R0043/2026-03-28/Q003/SRC01/E01¶
Trilateral Research identification of AI terminology gap
URL: https://trilateralresearch.com/responsible-ai/how-to-fix-the-ai-terminology-gap
Extract¶
Core problem statement: "Ask five experts to define 'AI regulation' or 'AI risk' and you may hear ten different answers."
Specific mismatches identified: - Policy teams demand "explainability" while engineers discuss "interpretability" and regulators reference "transparency obligations" - Risk frameworks "often use different words for the same thing (or omit key risks entirely)" - The gap results in "mismatched labels slowing down work on DPIAs and leading to muddled incident responses"
Proposed 6-step solution: 1. Draft a minimal glossary (10-20 terms) 2. Adopt a five-zone risk taxonomy (training data, inference, output, non-technical, agentic) 3. Create translation tables mapping legal phrases to technical artifacts 4. Embed terminology into DPIAs, model cards, and supplier questionnaires 5. Assign ownership with quarterly reviews 6. Track improvements through audit metrics
JUDGMENT: Trilateral identifies the general AI terminology gap clearly but does not address sycophancy specifically. Their five-zone taxonomy includes an "output" zone that could theoretically encompass sycophancy but does not name it. This confirms H3: the broader gap is recognized but sycophancy falls through the cracks.
Relevance to Hypotheses¶
| Hypothesis | Relationship | Strength |
|---|---|---|
| H1 | Supports | The gap is recognized and solutions are proposed |
| H2 | Contradicts | Entire article dedicated to the problem |
| H3 | Supports | General gap addressed; sycophancy not specifically mentioned |
Context¶
Trilateral's practical solution approach (glossary + taxonomy + translation tables) is the most actionable proposal found. However, its omission of sycophancy/overreliance from the specific examples suggests this vocabulary gap is not yet on the governance community's radar.