Glossary — AI

These are the jargon, acronyms, and constructs a non-expert hits in the AI whitepaper. Use this as a reading companion; the filing remains the source of truth.

• Decision density
  How often AI systems actually make consequential decisions — hiring, lending, housing, care — the quantity this filing says no federal instrument measures.
• Kill condition (KC)
  A pre-committed falsifier: if it fires, the inquiry's framing changes rather than getting papered over.
• Algorithm inventory / CAISI-style disclosure
  Rules that list which algorithms exist inside an organization, without publishing how often they decide or to whom.
• Consequential automated decision
  A decision with material effects on a person's rights, money, housing, or safety when software is in the loop.
• EEOC / CFPB / DOJ settlements
  The unglamorous enforcement trail (discrimination, fraud losses) the filing contrasts with population-scale claims that stay unfalsifiable.
• RAISE / state AI bills (e.g. SB)
  Named state and federal legislative vehicles discussed as architectures to score against measurement gaps.
• Voluntary commitments
  Industry pledges and white-house-style agreements that lack a measured decision series behind them.
• PJM / power & compute
  Grid and infrastructure constraints that appear when AI scaling is treated as an energy and siting problem, not only a model-safety problem.
• NCMEC / CSAM pipeline
  Child-exploitation reporting and detection constructs that show up when AI content harms are operationalized.
• TRAIGA / transparency regimes
  Disclosure-first regulatory designs that still don't create a national decision-frequency series.
• Unfalsifiable harm claim
  A claim that AI is (or isn't) causing widespread X when no population series exists that could prove either side wrong.
• Scorecard / architecture
  The desk's ranked set of policy designs scored against kill conditions and constraints.

Scope note: docs/glossary-scope.md. Challenge a definition: [email protected].