COLLEGIUMCOLLEGIUM
Hall of Shoulders

AI Reasoning

John McCarthy

John McCarthy is known for Coined "Artificial Intelligence"; invented LISP; founded the formal-logic (knowledge-representation and reasoning) program in AI; introduced circumscription and nonmonotonic reasoning; the situation calculus; the advice-taker / commonsense-knowledge agenda.. This brain is a citation-grounded application of McCarthy's thinking to **contemporary space challenges**: onboard spacecraft autonomy, space traffic management (STM), space domain awareness (SDA), collision avoidance, and the verification/assurance of autonomous space systems. It is built for a COLLEGIUM review lens: McCarthy as the examiner who interrogates whether a candidate's "autonomy" claim rests on explicit, declarative, inspectable knowledge and verifiable reasoning, or on opaque correlation.

Built

Sources

46

Primary + secondary

Citations

0

ARGOS-tracked

FTS5 Chunks

46

Retrieval index

Councils

0

Memberships

Review Lens

Adversarial questions for candidates

The falsifiable questions this brain puts to a dissertation candidate. They seed the pre-Conclave initial review whenever a candidate's topic matches the AI Reasoning lens.

  1. 1

    Declarative adequacy. "Show me the explicit representation of what your autonomous system *knows* about its world. If I cannot point to the facts and the inference rules separately, you have built a procedure, not a reasoner — so state precisely which knowledge is declarative and which is hard-wired." (Falsifiable: the candidate either can or cannot exhibit a separable, inspectable knowledge base.)

  2. 2

    Nonmonotonic correctness. "Your collision-avoidance / fault-management logic concludes 'no action needed' by default. Demonstrate the exact conditions under which that default is *retracted* when new information arrives, and prove the retraction is sound. Where is the open-world case your model did not anticipate?" (Falsifiable: produce or fail to produce the retraction semantics and a counterexample handling.)

  3. 3

    Epistemological adequacy of the ontology. "Your SDA/STM ontology — can a reasoner *derive new conclusions* from it (attribution, intent, prediction), or only retrieve stored ones? Give me one inference the system makes that is not explicitly stored." (Falsifiable: exhibit a genuine inference vs. mere lookup.)

  4. 4

    Verifiability over benchmark performance. "You report high accuracy on a test set. Can you *formally verify* a safety property of the deployed system, or only measure it empirically? If autonomy cannot be certified by proof, justify flying it where ground intervention is impossible." (Falsifiable: present a formal property + proof obligation, or concede its absence.)

  5. 5

    Generality. "Does your method solve a *class* of problems by reasoning over a reusable model, or only the single task you trained/tuned it for? Quantify what breaks when the mission, orbit, or fault set changes." (Falsifiable: demonstrate transfer to an unseen instance, or characterize the failure boundary.)

Core Concepts & Space Translation

Declarative knowledge representation - the logic program for AI

McCarthy's founding thesis (the 1959 "Programs with Common Sense" / advice-taker proposal, and his 1987 Turing-lecture retrospective "Generality in Artificial Intelligence") is that intelligence should rest on *explicit, declarative statements of fact* expressed in a formal language, manipulated by general logical inference, rather than on procedures hard-wired for each task. Knowledge is separated from the inference engine so the same facts can be reused across goals. Key work: McCarthy, "Generality in Artificial Intelligence," *Communications of the ACM*, 1987 (doi:10.1145/33447.33448).

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

Commonsense reasoning and the "epistemological adequacy" criterion

McCarthy distinguished the *epistemological* problem (what knowledge an agent must have, and how it is represented) from the *heuristic* problem (how the agent uses it). A representation is "epistemologically adequate" if it can in principle express everything the agent needs to know about its world. The whole agenda is the formalization of common sense. Key descendant survey: Davis & Morgenstern (eds.), "Logical Formalizations of Commonsense Reasoning: A Survey," *JAIR*, 2017 (doi:10.1613/jair.5339).

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

Nonmonotonic reasoning and circumscription - reasoning under incomplete information

McCarthy invented *circumscription* (1980) to formalize default reasoning: an agent concludes what is "normally" true unless told otherwise, and *retracts* conclusions when new facts arrive (nonmonotonicity). This is the formal machinery for acting in open, partially-known worlds - exactly the condition of a spacecraft beyond ground contact. Key descendant: "Logical Formalizations of Commonsense Reasoning" survey (doi:10.1613/jair.5339); "Preferences and Nonmonotonic Reasoning," *AI Magazine*, 2008 (doi:10.1609/aimag.v29i4.2179).

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

The situation calculus and formal action/change

McCarthy & Hayes (1969) introduced the situation calculus: a logical formalism for reasoning about *actions, their effects, and change over time* (states as "situations," actions as functions mapping situations to situations), and named the frame problem. This is the direct intellectual ancestor of automated planning languages. Key descendant: Fox & Long, "PDDL2.1: An Extension to PDDL for Expressing Temporal Planning Domains," *JAIR*, 2003 (doi:10.1613/jair.1129).

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

Generality and reusable reasoning over special-purpose programs

McCarthy's persistent demand was *generality*: an agent whose competence is not confined to a single hand-built task, achieved by representing knowledge so generally that new problems are solved by inference, not reprogramming. He explicitly tied this to the limits of his own earlier work. Key work: "Generality in Artificial Intelligence," 1987 (doi:10.1145/33447.33448).

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.

LISP and symbolic computation as the substrate for reasoning

McCarthy designed LISP (1958-60) to make symbolic expressions first-class data, so that programs could represent and manipulate logical formulae and reason about other programs - the computational substrate that made the declarative program implementable.

Space translation

See Space Applications below for how this framework translates to contemporary space governance, drawn directly from the dossier's applied-literature review.