Q001 — AI Training Limitations — ACH Matrix¶
Matrix¶
| Evidence | H1 — Adequate | H2 — Absent | H3 — Superficial |
|---|---|---|---|
| SRC01-E01 — NAVEX covers risks at high level | + | - | ++ |
| SRC02-E01 — GSA training exists with risk track | + | - | + |
| SRC03-E01 — Deloitte 30+ courses, Trustworthy AI | ++ | -- | + |
| SRC04-E01 — UK playbook explicit on limitations | ++ | -- | + |
| SRC05-E01 — Only 5/12 agencies acknowledge hallucinations | - | + | ++ |
| SRC06-E01 — 82% train, 59% gap | - | + | ++ |
| SRC07-E01 — Microsoft defines hallucinations, says "verify" | + | - | ++ |
| SRC08-E01 — Legal mandate, flexible implementation | + | - | ++ |
| SRC09-E01 — Single-sentence warning | + | N/A | ++ |
| SRC10-E01 — >50% find training inadequate | - | + | ++ |
| SRC11-E01 — NHS mandates oversight, flag outputs | + | - | + |
Legend¶
| Symbol | Meaning |
|---|---|
| ++ | Strongly consistent with hypothesis |
| + | Consistent with hypothesis |
| - | Inconsistent with hypothesis |
| -- | Strongly inconsistent with hypothesis |
| N/A | Not applicable |
Diagnosticity Analysis¶
Most diagnostic evidence:
- SRC06-E01 — The 82% availability vs. 59% skills gap finding discriminates strongly between H1 (adequate) and H3 (superficial). If training were adequate, the gap would not persist.
- SRC05-E01 — Only 5 of 12 agencies acknowledging hallucinations despite 9x AI growth discriminates between H1 and H2/H3.
Least diagnostic evidence:
- SRC08-E01 — The EU AI Act's flexibility means it is consistent with both H1 and H3 depending on implementation.
Outcome¶
H3 (superficial coverage) receives the most consistent support across all evidence items (9 of 11 items rated + or ++). H1 (adequate) is inconsistent with effectiveness data. H2 (absent) is inconsistent with the existence of training programs. H3 is the best-supported hypothesis.