# Deep Reinforcement Learning Agents are not even close to Human Intelligence

arXiv: https://arxiv.org/abs/2505.21731 · HTML: https://mlsanity.com/paper/2505.21731/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in Deep RL algorithms and representations, one of 15 directions in Deep reinforcement learning methods.
- First posted on arXiv 27 May 2025.

## Placed in

- Deep reinforcement learning methods / [Deep RL algorithms and representations](https://mlsanity.com/deeprl/deep-rl-algorithms-and-representations/)

## Built on by (most cited first, 8 of 8)

- [SLR: Automated Synthesis for Scalable Logical Reasoning](https://mlsanity.com/paper/2506.15787/) (2025)
- [ActivationReasoning: Logical Reasoning in Latent Activation Spaces](https://mlsanity.com/paper/2510.18184/) (2025)
- [AbideGym: Turning Static RL Worlds into Adaptive Challenges](https://mlsanity.com/paper/2509.21234/) (2025)
- [Human-Allied Relational Reinforcement Learning](https://mlsanity.com/paper/2510.16188/) (2025)
- [Boosting deep Reinforcement Learning using pretraining with Logical Options](https://mlsanity.com/paper/2603.06565/) (2026)
- [Gradient Iterated Temporal-Difference Learning](https://mlsanity.com/paper/2603.07833/) (2026)
- [GRAIL: Autonomous Concept Grounding for Neuro-Symbolic Reinforcement Learning](https://mlsanity.com/paper/2604.16871/) (2026)
- [Bridging Physical Reasoning and Task Generalization via Visual Action Outcome Reasoning Alignment](https://mlsanity.com/paper/2607.06522/) (2026)

## Abstract (arXiv)

Deep reinforcement learning (RL) agents achieve impressive results in a wide variety of tasks, but they lack zero-shot adaptation capabilities. While most robustness evaluations focus on tasks complexifications, for which human also struggle to maintain performances, no evaluation has been performed on tasks simplifications. To tackle this issue, we introduce HackAtari, a set of task variations of the Arcade Learning Environments. We use it to demonstrate that, contrary to humans, RL agents systematically exhibit huge performance drops on simpler versions of their training tasks, uncovering agents' consistent reliance on shortcuts. Our analysis across multiple algorithms and architectures highlights the persistent gap between RL agents and human behavioral intelligence, underscoring the need for new benchmarks and methodologies that enforce systematic generalization testing beyond static evaluation protocols. Training and testing in the same environment is not enough to obtain agents equipped with human-like intelligence.