Researchers have identified significant issues with how large language models generate explanations for multimodal dementia prognosis. A study using the Alzheimer's Disease Neuroimaging Initiative found that models often attribute predictions to modalities that were not observed, a problem termed "phantom attribution." Even when prompted to acknowledge missing data, the models struggled to accurately advise on the impact of obtaining these missing tests, with accuracy rates as low as 27% until specific restoration techniques were applied. The research also highlighted that the scale of the language model, rather than prompting strategies, was the primary factor in the faithfulness of the generated explanations. AI
IMPACT Highlights critical limitations in AI's ability to provide reliable explanations for medical prognoses, impacting trust and adoption in clinical settings.
RANK_REASON Academic paper detailing a novel methodology and findings in AI explainability. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- Alzheimer's disease
- Alzheimer's Disease Neuroimaging Initiative
- CatalyzeX Code Finder for Papers
- CORE Recommender
- DagsHub
- Gotit.pub
- Hugging Face
- Influence Flower
- large language models
- mild cognitive impairment
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
