More Like this

1. The ConceptARC Benchmark: Evaluating Understanding and Generalization in the ARC Domain

   Moskvichev, Arsenii Kirillovich; Odouard, Victor Vikram; Mitchell, Melanie (August 2023, Transactions on machine learning research)

   The abilities to form and abstract concepts is key to human intelligence, but such abilities remain lacking in state-of-the-art AI systems. There has been substantial research on conceptual abstraction in AI, particularly using idealized domains such as Raven's Progressive Matrices and Bongard problems, but even when AI systems succeed on such problems, the systems are rarely evaluated in depth to see if they have actually grasped the concepts they are meant to capture. In this paper we describe an in-depth evaluation benchmark for the Abstraction and Reasoning Corpus (ARC), a collection of few-shot abstraction and analogy problems developed by Chollet [2019]. In particular, we describe ConceptARC, a new, publicly available benchmark in the ARC domain that systematically assesses abstraction and generalization abilities on a number of basic spatial and semantic concepts. ConceptARC differs from the original ARC dataset in that it is specifically organized around "concept groups" -- sets of problems that focus on specific concepts and that are vary in complexity and level of abstraction. We report results on testing humans on this benchmark as well as three machine solvers: the top two programs from a 2021 ARC competition and OpenAI's GPT-4. Our results show that humans substantially outperform the machine solvers on this benchmark, showing abilities to abstract and generalize concepts that are not yet captured by AI systems. We believe that this benchmark will spur improvements in the development of AI systems for conceptual abstraction and in the effective evaluation of such systems. 

   more » « less
2. Abstraction Alignment: Comparing Model-Learned and Human-Encoded Conceptual Relationships

   https://doi.org/10.1145/3706598.3713406  

   Boggust, Angie; Bang, Hyemin; Strobelt, Hendrik; Satyanarayan, Arvind (April 2025, ACM)

   While interpretability methods identify a model’s learned concepts, they overlook the relationships between concepts that make up its abstractions and inform its ability to generalize to new data. To assess whether models’ have learned human-aligned abstractions, we introduce abstraction alignment, a methodology to compare model behavior against formal human knowledge. Abstraction alignment externalizes domain-specific human knowledge as an abstraction graph, a set of pertinent concepts spanning levels of abstraction. Using the abstraction graph as a ground truth, abstraction alignment measures the alignment of a model’s behavior by determining how much of its uncertainty is accounted for by the human abstractions. By aggregating abstraction alignment across entire datasets, users can test alignment hypotheses, such as which human concepts the model has learned and where misalignments recur. In evaluations with experts, abstraction alignment differentiates seemingly similar errors, improves the verbosity of existing model-quality metrics, and uncovers improvements to current human abstractions. 

   more » « less
3. Communicating Natural Programs to Humans and Machines

   Acquaviva, Sam; Pu, Yewen; Kryven, Marta; Wong, Catherine; Ecanow, Gabrielle E; Nye, Max; Sechopoulos, Theodoros; Tessler, Michael Henry; Tenenbaum, Joshua B (June 2021, ArXivorg)

   null (Ed.)

   The Abstraction and Reasoning Corpus (ARC) is a set of tasks that tests an agent’s ability to flexibly solve novel problems. While most ARC tasks are easy for humans, they are challenging for state-of-the-art AI. How do we build intelligent systems that can generalize to novel situations and understand human instructions in domains such as ARC? We posit that the answer may be found by studying how humans communicate to each other in solving these tasks. We present LARC, the Language-annotated ARC: a collection of natural language descriptions by a group of human participants, unfamiliar both with ARC and with each other, who instruct each other on how to solve ARC tasks. LARC contains successful instructions for 88% of the ARC tasks. We analyze the collected instructions as ‘natural programs’, finding that most natural program concepts have analogies in typical computer programs. However, unlike how one precisely programs a computer, we find that humans both anticipate and exploit ambiguities to communicate effectively. We demonstrate that a state-of-the-art program synthesis technique, which leverages the additional language annotations, outperforms its language-free counterpart. 

   more » « less
4. TLdR: Policy Summarization for Factored SSP Problems Using Temporal Abstractions

   Sarath Sreedharan, Siddharth Srivastava (May 2020, Proceedings of the International Conference on Automated Planning and Scheduling)

   As more and more people are expected to work with complex AI-systems, it becomes more important than ever that such systems provide intuitive explanations for their decisions. A prerequisite for holding such explanatory dialogue is the ability of the systems to present their proposed decisions to the user in an easy-to-understand form. Unfortunately, such dialogues could become hard to facilitate in real-world problems where the system may be planning for multiple eventualities in stochastic environments. This means for the system to be effective, it needs to be able to present the policy at a high-level of abstraction and delve into details as required. Towards this end, we investigate the utility of temporal abstractions derived through analytically computed landmarks and their relative ordering to build a summarization of policies for Stochastic Shortest Path Problems. We formalize the concept of policy landmarks and show how it can be used to provide a high level overview of a given policy. Additionally, we establish the connections between the type of hierarchy we generate and previous works in temporal abstractions, specifically MaxQ hierarchies. Our approach is evaluated through user studies as well as empirical metrics that establish that people tend to choose landmarks facts as subgoals to summarize policies and demonstrates the performance of our approach on standard benchmarks. 

   more » « less
5. TLdR: Policy Summarization for Factored SSP Problems Using Temporal Abstractions

   Sreedharan, Sarath; Srivastava, Siddharth; Kambhampati, Subbarao (May 2020, Proceedings of the International Conference on Automated Planning and Scheduling)

   J. Christopher Beck, Olivier Buffet (Ed.)

   As more and more people are expected to work with complex AI-systems, it becomes more important than ever that such systems provide intuitive explanations for their decisions. A prerequisite for holding such explanatory dialogue is the ability of the systems to present their proposed decisions to the user in an easy-to-understand form. Unfortunately, such dialogues could become hard to facilitate in real-world problems where the system may be planning for multiple eventualities in stochastic environments. This means for the system to be effective, it needs to be able to present the policy at a high-level of abstraction and delve into details as required. Towards this end, we investigate the utility of temporal abstractions derived through analytically computed landmarks and their relative ordering to build a summarization of policies for Stochastic Shortest Path Problems. We formalize the concept of policy landmarks and show how it can be used to provide a high level overview of a given policy. Additionally, we establish the connections between the type of hierarchy we generate and previous works in temporal abstractions, specifically MaxQ hierarchies. Our approach is evaluated through user studies as well as empirical metrics that establish that people tend to choose landmarks facts as subgoals to summarize policies and demonstrates the performance of our approach on standard benchmarks. 

   more » « less