Self-Evaluation can Help Agents Meet Social Expectations
Bibliographic record
Abstract
As artificial intelligence (AI) and multi-agent systems (MAS) become increasingly advanced and integrated into real-world applications, they are frequently used for content generation, problem-solving, and interactive communication. However, these systems still lack reflective capabilities, the ability to self-evaluate and reason about their own decisions. In particular, large language models (LLMs) often exhibit surface-level reflection that is more a product of linguistic prediction than reasoning. In this paper, we implement a case study on a recently proposed architecture designed to equip AI systems with reflective and self-evaluation capabilities. We implement and evaluate this architecture using a widely accessible LLM, Llama3, as the test bed and compare its baseline performance with the same model enhanced by this reflective architecture. To test the proposed system's ability to navigate social norms, we designed a normsensitive scenario involving a surprise birthday party. The model was prompted with 30 realistic questions that the guest of honor might ask, and its responses were evaluated across four metrics. The self-evaluator module is implemented using a second LLM to assess whether the base model's response aligns with defined norms and expectations. If not, the prompt is revised and reevaluated in an iterative loop. Experimental results show that this reflective setup improves the model's compliance with social and common-sense expectations, without requiring additional training or complex prompt engineering.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.007 | 0.032 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.004 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".