Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning.
The ability to acquire abstract knowledge is a hallmark of human intelligence and is believed by many to be one of the core differences between humans and neural network models. Agents can be endowed with an inductive bias towards abstraction through meta-learning, where they are trained on a distri...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Public Library of Science (PLoS)
2023-08-01
|
| Series: | PLoS Computational Biology |
| Online Access: | https://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1011316&type=printable |
| _version_ | 1827807291703820288 |
|---|---|
| author | Sreejan Kumar Ishita Dasgupta Nathaniel D Daw Jonathan D Cohen Thomas L Griffiths |
| author_facet | Sreejan Kumar Ishita Dasgupta Nathaniel D Daw Jonathan D Cohen Thomas L Griffiths |
| author_sort | Sreejan Kumar |
| collection | DOAJ |
| description | The ability to acquire abstract knowledge is a hallmark of human intelligence and is believed by many to be one of the core differences between humans and neural network models. Agents can be endowed with an inductive bias towards abstraction through meta-learning, where they are trained on a distribution of tasks that share some abstract structure that can be learned and applied. However, because neural networks are hard to interpret, it can be difficult to tell whether agents have learned the underlying abstraction, or alternatively statistical patterns that are characteristic of that abstraction. In this work, we compare the performance of humans and agents in a meta-reinforcement learning paradigm in which tasks are generated from abstract rules. We define a novel methodology for building "task metamers" that closely match the statistics of the abstract tasks but use a different underlying generative process, and evaluate performance on both abstract and metamer tasks. We find that humans perform better at abstract tasks than metamer tasks whereas common neural network architectures typically perform worse on the abstract tasks than the matched metamers. This work provides a foundation for characterizing differences between humans and machine learning that can be used in future work towards developing machines with more human-like behavior. |
| first_indexed | 2024-03-11T21:55:19Z |
| format | Article |
| id | doaj.art-fed6275593ea46bb83f1ad41bb5aa7bb |
| institution | Directory Open Access Journal |
| issn | 1553-734X 1553-7358 |
| language | English |
| last_indexed | 2024-03-11T21:55:19Z |
| publishDate | 2023-08-01 |
| publisher | Public Library of Science (PLoS) |
| record_format | Article |
| series | PLoS Computational Biology |
| spelling | doaj.art-fed6275593ea46bb83f1ad41bb5aa7bb2023-09-26T05:30:52ZengPublic Library of Science (PLoS)PLoS Computational Biology1553-734X1553-73582023-08-01198e101131610.1371/journal.pcbi.1011316Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning.Sreejan KumarIshita DasguptaNathaniel D DawJonathan D CohenThomas L GriffithsThe ability to acquire abstract knowledge is a hallmark of human intelligence and is believed by many to be one of the core differences between humans and neural network models. Agents can be endowed with an inductive bias towards abstraction through meta-learning, where they are trained on a distribution of tasks that share some abstract structure that can be learned and applied. However, because neural networks are hard to interpret, it can be difficult to tell whether agents have learned the underlying abstraction, or alternatively statistical patterns that are characteristic of that abstraction. In this work, we compare the performance of humans and agents in a meta-reinforcement learning paradigm in which tasks are generated from abstract rules. We define a novel methodology for building "task metamers" that closely match the statistics of the abstract tasks but use a different underlying generative process, and evaluate performance on both abstract and metamer tasks. We find that humans perform better at abstract tasks than metamer tasks whereas common neural network architectures typically perform worse on the abstract tasks than the matched metamers. This work provides a foundation for characterizing differences between humans and machine learning that can be used in future work towards developing machines with more human-like behavior.https://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1011316&type=printable |
| spellingShingle | Sreejan Kumar Ishita Dasgupta Nathaniel D Daw Jonathan D Cohen Thomas L Griffiths Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning. PLoS Computational Biology |
| title | Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning. |
| title_full | Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning. |
| title_fullStr | Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning. |
| title_full_unstemmed | Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning. |
| title_short | Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning. |
| title_sort | disentangling abstraction from statistical pattern matching in human and machine learning |
| url | https://journals.plos.org/ploscompbiol/article/file?id=10.1371/journal.pcbi.1011316&type=printable |
| work_keys_str_mv | AT sreejankumar disentanglingabstractionfromstatisticalpatternmatchinginhumanandmachinelearning AT ishitadasgupta disentanglingabstractionfromstatisticalpatternmatchinginhumanandmachinelearning AT nathanielddaw disentanglingabstractionfromstatisticalpatternmatchinginhumanandmachinelearning AT jonathandcohen disentanglingabstractionfromstatisticalpatternmatchinginhumanandmachinelearning AT thomaslgriffiths disentanglingabstractionfromstatisticalpatternmatchinginhumanandmachinelearning |