Q001-H2 — Training Does Not Cover Limitations¶
Statement¶
Standard corporate AI training courses do not meaningfully teach employees about AI limitations, focusing instead on adoption, productivity, and use-case promotion.
Status¶
Partially supported. Pure omission of limitations is no longer the norm — most training programs include at least brief mentions of hallucinations and the need for verification. However, the treatment is often so superficial that it may not effectively convey the nature or significance of the limitations.
Supporting Evidence¶
| Evidence | Summary |
|---|---|
| SRC05-E01 | Only 5 of 12 federal agencies explicitly acknowledge hallucinations |
| SRC06-E01 | Training is generic, one-time, disconnected from practice; 59% skills gap persists |
| SRC10-E01 | >50% of workers find training inadequate |
Contradicting Evidence¶
| Evidence | Summary |
|---|---|
| SRC01-E01 | NAVEX course explicitly covers risks and limitations |
| SRC04-E01 | UK playbook has explicit limitation warnings |
| SRC07-E01 | Microsoft training addresses hallucinations |
| SRC08-E01 | EU AI Act now legally requires risk coverage |
| SRC09-E01 | Sample policies include hallucination warnings |
Reasoning¶
H2 in its strong form is eliminated by the evidence. Training programs do mention limitations. However, H2 captures an important truth about the functional outcome: the superficiality of coverage means that employees may not actually understand the limitations even though they were nominally "taught."
Relationship to Other Hypotheses¶
H2 represents the extreme negative position. Evidence shows it is too strong — limitations are mentioned. But the evidence supporting H2 (inadequacy, gaps, generic treatment) feeds directly into H3, the nuanced middle position.