Ethologists and others studying animal behavior in a spirit are in need of a descriptive language and method that are neither anachronistically bound by behaviorist scruples nor prematurely committed to particular Just such an interim descriptive method can be found in intentional system theory. The use of intentional system theory is illustrated with the case of the apparently communicative behavior of vervet monkeys. A way of using the theory to generate data - including usable, testable data - is sketched. The (...) underlying assumptions of this approach can be seen to ally it directly with theorizing in evolutionary biology, which has recently come under attack from Stephen Gould and Richard Lewontin, who castigate it as the Their arguments, which are strongly analogous to B. F, Skinner's arguments against point to certain pitfalls that attend the careless exercise of such thinking (and rival varieties of thinking as well), but do not constitute a fundamental objection to either adaptationist theorizing or its cousin, intentional system theory. (shrink)

We consider the questions of whether or not large language models (LLMs) have beliefs, and, if they do, how we might measure them. First, we consider whether or not we should expect LLMs to have something like beliefs in the first place. We consider some recent arguments aiming to show that LLMs cannot have beliefs. We show that these arguments are misguided. We provide a more productive framing of questions surrounding the status of beliefs in LLMs, and highlight the empirical (...) nature of the problem. With this lesson in hand, we evaluate two existing approaches for measuring the beliefs of LLMs, one due to Azaria and Mitchell (The internal state of an llm knows when its lying, 2023) and the other to Burns et al. (Discovering latent knowledge in language models without supervision, 2022). Moving from the armchair to the desk chair, we provide empirical results that show that these methods fail to generalize in very basic ways. We then argue that, even if LLMs have beliefs, these methods are unlikely to be successful for conceptual reasons. Thus, there is still no lie-detector for LLMs. We conclude by suggesting some concrete paths for future work. (shrink)

The intentional stance is the strategy of prediction and explanation that attributes beliefs, desires, and other states to systems and predicts future behavior from what it would be rational for an agent to do, given those beliefs and desires. Any system whose performance can be thus predicted and explained is an intentional system, whatever its innards. The strategy of treating parts of the world as intentional systems is the foundation of but is also exploited in artificial intelligence and cognitive science (...) more generally, as well as in evolutionary theory. An analysis of the role of the intentional stance and its presuppositions supports a naturalistic theory of mental states and events, their content or intentionality, and the relation between levels of explanation and neurophysiological or mechanistic levels of explanation. As such, the analysis of the intentional stance grounds a theory of the mind and its relation to the body. (shrink)

This paper argues that questions can be bullshit. First it explores some shallowly interrogative ways in which that can happen. Then it shows how questions can also be bullshit in a way that’s more deeply interrogative.

The aim of the paper is to shed light on some of the most original elements of Fichte’s conception of morality as expressed in his account of specific obligations. After some remarks on Fichte’s original classification of ethical duties, the paper focuses on the prohibition of lying, the duty to communicate our true knowledge, and the duty to set a good example. Fichte’s account of those duties not only goes beyond the mere justification of universally acknowledged demands, but also deploys (...) different arguments than his contemporaries, most notably Kant. Fichte thereby sketches a conception of morality in which the agent is crucially required to contribute to the morality of others. The chapter explores the contrast between Fichte’s view and Kant’s thought of an end in itself and suggests that Fichte’s view of morality amounts to a form of normative perfectionism that is qualified by the underlying claim of the agent-neutral character of moral demands. (shrink)

Charitable interpreters of David Hume set aside his sprinkles of piety. Better to read him as lying than as clumsily inconsistent. We argue that the attribution of lies can pay dividends in historical scholarship no matter how strongly the theorist condemns lying. Accordingly, we show that our approach works even with one of the strongest condemners of lying: Immanuel Kant. We argue that Kant lied in his scholarly work and even in the first Critique. And we defend the claim that (...) this lie attribution, strange as it may sound, amounts to a kind of scholarly charity. (shrink)

Frankfurt gave an account of “bullshit” as a statement made without regard to truth or falsity. Austin argued that a large amount of language consists of speech acts aimed at goals other than truth or falsity. We don't want our account of bullshit to include all performatives. I develop a modification of Frankfurt's account that makes interesting and useful categorizations of various speech acts as bullshit or not and show that this account generalizes to many other kinds of act as (...) well. I show that this illuminates some of Graeber's classification of “bullshit jobs,” though it doesn't fully agree with it. (shrink)