Wikipedia, the Free Encyclopedia’s Post

Humans aren’t great at truth-seeking: Forming accurate beliefs grounded in reality requires using the scientific method. This involves starting with a default position of “I don’t know” and only reaching a belief after forming and testing hypotheses. Even after arriving at a belief, scientific thinking involves constant doubting and belief updating as new information arises. While humans are able to think scientifically, it’s difficult for us and doesn’t come naturally, incurring significant conscious effort - analogous to what it’s like to do mental arithmetic with large numbers. This has grounding in neuroscience, which finds that human brains are susceptible to a number of cognitive biases, including the most pernicious: confirmation bias. It can therefore be very difficult, even for highly intelligent people, to resist the pull of misleading content that tells us what we want to believe rather than what’s true.

It's out in Psychological Review! In this paper, we propose a method for testing whether theories explain empirical phenomena. Thanks Riet van Bork, Adam Finnemann, Jonas Haslbeck, Han L.J. van der Maas, Jill de Ron, jan sprenger, and Denny Borsboom! The paper is open-access and available here: https://t.co/JX0bpqOL9W 𝗔 𝗯𝗿𝗶𝗲𝗳 𝘀𝘂𝗺𝗺𝗮𝗿𝘆 𝗼𝗳 𝘁𝗵𝗲 𝗽𝗮𝗽𝗲𝗿: Our paper introduces a framework for evaluating explanations in psychological science. By representing theories as formal models and capturing phenomena as statistical patterns that are observed across studies, we aim to clarify how theories explain these phenomena. Productive explanation involves three steps: a) Explicate the verbal theory as a mathematical or computational model b) Represent the phenomenon as a statistical pattern, c) Simulate data from the model and test if it matches the phenomenon's pattern. We propose three criteria to evaluate explanations: Precision: How much of the formalization is determined by the theory Robustness: How well the statistical pattern is produced under parameter variations Empirical Relevance: The necessity of theory components in the model. We demonstrate our framework on the regulatory resource theory of ego depletion. This theory is much debated and replication studies failed to observe the effect of ego depletion under rigorous conditions. We draw some interesting conclusions which I'm not going to spoil here ;-)

7.2 The postmodern position B&K presuppose a set of ideological givens: Munsell, psychological reality, the innate colour space. Yet it makes as much sense to treat these givens as ideological, as a body of ruling ideas created by a ruling class. It makes as much sense to treat these givens as a form of misrepresentation containing distorted ideas about, say, colour science on the one hand, and intercultural relations on the other. Doing so would help correct the fundamental erasure of social and historical relations, practices and ideas, in B&K's work, which makes invisible historically particular impressions, noticings, reportings. The interested reader is directed to Saunders and Brakel (1997) which deals with the (supposedly) factual basis for the Berlin and Kay tradition — it includes commentary by a number of colour scientists and other interested parties — and to, for example, Brakel and Saunders (2002) which works to articulate a sociohistorical reconception of colour. 7.3 The communicative position This contemporary focus, which is here named the communicative position, might seem to build bridges between the universal constraints position and the postmodern position: focusing on the “everyday” cognitive tasks rather than “innate neural processes” is surely of importance to the postmodernist. Agreement between these positions is, however, unlikely. While compatible with the universal constraints position, in that it aims to give an explanation for the constraints in terms of communicative function, the research trend Lindsay and Brown identify is wholly within the orbit of contemporary cognitive science and psychology, and largely operates within the confines of the BCT theoretical construct. The cognitive tasks that people preform “every day” are abstracted away from contextualized human experience and relocated to the laboratory (e.g., Bae 2015) or the computational model (e.g., Belpaeme and Bleys 2009; Zasalavsky 2019). There is nothing wrong with these approaches—unless one thinks that the problem is with the application of a scientific worldview to lived experience. 7.4 The hybrid position Indeed, as cognitive science moves away from innatist models of the mind and toward inductive, probabilistic accounts, learning relative to one's environment (including one's social environment) will play a more significant role. https://lnkd.in/gAqw7EVW

In Anthropic's latest research, they attempt to unravel the mysteries within their LLM, Claude. This immediately reminded me of Phineas Gage. Let me explain. Anthropic peaked inside Claude to find concepts (they call them 𝑓𝑒𝑎𝑡𝑢𝑟𝑒𝑠) that exist inside it — everything from "The Golden Gate Bridge" to "secrecy" or "rude aggressive driving behaviors," and then 𝗰𝗹𝗮𝗺𝗽𝗲𝗱 some of these features up or down. For example, they clamped/dialed up the Golden Gate Bridge feature and Claude started responding as if it was the Golden Gate Bridge. You could 𝗰𝗹𝗮𝗺𝗽 the "scam emails" concept and it will write scam emails, which it normally shouldn't allow. To me this is really fascinating — it suggests that LLMs are a massive, complex amalgamation of 𝘢𝘭𝘭 𝘩𝘶𝘮𝘢𝘯𝘴’ 𝘮𝘪𝘯𝘥𝘴, and what makes me, well... 𝗺𝗲 might just be my particular clamping of different concepts and attributes in my brain. This brings me to the story of Phineas Gage (albeit likely exaggerated over the years)—the idea that Gage’s personality changed after his famous accident feels akin to his brain’s features being “clamped” by the incident. And I know *exactly* what clamping feels like: after a few overnight shifts and poor sleep, my emotion clamp is turned WAY up. I'm more likely to cry listening to a song, more likely to be irritable and snip at my husband, more likely to be slap-happy. And while phrenology is of course ridiculous pseudoscience, I loved the drawing below suggesting that all of these different values — "features" as Anthropic calls them — exist inside our brains. 👏 Applause to the Anthropic Research team — not only is this research fascinating, it's crucial for understanding this new technology and identifying its blindspots, gaps, and challenges. Daniela Amodei Jason Kim Rachel Lui Pang

A recent free access publication by Robert Lawrence Kuhn entitled “A landscape of consciousness: Toward a taxonomy of explanations and implications” is available to download at, https://lnkd.in/eCmJbcHy The reasons behind this monumental paper on consciousness, in Robert Lawrence Kuhn’s words, are: “In the Landscape paper, I seek an organizing framework for diverse theories of consciousness and to explore their impact on big questions. My central theses are twofold: (i) understanding consciousness at this point cannot be limited to selected ways of thinking or knowing, but should seek expansive yet rational diversity, and (ii) issues related to consciousness, such as AI consciousness, virtual immortality, meaning/purpose/value, free will, etc., cannot be understood except in the light of particular theories of consciousness. In addition, I want to present consciousness and its significance to broad scientific and scholarly communities.”

A unified theory of #consciousness (as an instance of #holism) needs to be able to account for itself, first, and only then to expand scope. This is the most sensible way to engage the inevitability of downstream self-referential antinomies, IMHO. Threading this particular needle backwards through itself is a subtle matter of conceptual engineering in #systems #philosophy. On the eve of AGI, if indeed we are, these deepest of questions make a stunning return. In reality, they never once left us and questions of #truth, #freedom, #political #economy, sustainable technological civilisation and environmental #complexity are deeply implicated. Why? Because holism is something that is not simply observed, defined or decomposably isolated in one artefact, entity or system. Solving holism in one place in some sense solves it everywhere. Personally, I think this is a significant and interesting personal challenge for 𝙖𝙡𝙡 of us.

I have been ruminating deeply on the cascading effects that the simple act of daily reading can instigate, transcending the mere acquisition of knowledge to become a cornerstone of personal evolution. As the cognitive faculties engage with the written word, a complex interplay ensues, fostering an iterative progression of mental fortification and enlightenment. Within this intricate tapestry, reading transcends mere information absorption to become a catalyst for critical analysis, synthesis, and refinement of thought… Through the immersive exploration of diverse perspectives and narratives, the reader is compelled to embark on a journey of introspection, challenging entrenched beliefs and broadening the horizons of comprehension. Furthermore, the cognitive demands inherent in sustained textual immersion stimulate neural networks, augmenting memory retention, analytical acuity, and problem-solving prowess. This intellectual crucible, perpetually stoked by an insatiable thirst for knowledge, propels the reader toward ever-greater echelons of intellectual enlightenment, enriching not only individual cognition but also the collective consciousness of society. Thus, in the relentless pursuit of enlightenment through daily literary engagement, one discovers not only a reservoir of wisdom but also an inexhaustible wellspring of intellectual vitality.

What might be a thought experiment? It is a form of emptying ones pockets and coming up with ideas based on what you have there. I was thinking about the nature of memory and the utter messiness of neuroscience. There used to be a connectionist model from the early days of Sherrington to Hebb, and then doctrines were cast aside as it was discovered more chemical and biological processes involved. From my perspective I was taken up with the notion of evolved memories and how one might simulate through abstraction the kind of memories that our pre-human ancestors had. How would go about creating the means and methods of working out what might be trace memories - like habits and functions we have inherited. Certainly one can see in performance and in physical development the expression of inherited information - but how far back does it go? Of course it is a thought experiment and as such has no scientific validity - nevertheless it is not on the same level as looking for lodestones:-)))) It is a whimsical thought that entertained me.

Notevenwrongisms "Not even wrong" is a famous quote attributed to the physicist Wolfgang Pauli who apparently dismissed in this way poorly constructed theoretical work that could neither be proved nor disproved. The quote has become the favourite attack against grandiose theories that try to explain very hard scientific problems in formal mathematical terms. Peter Woit attacked string theory in this way, writing a book with this title. I am not an expert, but I wonder if string theorists would not frame in the same way the theory of "Euclidean Twistor Unification" proposed by Woit to unify quantum field theory and gravity. The latest episode in "notevenwrongism" is the controversy sorrounding theories of consciousness which was reported yesterdat by Nature. A group of 124 scientistist wrote an open letter (https://lnkd.in/dmhkTWaW) attacking the integrated information theory of conciousness (which indeed is full of mathematical formulas https://lnkd.in/dajraZim). Again I am not an expert but I notice the similarity with the string theory controversy. Both are big complex problems (how does the universe work? what is consciousness?) and it would be so nice to have a closed mathematical theory to explain eveything. While I appreciate the efforts, the "theory of everything" might not even exist and we might to need to contnue work one step at a time trying to understand little bits of the world. https://lnkd.in/dZimZgtD

This is an interesting paper with implications for the philosophy of law. Where LLMs model language as complex systems with emergent properties, and legal informatics often seeks reductive approaches to understanding the semantic meaning of legally significant texts.