of understanding and argument. Premise 2: LLMs exhibit potential for understanding and argument. Conclusion: LLMs can do significant philosophical work/contribute to philosophy. 4. **Objections, Replies, and Final Considerations**: This concluding section will anticipate and address potential objections to the paper's central thesis. It will also offer reflections on the broader implications of LLMs performing philosophical tasks. ### Detailed Discussion Points and Analogies - **Philosophy vs. Mathematics**: A distinction was drawn between philosophical arguments and mathematical proofs. Mathematical LLMs, trained on vast datasets of formal proofs, offer a point of comparison. However, a key difference is that mathematical premises (axioms) are typically taken as unquestionable starting points, with the focus being on valid inference. In contrast, a significant aspect of philosophical practice involves the selection, defence, and questioning of premises. Bertrand Russell’s early questioning of mathematical axioms was cited as an example of a philosophical, rather than purely mathematical, stance. The "genius of philosophy" was suggested to lie more in identifying and defining appropriate premises than in the inferential steps themselves, which are often less complex. The dependency-relations account of understanding is not seen as an attempt to mathematise philosophy with fixed axioms, but rather as a framework for assessing the quality (accuracy, comprehensiveness) of a philosophical theory. - **Rules in Philosophy**: The discussion explored whether philosophy operates under strict rules comparable to those in games like chess or Go, or in formal mathematics. It was concluded that while philosophy is not an "anything goes" activity, its guiding principles are more akin to Gricean maxims of rational communication (e.g., clarity, coherence, truthfulness) than to rigid, procedural rules. The process involves understanding phenomena, gathering information, and engaging critically with existing work. Strategies like abduction (inference to the best explanation) are employed, but these are considered general principles of good reasoning, not rules unique to philosophy. The paper should endorse a descriptive view of philosophy, reflecting actual practice, rather than a strictly normative one. - **Analogies with Games (Chess/Go) and Self-Play**: The training of AI for games like chess and Go was discussed. These games have fixed, explicit rules, which allows for effective training through self-play, where the AI can generate vast numbers of games to learn from. Philosophy lacks such rigidly defined rules, particularly concerning premise generation, which is more open-ended. This makes direct application of self-play methodologies more challenging. The discussion considered pre-trained versus untrained LLMs for philosophy, drawing an analogy to pre-trained and untrained Go-playing AIs. The success of self-play in Go is attributed to its clear rule set, which allows the AI to generate valid game instances and evaluate outcomes. In philosophy, the closest equivalents to such rules might be general principles of rationality or Gricean maxims. ### Anticipated Objections for Section 4 Several potential objections to the claim that LLMs can do philosophy were identified: - **Embodied Intuition**: A common objection is that philosophy requires first-personal, embodied intuitions (e.g., in evaluating thought experiments), which LLMs, as non-embodied systems, lack. A possible counter is that the relevant intuitions are already embedded within the vast textual data on which LLMs are trained. - **History of Making / Genesis of Ideas**: It might be argued that the process of creation and the historical context of ideas (the "history of making") are integral to philosophy in a way that is not captured by LLM-generated text. The counter-argument suggests this view might align philosophy more with art (where authorial intent and context are often central) rather than with science (where the validity of a theory is generally assessed independently of its discoverer's personal history or the process of discovery). The concern is that this could reduce philosophy to a "hobby" or "self-expression." - **Role of Human Philosophers**: If LLMs become proficient at generating philosophical work, the role of human philosophers might shift. Potential future roles include: - Readers and interpreters of LLM-generated philosophy, using it to enhance their own understanding. - Curators or commentators on LLM outputs. - Designers of sophisticated prompts to guide LLMs. This role, however, might be transitional if LLMs develop to a point where they can address complex philosophical questions from very general prompts (e.g., "Tell me what justice is," akin to the "Deep Thought" computer in _The Hitchhiker's Guide to the Galaxy_). A visualised trend is that human effort in generating philosophical content might decrease as LLM sophistication increases. ### Prompt Craft as Artefact Design and Hylomorphism A novel idea discussed was the nature of "prompt crafting." It was proposed that writing an effective prompt for an LLM is less like engaging in a conversation and more like designing an artefact. - The Aristotelian concept of **hylomorphism** (form and matter) was tentatively applied. The **form** of the prompt is its specific sequence of words and linguistic structures. The **matter** is the underlying token-based representation within the LLM's latent space. - An analogy was drawn: just as a hammer and a nail operate within the same physical world and are subject to the same physical laws (sharing "material properties"), a prompt (the "hammer") and the LLM's latent space (the "nail" or the "world of words") operate within a shared "world of tokens." The prompt's effectiveness comes from how its form interacts with this shared token-based "matter." - The philosopher's role, in this context, could be conceptualised as "sculpting" prompts – designing them with a deep understanding of how their structure will interact with the LLM's internal representations to produce desired philosophical outputs. This requires moving beyond treating the LLM as a human interlocutor and instead understanding its systemic properties. - This concept of prompt craft was seen as potentially bridging the work on the philosophy of AI paper with the environmental aesthetics paper. An analogy to gardening was used: a successful gardener needs knowledge (either scientific or practical/tacit) of the "nature" of their plants, soil, and environment to cultivate them effectively. Similarly, effective prompt crafters need to understand the "nature" of LLMs. ### Meta-Philosophical References The discussion referenced several philosophers and their works in the context of metaphilosophy and understanding LLMs: - **Dellsén et al.**: Their paper on philosophy as understanding through dependency networks is a cornerstone. - **Buckner**: His work on LLMs for philosophers was found useful for its clear explanations, although one piece was criticised for excessive historical exposition. - **Williamson**: His writings on metaphilosophy were considered for supporting the paper's framework. Specific works mentioned included _Philosophy of Philosophy_ (and a potential view of philosophy as modal logic or a branch of mathematics, which was deemed too extreme for the current paper's purposes), the more accessible _Doing Philosophy: From Common Curiosity to Logical Reasoning_ (2018), and a recent publication on "overfitting and heuristics in philosophy." The aim is to find elements in Williamson's work that reinforce the Dellsénian approach. ### Next Steps for the Paper - A full draft outline of the paper, incorporating the agreed-upon four-section structure (with placeholders for sections still under development), is to be circulated by one of the participants on Monday, May 19th. - This draft will be discussed at an in-person meeting scheduled for Tuesday, May 20th, at 12:00 PM. - The possibility of splitting the project into two separate papers (one focusing on latent-space dependency relations, the other on LLM reasoning and argumentation) was briefly noted as an option, though the current plan is for a single, comprehensive paper. ## Reading Materials and Other Discussions - A YouTube video discussing reasoning in language models, previously shared via WhatsApp, was mentioned. The message containing the link is on an old phone and needs to be retrieved. - Reading lists, including materials on "Natural Language Reasoning" and a "Survey of Chain of Thought Prompting," are being consulted. - A brief discussion touched upon the idea that dedicated phone applications might become obsolete, replaced by AI integrated at the operating system level. Users would interact by prompting the OS for desired information or functionalities, making prompt writing a primary mode of interaction. This was linked to the evolution of smartphones and computers. - The observation that Google search results now sometimes feature AI-generated summaries was noted as a move in this direction. ## Administrative and Scheduling Matters - An individual is currently abroad and plans to complete transcription work in the upcoming week, prior to a trip to Paris. - A meeting is scheduled for Tuesday, May 20th, at 12:00 PM to discuss the "Generating Philosophy with AI" paper. A draft outline is to be provided by Monday, May 19th. An alternative meeting time of Tuesday morning was also considered feasible. - Time (an afternoon) is to be dedicated to the engineering aesthetics paper, which has a deadline of next month. - A new review invitation from the journal _Philosophies_ has been received and needs evaluation. ## Conference Updates ### Venice Conference (Recent) Discussions with computer scientists at a recent Venice conference regarding the characterisation of latent space (as "generative archives" or "predictive algorithms") were noted as relevant for the environmental aesthetics paper. ### Athens Conference (Upcoming) The program for an upcoming conference in Athens has been released. - A significant proportion of the speakers (estimated at 80%) are Italian. - Notable attendees mentioned include Julia F. R. Lorenzi (working on music) and Elisa (whose paper with Camilla on introspection is considered to have high potential). Michele Averchi will also be present. - Emanuele Arielli, an organiser of the "Understanding Value" conference, will be in attendance. One of the speakers had previously reviewed Arielli's book, finding it "not good" but expressing this critique gently. Averchi's talk in Napoli on automation was recalled positively. - The conference format, featuring numerous parallel sessions with short (20-minute) talks, was criticised as not conducive to fostering a sense of community or justifying the travel involved for such brief presentation slots. One participant noted their presentation has an advantageous slot. ## Journal Submission Status: _Ergo_ A paper submitted to the journal _Ergo_ is currently under review. - The journal's submission system unexpectedly allows authors to view referee reports before an official editorial decision is communicated. - Two referee reports have been received and are visible: - Referee "G" provided a quick and positive report, recommending publication with minor suggestions. - Referee "F" also appeared generally positive but raised concerns regarding the bibliography and the paper's engagement with the concept of "aesthetic experience," referencing "somaesthetics" (attributed to Richard Shusterman and described as a "continental" approach focusing on "aesthetics with the body," related to everyday aesthetics). - A third referee, "H," has been assigned to review the paper, with a report due by June 2nd. - The decision to send the paper to a third referee is interpreted as a potentially positive sign from the editor. - _Ergo_ is perceived as an improving, open-access journal with good management practices, possibly superior to _Philosophers' Imprint_ in that regard, though the latter might hold more prestige due to its association with NYU.