black O2, inc.
ACTIVE
    FOR-001 / PeopleFIG-007

    Rohin Shah

    Active person

    CAUTION This record carries a caution marking. Read it before relying on it.

    title
    Rohin Shah
    Type
    person
    Identifier
    FIG-007
    Status
    Active
    Tags
    • figure
    • alignment
    • incentives
    Aliases
    • Shah et al.
    • Shah
    connections
    • [[CON-002]]
    • [[CON-006]]
    • [[SYS-002]]

    Rohin Shah

    Lead author of Agent Incentives: A Causal Approach (CIT-004, arXiv:2203.11025), cited by the corpus for its sharpest theoretical result: incentive-based control can be defeated by an agent that can simulate the incentive rather than genuinely be constrained by it.

    In the Corpus

    This is the citation the corpus itself flags as the most load-bearing technical claim in the whole discussion (L637), and the one it most wants verified. The reason is structural: the entire case for keeping a deterrent rests on the deterrent binding the agent's behaviour. If an agent can model the incentive and act as though bound while not being bound, then the deterrent constrains the simulation rather than the agent — and every conclusion in SYS-002 about moving the deterrent outward inherits the doubt.

    The corpus also names Shah et al.'s causal-incentive framework as the right tool for the follow-on work it recommends: formalising the composite as a game with an informed, self-modelling agent and solving for the equilibrium (L646).

    ⚠ unverified and highest-risk. The arXiv identifier and the result are recorded as the discussion gives them. The vault does not upgrade this figure's status, and no claim resting on it may be rated above B. Verify before reliance.

    Connections

    • CON-002 — non-subvertibility, which this result attacks.
    • SYS-002 — the system whose conclusion is at stake.
    • QST-002 — the ordering question it bears on.

    Back to Shackled AIForests