# Ultra Strong Machine Learning: Teaching Humans Active Learning Strategies via Automated AI Explanations

**Authors:** Lun Ai, Johannes Langer, Ute Schmid, Stephen Muggleton  
**arXiv:** [2509.00961](https://arxiv.org/abs/2509.00961)  
**Date:** 2025-08-31  
**Summary page:** https://www.summarizepaper.com/en/paper/2509.00961v1/

## Key points

- LENS is a novel approach introduced to enhance machine and human performance
- LENS combines machine learning with natural language processing to automatically explain logic programs
- LENS operates through three key pipelines: learning, explanation, and scoring
- In the learning pipeline, LENS creates a library of learned programs for reference using machine learning techniques
- The explanation pipeline interprets individual programs using multiple coding models and generates consensus explanations inspired by improved performance in factual reasoning tasks
- The scoring pipeline objectively evaluates explanations against human-crafted reference answers using language models to mitigate biases
- The goal of this work is to explain machine-learned logic programs effectively and enhance human understanding
- While Explainable AI research focuses on making systems intelligible through explanations, this study aims at evaluating human comprehension by comparing predictive performance
- Systematic evaluation has shown that LENS excels in generating superior explanations but may overwhelm users with comprehensive responses for simpler problems
- A human learning experiment did not show significant improvements when utilizing transferable active learning strategies generated by LENS across related domains
- Nevertheless, this work establishes a foundation for developing effective tools supporting human learning processes
- The source code for LENS is available on GitHub for further exploration and development opportunities

## Summary

In the realm of , a novel approach known as  has been introduced to enhance both machine and human performance. LENS combines  to automatically explain machine-learned logic programs in natural language. It operates through three key pipelines: learning, explanation, and scoring. In the learning pipeline, LENS uses  to create a library of learned programs for reference. The explanation pipeline employs multiple coding LLMs to interpret individual programs and generate a consensus explanation using a reasoning LLM. This process is inspired by strategies that have shown improved performance in factual reasoning tasks. In the scoring pipeline, LLMs are used to objectively evaluate explanations against human-crafted reference answers based on specific prompts, mitigating potential biases. The goal of this work is to effectively explain machine-learned logic programs and enhance human understanding. While Explainable AI research aims to make learning systems more intelligible through explanations, this study focuses on objectively evaluating human comprehension by comparing predictive performance. Through systematic evaluation using both LLM judges and human validation, it has been demonstrated that  in generating superior explanations. Despite its promising results in automating the explanation of logic programs, a human learning experiment conducted across three related domains did not show significant improvements in human performance when utilizing transferable active learning strategies generated by LENS. This suggests that comprehensive responses from LLMs may overwhelm users for simpler problems, rather than providing effective learning support. Nevertheless, this work lays a solid foundation for developing effective  that can support human learning processes. The source code for LENS is available on GitHub for further exploration and development opportunities.