Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
79 72 85 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
80 72 88 -
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 1 promise resolved, 5 needed for a score.
All types combined: 79 (likely 74 to 84, high confidence), from 102 claims: 64 factual claims, 38 predictions. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Gary Marcus's claims
42 claims
“It’s not an end to end system.”
Gary Marcus, podcast guest 14 November 2019 EP25 Gary Marcus on Rebooting AI (The Jim Rutt Show)
Cyc is not an end-to-end system.
Resolved true The record shows it happened or was so.
The second judge said mostly true; the verdict shown is the first judge's.
What would count as true or false
Judge as of 2019-11-14. An 'end-to-end system' means one trained as a single system mapping raw inputs directly to outputs, typically through learning. True: authoritative descriptions show that Cyc in 2019 was a hand-curated knowledge base with a separate inference engine and other components, and was not trained or operated end-to-end. Mostly true: Cyc was mostly modular and hand-built but had some learned or integrated pipeline elements. Mixed: the evidence is ambiguous about how integrated it was. Mostly false: Cyc worked largely as an integrated end-to-end pipeline, with modest modular parts. False: Cyc was an end-to-end system.
Evidence
- Selected Lexical/Ontology/Knowledge Base Projects METU OCW (course reading, prepared for COG 515 / COGS 523) · 13 October 2025
“It basically consists of a knowledge base and an associated inference engine, along with several other components”
- Guiding Inference with Policy Search Reinforcement Learning AAAI (FLAIRS 2007), by Taylor, Matuszek, Smith and Witbrock (Cycorp, Inc. co-authors) · 2007
“Inference in Cyc is sound but is not always complete: in practice, memory- or time-based cutoffs are typically used to limit very large or long-running queries.”
- Guiding Inference with Policy Search Reinforcement Learning AAAI (FLAIRS 2007), by Taylor, Matuszek, Smith and Witbrock (Cycorp, Inc. co-authors) · 2007
“The RL Tactician provides the inference Engine with an ordered list of tactics. However, the inference Engine may use”
- Cyc: Knowledge-Intensive AI (Chapter 4) - Artificial Dreams Cambridge University Press · 4 April 2008
“The first stage was envisaged as ending in 1994, the second by the late 1990s, and, as Doug Lenat, the main figure behind the project, put it, “Soon after, say by 2001, we planned to have it learning on its own, by automated discovery methods””
Reasoning
E1 describes Cyc as a knowledge base with an associated inference engine and several other components, which is a modular architecture. E4 records that Cyc's original plan was staged, with learning by automated discovery expected only at a later stage, around 2001. E3 shows an experimental 2007 reinforcement-learning component that supplies ranked tactics to the inference engine. It supplies tactics to a separate inference engine rather than making the system trained end-to-end, and no evidence shows it was in use in 2019. On this evidence Cyc was not an end-to-end system.
Cite this claim
Gary Marcus, “Cyc is not an end-to-end system.” (said 14 November 2019). Verdict: resolved true. Reliability API, claim c_b128e65af040, method 0.1.0+c876030c. https://epistacklabs.com/claims/c_b128e65af040
Claims without a verdict
Is a verdict here wrong, or do you speak for Gary Marcus? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.