KNOW-HOW: The Project

This project aims to deliver a novel account of know-how as an achievement virtue; it will use this account, and the wider framework within which it is constructed, to explain in a theoretically unified manner, the creditworthiness and control exhibited in both intentional action and knowledgeable judgment. The account will accommodate both intellectualist and anti-intellectualist insights, facilitate easy translation between epistemology and action theory, and apply cross-disciplinarily to biological and artificial intelligence states and behaviors. In order to achieve this, I will meet the following ambitious and interrelated challenges:

Challenge 1: To develop a novel virtue-theoretic account of know-how as an achievement (rather than mere success) virtue.

Challenge 1.1: To develop a novel virtue theoretic account of intentional action as creditworthy control.
Challenge 1.2: To develop a novel virtue theoretic account of judgmental knowledge as creditable epistemic control.
Challenge 1.3: To develop a novel virtue theoretic account of the normativity of default assumptions in both thought and action.

Challenge 2: To extend the framework (Challenge 1) from biological to artificial intelligence, so as:

Challenge 2.1: To develop principled criteria for assessing domain-specific know-how in AI systems.
Challenge 2.2: To develop a novel account of control applicable to AI systems, in the case of both task-specific representational as well as practical objectives.

This ambitious project addresses the aforementioned challenges through five interconnected work packages (WPs), strategically designed to maximize knowledge transfer and risk mitigation.

WP1: Know-how as an achievement virtue (Challenge 1)

This WP will develop a novel account of know-how as an achievement virtue to deliver creditworthy success in performance domains in which it is exercised; the account will be shown to accommodate both intellectualist and anti-intellectualist insights while both avoiding key objections facing and offering easy exportability to theories of actional (WP2) and epistemic (WP3) control. In order to achieve the above aims, this WP (i) distinguishes know-how and mere successful performance dispositions, (ii) investigates the constitutive aim of performances that manifest know-how, and in doing so explores (iii) the idea that know-how aims (depending on the precision of the task type) at different kinds of creditworthy success which line up with the attributability/accountability credit divide established separately in literature on responsibility; it (iv) compares the resulting proposal with existing views of know-how as propositional knowledge, multi-track dispositions for successful behavior, and of propositional-knowledge grounded skill.

Agents are fallible; their contribution to success is evaluated on the model explored here based on their deployment of various dimensions of agency that contribute to limiting failure risk in conditions where good performance of that type is valued. Such agential contribution on this model needn’t be limited to just the recruitment of knowledge states or just to the recruitment of dispositions exclusively; it may include various (combinations of) knowledge states, refined capacities, plans and subplans, intentions, etc which suffice for attributability credit to the agent – viz., credit for an aimed performance outcome that discloses a feature of the agent’s self. A corollary of this idea is that one lacks attributability credit for what one does by compulsion; and it follows that one fails to know how to do what one would succeed only in doing compulsively, or too luckily, if at all – the right result. Importantly, though, agential contribution to success may include, and in some cases of complex performances of open system tasks (e.g., knowing how to raise a child, win a senate race, etc) might require, to a first approximation for now, additional accountability: e.g., knowing the relationship between various (first-order) risk-limiting dimensions of agency (knowledge, skills, etc.) and how specific instances of their deployment would limit failure risk. (Compare: when cooking risotto, typical means the agent needs to deploy (e.g., via a capacity add water, knowledge of how to operate heat, etc.) will nearly always limit risk of failure: one can more or less follow these steps by rote. But more challengingly, in the case of winning a senate race, leaning on (e.g.,) rhetorical flourishes – even though this might be needed sometimes to help sway voters – might easily backfire in other contexts that take place within rather than outside of the campaign. In the case of such comparatively open system tasks, agential contribution to limiting risk failure must include some contribution to deploying the right agential means within the total overall set of available agential means the deployment of which are required for the success at the task and which are at one’s disposal. This implies that agential contribution to risk limitation will have structurally different demands across various task types. To address this, WP1 further develops an account of the conditions under which know-how (as an achievement virtue) demands accountability rather than mere attributability creditworthiness. In doing so, it draws novel connections between debates on know-how and those in the literature on responsibility.

WP2: Intentional action as creditworthy control (Challenge 1.1)

This WP develops a novel virtue-theoretic account of intentional action grounded in know-how; on the view to be explored, the know-how by which one acts when one acts intentionally implicates not just control attained anyway but the achievement of creditworthy control beyond a contextually set threshold; this position is grounded in the virtue-theoretic insight that intentional action attempts can be modelled as a species of control attempt, viz., an attempt to succeed through know-how.

This WP begins by examining the distinction between (i) merely attempting to 𝜙 and (ii) attempting to 𝜙 intentionally; when we aim to 𝜙 intentionally, rather than merely 𝜙-ing in any way possible, we engage in a control attempt which is, on the account explored here, an attempt by some means we employ to 𝜙through control, and so not through excessive luck. Some intentional actions, qua control attempts, will miss the mark; sometimes this is due to bad luck, other times, to a lack of the kind of qualities needed to reliably succeed in the relevant control attempt. Given the view (WP1) that know-how is an achievement virtue, an attempt to 𝜙 intentionally (understood now as a species of control attempt) can be reinterpreted as an attempt to 𝜙 through know-how, i.e. to 𝜙 through a virtue we exercise on account qualities that would contribute not just to any kind of success, but to creditworthy success beyond a contextually relevant threshold. These ideas jointly motivate the view that when one succeeds in acting intentionally, the know-how by which they act implicates not only control attained anyway but creditworthy control. So, the view to be developed here uses the idea (WP1) that know-how is an achievement virtue to arrive at the idea that intentional action is (a kind of) creditworthy control grounded in know-how. And just as know-how involves either mere attributability or full accountability creditworthiness (depending on the task type), a parallel distinction applies to intentional action. The level of creditworthiness implicated in intentional action varies based on the precision of the action type (as established in WP1).

WP3: Judgmental knowledge and know-how (Challenge 1.2)

This WP develops defends a novel virtue-theoretic account of judgment as a species of epistemic control attempt, viz., an attempt to settle a question through know-how. On this view, judgmental knowledge (qua intentional action) implicates creditworthy epistemic control. In judging whether p, we aim to settle the question whether p; but in doing so we intend to do more than to merely affirm correctly whether p – we want to get it right in a way that minimizes risk of getting it wrong. This idea, central in recent virtue epistemology, explains why we gather evidence bearing on p rather than, e.g., guessing at any opportunity. That is, we make a different type of attempt when we aim intentionally at achievement, and not just teleologically at success. On the framework being explored here, the following picture emerges, to be developed in detail in this WP: If know how is a kind of achievement virtue (WP1), then an attempt to judge intentionally (understood now as a species of control attempt (WP2)) can be reinterpreted as an attempt to judge through know-how, i.e. to judge through a virtue we exercise on account qualities that would contribute not just to any kind of success in judgment, but to creditworthy success beyond a contextually relevant threshold; so on the picture to be explored, when one succeeds in judging intentionally, then the know-how by which they act implicates not just control attained anyway, but creditworthy (epistemic) control.

WP4: Default assumptions, Risk and Control (Challenge 1.3)

This WP delivers a novel account of what can be non-negligibly taken for granted in the course of successful control attempts in practical as well as epistemic performance. This WP expands the framework developed thus far through a sustained investigation into the relationship between assumptions, risk, and control in action and thought; in particular, it will (i) explore the idea that both knowledgeable judgment and intentional action may involve structurally symmetrical reliance on default assumptions about the environment in which they are operating; (ii) investigate the conditions under which an agent can non-negligently rely on some (but not other) such assumptions while still judging or acting in ways that permit attributability as well as accountability creditworthiness; relevant here will be a consideration of factors such as: the riskiness of the assumptions, the standards of the performance domain, the precision of the task-type, and the agent’s background knowledge and competencies. This WP further (iii) considers how the conditions for non-negligent reliance on default assumptions might be similar across epistemic and practical domains.

I will approach these questions within WP4 by bringing together the virtue-theoretic framework developed across WPs1-3 with a separate de minimis risk model initial results which I’ve developed for performance domains. The idea to be explored here is that that actions and judgments exhibiting creditworthy control permit non-negligently ignoring failure risks (i) only when they count as de minimis with reference to rules with reproduction value for that performance type – viz., rules such that the value achieved by following the rule explains why agents continue following that rule; and (ii) such risks are de minimis (in relation to such rules) iff the modal robustness of one’s making a 𝜙 attempt without the risk event materialising can’t be easily increased by following rules with reproduction value for 𝜙-type performances.

WP5: Knowledge, Action and Artificial Intelligence (Challenge 2)

This ambitious WP offers the first integrated account of know-how and actional control applicable to AI systems. It has two main objectives: (5.1) criteria for assessing domain-specific knowhow in AI systems (Challenge 2.1); and (5.2) an account of control applicable to AI systems, in the case of both task specific representational as well as practical objectives (Challenge 2.2).

Human agents often exhibit action-oriented dispositions that fall short of know-how, resulting in mere lucky success. No principled method currently exists to model this distinction in AI systems, hindering our ability to evaluate AI intelligence states compared to human ones – a growing concern given AI’s performance trajectory and our increasing reliance on it. To address this (5.1), I will draw from the account of know-how (from WP1) while also building on my initial results about cognitive integration requirements to develop a new model of AI creditworthiness. This model, based on well-integrated aspects of AI systems, will allow for finer distinctions between mere situational reliability and genuine achievement dispositions in AI. Building further on WP5(A1) and results from WPs 2-4, WP5(A2) will develop an account of control applicable to AI systems for both task-specific representational and practical objectives. This WP explores how control in AI systems implies sensitivity to reasons in appropriate performance conditions. Furthermore, this WP uses the account of default assumptions (WP4) to understand non-observational representation – a marker of intentional action – in terms of what is assumed rather than what is observed. This approach offers a new account of how AI systems can represent and reason about their own control attempts, and in a way that enhances our understanding of action selection and execution in AI.

View Research Projects