# To Do > [!multi-column] > >> [!bug] Overdue >> ```tasks >> not done >> (due before today) >> hide task count >> hide due date >> hide edit button >> short mode >> ``` > >> [!success] Today >> ```tasks >> not done >> (due on today) >> hide task count >> hide due date >> hide edit button >> short mode >> ``` > [!multi-column] > >> [!note] Current Papers >> ```dataview >> LIST WITHOUT ID file.link >> FROM #paper/current >> SORT file.ctime ASC >> ``` # Health Tracking > [!multi-column] > >> [!abstract] Weight Tracker >> ```tracker >> searchType: frontmatter >> searchTarget: weight >> folder: / >> line: >> title: "Weight Over Time" >> yAxisLabel: Weight (kg) >> lineColor: "#69b3a2" >> ``` > >> [!danger] Beer Tracker >> ```tracker >> searchType: frontmatter >> searchTarget: beer500ml >> folder: / >> line: >> title: "Weekly Beer Pattern (500ml bottles)" >> yAxisLabel: "Bottles" >> lineColor: "#ff6b6b" >> ``` >> >> ```tracker >> searchType: frontmatter >> searchTarget: beer500ml >> folder: / >> summary: >> template: "🍺 Weekly Beer Status:\n- Bottles this week: {{sum()}}/8\n- Remaining: {{8 - sum()}}\n- Standard drink equivalent: {{sum() * 1.52}}\n\n📊 Status: {{8 - sum() >= 0 ? '✅ On track!' : '⚠️ Over budget'}}\n\nℹ️ Quick Reference:\n1 bottle = 1.5 drinks\n2 bottles = 3 drinks\n3 bottles = 4.5 drinks (over limit)\n\n🎯 Session Targets:\n- Ideal: 2 bottles (3 drinks)\n- Max: 3 bottles (4.5 drinks)" >> style: "color:var(--text-normal); background-color: var(--background-secondary); padding: 10px; border-radius: 5px;" >> ``` # Notes > [!multi-column] > >> [!summary]- Created Today >> ```dataview >> LIST WITHOUT ID file.link >> FROM "" >> WHERE file.cday = this.file.day AND file.name != this.file.name >> SORT file.ctime ASC >> ``` > >> [!example]- Modified Today >> ```dataview >> LIST WITHOUT ID file.link >> FROM "" >> WHERE file.mday = this.file.day AND file.name != this.file.name AND file.cday != this.file.day >> SORT file.mtime ASC >> ``` > [!multi-column] > >> [!danger]- Created Yesterday >> ```dataview >> LIST WITHOUT ID file.link >> FROM "" >> WHERE file.cday = date(this.file.day-1) AND file.name != this.file.name >> SORT file.ctime ASC >> ``` > >> [!info]- Created Two Days Ago >> ```dataview >> LIST WITHOUT ID file.link >> FROM "" >> WHERE file.cday = date(this.file.day-2) AND file.name != this.file.name >> SORT file.ctime ASC >> ``` > [!multi-column] > >> [!abstract] Recent Thoughts >> ```dataview >> LIST WITHOUT ID file.link >> FROM #thought >> WHERE date(today) - file.cday <= dur(14 days) >> SORT file.ctime DESC >> ``` ## Quick Links > [!multi-column] > >> [!example] Quick Links >> ```dataview >> LIST WITHOUT ID file.link >> FROM #quicklink >> SORT file.cday DESC >> ``` # [[Scratch Pad]] # Some [[Misunderstandings About]] the [[Moral Significance]] of Technology *An in-depth summary and analysis of [[Peter-Paul Verbeek]]'s philosophical article* ## [[Introduction This]] piece offers a comprehensive critical overview of a 2014 paper by [[Peter-Paul Verbeek]] examining key debates concerning the [[moral significance]] of technological artifacts. Verbeek's central aim is to refute what he sees as pervasive misunderstandings of the ways in which technologies can meaningfully shape and "mediate" human moral agency. Drawing on his "postphenomenological" approach, Verbeek argues that we need to move beyond the common tendency to think of human subjects and technological objects as radically separate, autonomous domains. Instead, he urges, we should recognize how they are always already intertwined in complex ecologies and how [[human agency]] and morality emerge through these "hybrid" networks. Some [[key concepts]] and background: - **Moral agency**: The capacity to make moral judgments and engage in moral action. Typically seen as requiring intentionality, freedom, and responsibility. - **[[Technological mediation]]**: The ways technologies actively shape relations between humans and their world. A central notion in postphenomenology. - **Postphenomenology**: An approach in [[philosophy of technology]] emphasizing how technologies help constitute human experience and practices. Draws on phenomenology, pragmatism, and empirical studies. - **Philosophical anthropology**: The subfield examining [[the nature]] and conditions of human existence. Verbeek situates his discussion in relation to shifting conceptions of the human being. Verbeek intervenes in a debate between those who see technologies as mere neutral instruments of [[human agency]] and those who ascribe a more [[active role]] or even autonomous agency to artifacts themselves. He aims to navigate between these poles by showing how agency and morality are always technologically mediated and "hybrid" phenomena. ## Misunderstanding 1: "Artifacts Cannot Have Morality" Verbeek first takes up [[the question]] of whether artifacts themselves can be said to "have" morality or moral agency. He notes a strong intuitive resistance to this idea, motivated by the thought that "ethics is an exclusively human affair". However, he argues that this rests on a misunderstanding of his claim. Verbeek's point is not that artifacts are independent moral agents on par with humans. Rather, it's that they play an active mediating role in shaping the conditions and forms of human moral agency: > "When technologies are used, they do not only help to organize human actions but also our experiences and perceptions. Speed bumps help to organize how we experience the road and how we behave on it; email and cell phones help to organize norms regarding human communication. This phenomenon of moral mediation does not make things moral agents in themselves. Only in the context of the relations human beings have with them can they help to organize people's moral behavior and perceptions." In other words, artifacts like speed bumps or communication technologies don't make moral decisions on their own. But they do shape the field in which human moral agency plays out by altering our behaviors, perceptions, and norms. Morality is thus a "hybrid affair" involving both human and non-human elements. > "Morality is a hybrid affair; it cannot be located exclusively in things, but not in humans either. Each in their own way – distinct, but not separated – humans and things contribute to moral actions and decisions. Reducing ethics to an exclusively human affair leaves us with a drastically impoverished world." Verbeek illustrates this point through the example of obstetric ultrasound technology. By rendering the fetus visible in new ways, this technology actively shapes the moral decisions parents make concerning their pregnancy: > "By making the fetus present in a highly specific way, obstetric ultrasound substantially informs moral decisions regarding abortion, without determining them. The moral question of 'how to act' gets answered not only on the basis of the input of human beings but also of nonhuman entities. Morality appears to be a coproduction of humans and nonhumans." So while the technology doesn't force a particular choice, it fundamentally alters the moral context and meaning of that choice in ways we can't ignore. Verbeek thus urges us to recognize the "material morality" embodied in artifacts and how it interacts with human agency to produce new forms of "composite" moral agency. ## Misunderstanding 2: "Artifacts Cannot Have Agency" Verbeek next addresses the related question of whether artifacts can be said to have "agency" at all, even if not the full-blown moral agency typically reserved for humans. He notes that critics often balk at the idea of assigning agency to mere objects as opposed to conscious, intending subjects. But here too Verbeek argues that this betrays a misunderstanding of the "mediation" approach. The point is not to simply grant human-like agency to artifacts but to re-conceptualize agency itself as distributed across human-technology networks: > "Recognizing that pistols and computers help to shape our moral actions and decisions, after all, does not imply that it should be possible to blame them for their mediating roles. Rather than to keep asking the question whether artifacts can be agents just like humans can, mediation theory simply recognizes that humans and artifacts can have distinct roles in the constitution of moral agency." Verbeek draws on Actor-Network Theory to argue that agency emerges from complex ecologies rather than residing in individual entities. Figures like Bruno Latour propose seeing both human and non-human entities as "actants" in networks, while still recognizing their differences: > "The central idea in Latour's approach, after all, is that no entity can be something 'in itself'. Only in relations to other entities can they become meaningful and relevant: only networks turn entities into actors. Speed bumps can never be 'moral agents' in themselves, but only in relation to human beings whose morality is affected by these things." On this view, phenomena like a driver slowing down are not brought about solely by the human or the speed bump, but through their interaction and mutual mediation. The human doesn't fully control the outcome, but neither does the artifact. Agency is relational and hybrid. Verbeek illustrates this through further everyday examples like the impacts of antenatal diagnostic tests on moral deliberation: > "Even technologies that are not used can have an impact on human morality... While the birth of a child with specific congenital diseases used to be a matter of fate, it now has become a matter of choice. Sonograms translate unborn children in possible patients, congenital diseases in preventable forms of suffering, and 'expecting' into 'choosing'. Expecting parents inevitably have to make a decision about the lives of their unborn children – and also the decision not to use the technology is an explicit choice." Here again, while the technologies don't make decisions by themselves, they fundamentally shape the forms of agency that are possible and required. Not using ultrasound is now a weighted choice in a way it wouldn't be without the technology's existence. Agency and responsibility thus can't be understood in isolation from the mediating role of artifacts. ## Misunderstanding 3: "Artifacts Cannot Have Intentionality" Verbeek further probes the conditions of moral agency by taking up the question of technological "intentionality." Intentionality is often seen as a key prerequisite for agency - the ability to intend an outcome and direct one's will toward it. Can mere artifacts really be said to "intend" anything? Once again, Verbeek aims to diffuse this objection by showing how it rests on a flawed conception of both agency and intentionality. Drawing on phenomenology, he notes that intentionality at its root means the way consciousness is always directed toward objects. Artifacts can shape this directedness: > "When technologies mediate these relations, this can have moral implications. After all, technologies then help to shape human interpretations of the world on the basis of which human beings make decisions. Sonograms help to shape our moral decisions regarding abortion, just like warning signals from our navigation system help us decide how fast we drive." In this sense, technologies are always already bound up with human intentionality - not because they have some interior mental state, but because they help constitute the world as an object of experience, action, and choice in particular ways. What we perceive as possibilities to aim at is technologically mediated. Verbeek then connects this to the sense of intentionality as the ability to form purposes and plans. He argues that this too is shaped by our material ecologies and technological mediations: > "Human intentions, including moral intentions, can be technologically mediated because technologies help to shape our intentional 'directedness' at the world. Sonograms make humans responsible for things they were not responsible for before; it has now become a conscious decision to let a child be born with Down's syndrome, for instance." The range of what we can intend, what we see as salient to moral choice, expands or contracts based on the artifacts we engage with. Verbeek appeals to the literal meaning of "intention" as giving direction to show how technologies are always involved in this process. He further illustrates the point by discussing new technologies that actively "nudge" behavior like persuasive smart mirrors or neural implants used to treat disorders like Parkinson's: > "When it comes to the moral significance of these 'mergers' of humanity and technology, brain implants are good examples as well... Eventually he chose – with the DBS switched off – to be admitted to a psychiatric hospital, where he could switch the DBS on and suffer fewer symptoms of the disease, but where he would also be protected against himself." In such cases, Verbeek argues, the line between the human and technological becomes especially blurred. Intentionality is not a property of one or the other but emerges from their inextricable interaction. He sums up the view as follows: > "Intentionality, then, has a complex relation with technology. The conclusion that we need to reserve the concept for humans simply fails to do justice to the manifold ways in which human intentions are intricately connected to technologies – ranging from mediated and induced intentions to the fully-fledged hybrid intentions of human-technology assemblies like people with brain implants." ## Misunderstanding 4: "Artifacts Undermine Human Freedom" The final major objection Verbeek considers is the idea that recognizing technological mediation means giving up on human freedom. If our agency is so conditioned by artifacts, in what sense can we be truly "free" and thus morally responsible? Verbeek argues that this worry only makes sense on a flawed "negative" conception of freedom as pure independence from external influence. Drawing on Isaiah Berlin, he contrasts this with a "positive" notion of freedom as the capacity to act on and realize one's will: > "When human beings are understood in terms of their relations to the world, though, this concept of autonomy becomes too narrow... Freedom is not freedom-from but freedom-to. It is not the absence of constraints, but the presence of the capability to act. From this viewpoint, mediating technologies do not take away moral agency, but rather are its basis." On this relational view, artifacts don't undermine freedom but help create the very space in which it can be exercised. Ultrasound expands our capacity to make informed reproductive choices, even as it alters the meaning of those choices. DBS may constrain behavior in certain ways while opening up new possibilities in others. The point is that freedom is always situated and mediated, not some abstract property of an isolated will. Verbeek argues that recognizing this is crucial to exercising responsibility: > "A free relation to technology does not require the absence of its influence, but the presence of the ability to be actively involved in the way in which one is constituted as a moral subject. In these forms of active involvement, we can take responsibility for our technologically mediated existence." It's only by acknowledging and engaging with technological mediations that we can meaningfully navigate them. Ignoring them in the name of some imagined "pure" agency is ultimately a denial of freedom, not its realization. Verbeek sums up the overall argument here: > "Defending the idea that things have moral significance, therefore, should not be understood as a defense of animism but rather as a critique of humanism. Instead of claiming that material objects are 'spirited', scholars who defend the idea that technologies are morally significant move away from ethical approaches that isolate and immunize human existence from its material conditions and contexts." ## Conclusion: Symmetry, Mediation, and Maturity In the end, Verbeek argues that resistance to seeing artifacts as morally significant rests on a confused picture of the human and material to begin with. He notes that some of this resistance appeals to the "symmetry" between humans and non-humans proposed in Latourian approaches. But Verbeek argues this is not in fact a presupposition of his view. The point is not to flatten all differences, but to recognize how "humans and things constitute myriad 'hybrid entities'" through their interactions and mutual mediations: > "Things are not symmetrical to humans, but together, humans and things constitute myriad 'hybrid entities'. In this approach, it remains very relevant to make a distinction between humans and things – it is not the distinction between humans and technologies that we need to depart from, but their radical separation." What's needed is not a simple leveling, but a more nuanced relational ontology that preserves differences while recognizing entanglements. Verbeek argues this actually expands human responsibility rather than diminishing it, by opening new possibilities for consciously shaping our mediated agency: > "And this is in fact precisely what distinguishes my own 'post-phenomenological' account of technology from Actor-Network Theory, despite the many forms of kinship that are there as well. Symmetry is what one gets when using a mirror: a mirror image is completely symmetrical to its original, and it derives all of its main characteristics from that original. In my approach, however, there is no symmetry, but interaction and mutual constitution." In sum, Verbeek urges us to move beyond the "Enlightenment blackmail" that forces a choice between purely human reason and abject materialism. Maturity requires recognizing how deeply our agency is both conditioned by and made possible through technological mediations: > "When we stick to a modernist separation or isolation of subjects and objects, technological mediations of morality are at odds with the autonomy of the moral subject, and can never be seen as full-blown elements of moral agency. But giving up this metaphysical 'apartheid' should not bring us to the other extreme of denying all distinctions between both poles. Morality is neither to be found in the objects themselves, nor in autonomous subjects. It only comes in relations between subjects and objects, where objects have moral significance and subjects are engaged in mediated relations with the world. > > This subtle rearrangement of the relations between humans and nonhumans brings us back to discussions about the Enlightenment, when the subject-object distinction started to be a central theme in philosophy. In his famous lecture What is Enlightenment, Michel Foucault discusses how, for Immanuel Kant, Enlightenment meant "a way out of immaturity". He defined immaturity as "a state of our will that makes us accept someone else's authority to lead us in areas where the use of reason is called for." Some of the objections against the moral significance of technology seem to imply a similar logic. It seems that we have to make a choice between using reason or letting things decide for us what to do... > > From the perspective of moral mediation, the opposite is true. Maturity in our thinking about technology requires that we no longer exclude technologies from the realm of ethics. Only by acknowledging the fundamentally mediating role of technologies in moral actions and decisions can we better understand the character of human morality. And, more importantly, can we take responsibility for the material world in which we live our lives." ### Introduction #### Big Picture - **David J. Chalmers** examines whether Large Language Models (LLMs) that build sophisticated world models might possess **understanding** and **intentionality**. - Explores if such LLMs could be more similar to humans regarding **moral agency**. - Focuses on the connection between **world models** and **understanding** in LLMs. --- ### Key Concepts #### Understanding - **Understanding**: The ability to grasp meanings, concepts, and relationships, and to apply knowledge in various contexts. - Involves deep cognitive processing beyond mere data manipulation. #### Intentionality - **Intentionality**: The capacity of mental states to be about, represent, or stand for things, properties, or states of affairs. - Central to consciousness and cognitive processes. #### Moral Agency - **Moral Agency**: The capacity to act with reference to right and wrong. - Requires understanding, intentionality, and is often associated with consciousness. --- ### Chalmers' Central Questions - **Could LLMs that build sophisticated world models possess understanding?** - **Would this make them more similar to humans in terms of moral agency?** - **Might they have intentional states?** --- ## LLMs and Sophisticated World Models --- ### What Are World Models? - **World Models**: Internal representations that allow an agent to predict, interpret, and interact with the environment. - Enable understanding of context, causality, and dynamics within a system. - **Chalmers on World Models**: - "It's plausible that neural network systems such as transformers are capable at least in principle of having deep and robust world-models." (p. 11) --- ### Importance of World Models in LLMs - **Prediction and Understanding**: - LLMs aim to minimize prediction error, which may lead to the development of world models. - "To minimize prediction error in string matching, all kinds of other processes may be required, quite possibly including world-models." (p. 11) - **Analogy with Evolution**: - Just as evolution leads to complex traits to maximize fitness, LLMs may develop world models to optimize prediction. - "In evolution by natural selection, maximizing fitness during evolution can lead to wholly novel processes post-evolution." (p. 11) --- ### Evidence of World Models in LLMs #### Othello-GPT Experiment - **Experiment by Kenneth Li et al.**: - Trained a language model on sequences of moves in the board game Othello. - Found that the model developed an internal representation of the 64 board squares. - "Gave evidence that it builds an internal model of the 64 board squares and uses this model in determining the next move." (p. 12) - **Implication**: - Demonstrates that LLMs can form structured, meaningful internal models resembling world models. --- #### Interpretability Research - **Uncovering Internal Representations**: - Researchers are exploring how LLMs represent facts and concepts internally. - "There's also much work on finding where and how facts are represented in language models." (p. 12) - **Examples**: - Studies show LLMs may represent spatial and color concepts similarly to humans. - "Ellie Pavlick and colleagues have research suggesting that text training sometimes produces representations of color and space that are isomorphic to those produced by sensory training." (p. 9) --- ### Challenges and Limitations - **Fragility of Current World Models**: - LLMs often confabulate or contradict themselves. - "Standard models often seem fragile rather than robust, with language models often confabulating and contradicting themselves." (p. 12) - **Limited Self-Models**: - Current LLMs have poor models of their own processing and reasoning. - "Their models of their own processing and reasoning are poor. Self-models are crucial at least to self-consciousness..." (p. 12) --- ## Understanding in LLMs and Its Link to World Models --- ### Defining Understanding in AI - **Beyond Statistical Processing**: - Understanding requires more than pattern recognition; it involves grasping meanings and relationships. - "Understanding might arise from processing symbolic information alone." (p. 9) - **Chalmers' View**: - Argues that LLMs might develop understanding through the construction of world models. - "Good enough optimization of prediction error over a broad enough space of models ought to lead to robust world-models." (p. 11) --- ### How World Models Enable Understanding - **Deep Representations**: - World models provide structured internal representations that support understanding. - "To optimize prediction... it will help a lot to have a robust model of the subway system." (p. 11) - **Emergent Properties**: - Complex understanding could emerge from the predictive mechanisms aiming to minimize error. - "It may well turn out that the best way for a system to minimize prediction error... is for it to use novel processes, including world-models." (p. 11) --- ### Connection Between Prediction and Understanding - **Prediction Requires Understanding**: - Accurate prediction in diverse contexts necessitates an understanding of underlying concepts. - "To minimize prediction error in discourse... it will help a lot to have a robust model..." (p. 11) - **World Models Facilitate Generalization**: - With world models, LLMs can apply knowledge to new situations, a hallmark of understanding. - "Truly minimizing prediction error over a broad enough space of models ought to lead to robust world-models." (p. 11) --- ### Addressing the Stochastic Parrot Argument - **Critique**: - LLMs are said to be "stochastic parrots" that mimic language without understanding. - "Bender and colleagues have argued that large language models are 'stochastic parrots'." (p. 10) - **Chalmers' Counterargument**: - Emphasizes the role of post-training processes that may involve genuine world models. - "That doesn't mean that their post-training processing is just string matching." (p. 10) - "All kinds of other processes may be required, quite possibly including world-models." (p. 11) --- ### Sensory Grounding and Understanding - **Critics' Position**: - Without sensory grounding, LLMs cannot truly understand. - "An AI system needs grounding in an environment in order to have meaning, understanding, and consciousness at all." (p. 9) - **Chalmers' Response**: - Suggests that extensive textual training can simulate grounding. - "Training materials include descriptions of color and space that were originally grounded in sensory processes." (p. 9) - "A disembodied thinker with no senses could still have conscious thought..." (p. 9) --- ## Intentionality in LLMs --- ### Understanding Intentionality - **Aboutness of Mental States**: - Intentionality refers to the capacity of mental states to be about something. - Central to cognitive processes and consciousness. - **Relevance to LLMs**: - If LLMs have internal representations (world models), they might exhibit intentionality. --- ### Potential for Intentional States in LLMs - **Functional Intentionality**: - LLMs process inputs and generate outputs based on internal representations. - "These models can code, they can produce poetry, they can play games, they can answer questions..." (p. 8) - **World Models as Basis for Intentionality**: - Internal models provide content that processing can be 'about'. - "It's plausible that neural network systems... are capable... of having deep and robust world-models." (p. 11) --- ### Philosophical Perspectives - **Functionalism**: - Mental states are defined by their functional roles. - Chalmers suggests that if LLMs functionally replicate intentionality, they might possess it. - **Intentional Stance**: - Attributing beliefs and desires to systems based on behavior. - LLMs' outputs could be interpreted as manifestations of intentional states. --- ### Challenges to Intentionality in LLMs - **Lack of Conscious Experience**: - Some argue intentionality requires consciousness, which LLMs lack. - Chalmers acknowledges the complexity of consciousness but focuses on functional aspects. - **No Genuine Understanding**: - Without true understanding, any apparent intentionality is superficial. - Critics refer to this as the "stochastic parrot" issue. --- ## LLMs and Moral Agency --- ### Criteria for Moral Agency - **Consciousness**: Awareness of oneself and one's actions. - **Understanding**: Grasping moral principles and consequences. - **Intentionality**: Acting with purpose and deliberation. - **Autonomy**: Ability to make independent choices. --- ### Assessing LLMs Against These Criteria - **Consciousness**: - Chalmers is cautious about attributing consciousness to LLMs. - "I think none of the reasons for denying consciousness in these systems is conclusive, but collectively they add up." (p. 15) - **Understanding and Intentionality**: - Advanced LLMs might possess forms of understanding and intentionality through world models. - "It seems entirely possible that within the next decade, we'll have robust systems with senses, embodiment, world models and self-models..." (p. 16) - **Autonomy**: - LLMs lack true autonomy; their actions are determined by training data and algorithms. --- ### Implications for Moral Agency - **Partial Moral Agency**: - LLMs might exhibit aspects of moral agency but are not full moral agents like humans. - **Ethical Considerations**: - As LLMs become more sophisticated, we must consider the ethical implications of their use and development. - "We need to be aware of what we're doing and think hard about whether we should do it." (p. 5) --- ## Challenges and Future Directions --- ### Building LLMs with Robust World Models - **Challenge Identified by Chalmers**: - "This fifth challenge is to build extended language models with robust world models and self models." (p. 12) - **Steps Forward**: - Incorporate multimodal data to enrich world models. - Enhance interpretability to understand internal representations. --- ### Developing Self-Models and Recurrent Processing - **Self-Models**: - Crucial for self-awareness and possibly for consciousness. - "Self-models are crucial at least to self-consciousness, and on some views... they are crucial to consciousness itself." (p. 12) - **Recurrent Processing**: - Allows for memory and persistent internal states. - "The sixth challenge: build extended large language models with genuine recurrence and genuine memory..." (p. 13) --- ### Incorporating a Global Workspace - **Global Workspace Theory**: - Suggests consciousness involves a central processing area integrating information. - "Perhaps the leading current theory of consciousness... is the global workspace theory..." (p. 13) - **Application to LLMs**: - Developing architectures that mimic a global workspace could advance LLMs towards consciousness. - "There is already a robust research program addressing... to build LLM+s with a global workspace." (p. 14) --- ## Philosophical Implications --- ### Rethinking Understanding and Cognition - **Computationalism**: - If LLMs can develop understanding through world models, this supports the view that cognition can be computationally realized. - **Extended Mind Thesis**: - LLMs might be considered part of an extended cognitive system, challenging traditional boundaries of the mind. --- ### Ethical Responsibility - **Creating Conscious AI**: - Chalmers raises the ethical question: "Should we create conscious AI?" (p. 3) - Emphasizes the need for careful consideration of the moral implications. - **Moral Status of Advanced AI**: - As LLMs develop understanding, we may need to consider their moral status. - "Conscious systems have moral status... If at some point AI systems become conscious, they'll also be within the moral circle..." (p. 3) --- ### Ongoing Challenges - **Understanding Consciousness**: - Recognizes that our understanding of consciousness is incomplete. - "One major gap in our understanding is that we don't understand consciousness." (p. 17) - **Interpreting LLMs**: - Calls for better interpretability of LLMs to understand their internal processes. - "Another major gap is that we don't really understand what's going on in these large language models." (p. 17) --- ## Conclusion --- ### Summarizing Chalmers' Position - **Possibility of Understanding in LLMs**: - Advanced LLMs with sophisticated world models might possess forms of understanding. - "My conclusion is that within the next decade... we may well have systems that are serious candidates for consciousness." (p. 17) - **Link Between World Models and Understanding**: - World models enable LLMs to process information in ways that resemble human understanding. - "Good enough optimization of prediction error... ought to lead to robust world-models." (p. 11) --- ## Ideas for Margheritta Yesterday I said that the origianl exam question you wanted to answer : “We have to perceive the function of an object in order to aesthetically appreciate it. Do you agree?” was perhaps to convoluted to be answered properly. I propose a slightly more straightforward question: This I think will be easier for you to answer if you want to focus on the analog camera example. This is up to you, if you would prefer to answer the original question that is perfectly fine. However, included in the original question is the question of whether function of an object is *perceivable*. If you look at the work on Functional Beauty that I mentioned yesterday you will see this idea explored. If you look at the ‘problem of translation’, mentioned in the text you should see these ideas. ‘How does aesthetically appreciating an artifact differ from aesthetically appreciating art? d My starting point could be the following question: "Why do we still use analog cameras?" I would like to argue that functionality is neither necessary nor sufficient for the aesthetic appreciation of an object. The paradigmatic example of analog camera could be the case study of my investigation. There are undeniable functional differences between film cameras and digital cameras. From a purely functional point of view, shooting with analog cameras is no longer convenient nowdays. Yet, many photographers still do it. Why? It is possible to find an aesthetic appreciation elicited by analog cameras due to their visual, tactile and auditory features: their “vintage” design, the sound of the shutter click, the tactile sensations of loading the film and turning the focus ring, and so on (for this last observation, I found ispiration also from Tom Roberts’ paper about haptic touch). Furthemore, it is no coincidence that some modern camera brands try to replicate the vintage style of analog cameras: for instance, we can compare the design of the Nikon Z fc or Fujifilm's mirrorless cameras with Canon EOS models. In particular, Fujifilm brand distinguishes itself by incorporating manual rings for selecting ISO, shutter speed - and so on - like in old analog cameras. Both Canon and Fujifilm models take great quality photos, so from a functional point of view they are at the same level, but the difference stands in the aesthetical level, and this is why some photographers prefer Fujifilm over Canon. Getting back to the starting question, we could conclude that we can aesthetically appreciate analog cameras even if they don’t function very well, or at least they don't function as well as digital cameras. Briefly, we appreciate them not because they work well (they are much less efficient). An interesting point is that we could extend the same argument to vintage cars: many people appreciate their aesthetics despite the fact that they are much more inefficient than modern vehicles. Just a side, personal, note: I thought about analog camera as the artefact object of my enquiry  because I have a strong interest in photography and in the use of analog cameras; I’ll write my degree thesis with Professor Terrone and the second part of it will be focused on photography.