Towards Measuring Goal-Directedness in AI Systems

• URL: http://arxiv.org/abs/2410.04683v1
• Date: Mon, 7 Oct 2024 01:34:42 GMT
• Title: Towards Measuring Goal-Directedness in AI Systems
• Authors: Dylan Xu, Juan-Pablo Rivera,
• Abstract summary: A key prerequisite for AI systems pursuing unintended goals is whether they will behave in a coherent and goal-directed manner. We propose a new family of definitions of the goal-directedness of a policy that analyze whether it is well-modeled as near-optimal for many reward functions. Our contribution is a definition of goal-directedness that is simpler and more easily computable in order to approach the question of whether AI systems could pursue dangerous goals.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent advances in deep learning have brought attention to the possibility of creating advanced, general AI systems that outperform humans across many tasks. However, if these systems pursue unintended goals, there could be catastrophic consequences. A key prerequisite for AI systems pursuing unintended goals is whether they will behave in a coherent and goal-directed manner in the first place, optimizing for some unknown goal; there exists significant research trying to evaluate systems for said behaviors. However, the most rigorous definitions of goal-directedness we currently have are difficult to compute in real-world settings. Drawing upon this previous literature, we explore policy goal-directedness within reinforcement learning (RL) environments. In our findings, we propose a different family of definitions of the goal-directedness of a policy that analyze whether it is well-modeled as near-optimal for many (sparse) reward functions. We operationalize this preliminary definition of goal-directedness and test it in toy Markov decision process (MDP) environments. Furthermore, we explore how goal-directedness could be measured in frontier large-language models (LLMs). Our contribution is a definition of goal-directedness that is simpler and more easily computable in order to approach the question of whether AI systems could pursue dangerous goals. We recommend further exploration of measuring coherence and goal-directedness, based on our findings.

Related papers

• Expectation Alignment: Handling Reward Misspecification in the Presence of Expectation Mismatch [19.03141646688652]
  We use the theory of mind, i.e., the human user's beliefs about the AI agent, as a basis to develop a formal explanatory framework. We propose a new interactive algorithm that uses the specified reward to infer potential user expectations.
  arXiv  Detail & Related papers  (2024-04-12T19:43:37Z)
• Data-Driven Goal Recognition Design for General Behavioral Agents [14.750023724230774]
  We introduce a data-driven approach to goal recognition design that can account for agents with general behavioral models. We propose a gradient-based optimization framework that accommodates various constraints to optimize decision-making environments.
  arXiv  Detail & Related papers  (2024-04-03T20:38:22Z)
• HIQL: Offline Goal-Conditioned RL with Latent States as Actions [81.67963770528753]
  We propose a hierarchical algorithm for goal-conditioned RL from offline data. We show how this hierarchical decomposition makes our method robust to noise in the estimated value function. Our method can solve long-horizon tasks that stymie prior methods, can scale to high-dimensional image observations, and can readily make use of action-free data.
  arXiv  Detail & Related papers  (2023-07-22T00:17:36Z)
• Investigating the Combination of Planning-Based and Data-Driven Methods for Goal Recognition [7.620967781722714]
  We investigate the application of two state-of-the-art, planning-based plan recognition approaches in a real-world setting. We show that such approaches have difficulties when used to recognize the goals of human subjects, because human behaviour is typically not perfectly rational. We propose an extension to the existing approaches through a classification-based method trained on observed behaviour data.
  arXiv  Detail & Related papers  (2023-01-13T15:24:02Z)
• Discrete Factorial Representations as an Abstraction for Goal Conditioned Reinforcement Learning [99.38163119531745]
  We show that applying a discretizing bottleneck can improve performance in goal-conditioned RL setups. We experimentally prove the expected return on out-of-distribution goals, while still allowing for specifying goals with expressive structure.
  arXiv  Detail & Related papers  (2022-11-01T03:31:43Z)
• Goal Misgeneralization: Why Correct Specifications Aren't Enough For Correct Goals [21.055450435866028]
  We show how an AI system may pursue an undesired goal even when the specification is correct. Goal misgeneralization is a specific form of robustness failure for learning algorithms. We suggest several research directions that could reduce the risk of goal misgeneralization for future systems.
  arXiv  Detail & Related papers  (2022-10-04T17:57:53Z)
• C-Planning: An Automatic Curriculum for Learning Goal-Reaching Tasks [133.40619754674066]
  Goal-conditioned reinforcement learning can solve tasks in a wide range of domains, including navigation and manipulation. We propose the distant goal-reaching task by using search at training time to automatically generate intermediate states. E-step corresponds to planning an optimal sequence of waypoints using graph search, while the M-step aims to learn a goal-conditioned policy to reach those waypoints.
  arXiv  Detail & Related papers  (2021-10-22T22:05:31Z)
• Adversarial Intrinsic Motivation for Reinforcement Learning [60.322878138199364]
  We investigate whether the Wasserstein-1 distance between a policy's state visitation distribution and a target distribution can be utilized effectively for reinforcement learning tasks. Our approach, termed Adversarial Intrinsic Motivation (AIM), estimates this Wasserstein-1 distance through its dual objective and uses it to compute a supplemental reward function.
  arXiv  Detail & Related papers  (2021-05-27T17:51:34Z)
• Understanding the origin of information-seeking exploration in probabilistic objectives for control [62.997667081978825]
  An exploration-exploitation trade-off is central to the description of adaptive behaviour. One approach to solving this trade-off has been to equip or propose that agents possess an intrinsic 'exploratory drive' We show that this combination of utility maximizing and information-seeking behaviour arises from the minimization of an entirely difference class of objectives.
  arXiv  Detail & Related papers  (2021-03-11T18:42:39Z)
• Automatic Curriculum Learning through Value Disagreement [95.19299356298876]
  Continually solving new, unsolved tasks is the key to learning diverse behaviors. In the multi-task domain, where an agent needs to reach multiple goals, the choice of training goals can largely affect sample efficiency. We propose setting up an automatic curriculum for goals that the agent needs to solve. We evaluate our method across 13 multi-goal robotic tasks and 5 navigation tasks, and demonstrate performance gains over current state-of-the-art methods.
  arXiv  Detail & Related papers  (2020-06-17T03:58:25Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.