Bootstrapping new knowledge from abstract representations
A long tradition in developmental psychology has used formal scientific inquiry as a basis for understanding learning in early childhood. But much of this work has focused on situations in which children observe firsthand the covariation between parts of a causal system, or can intervene directly on...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2022
|
| Online Access: | https://hdl.handle.net/1721.1/140120 |
| _version_ | 1811068621093339136 |
|---|---|
| author | Pelz, Madeline C. |
| author2 | Schulz, Laura E. |
| author_facet | Schulz, Laura E. Pelz, Madeline C. |
| author_sort | Pelz, Madeline C. |
| collection | MIT |
| description | A long tradition in developmental psychology has used formal scientific inquiry as a basis for understanding learning in early childhood. But much of this work has focused on situations in which children observe firsthand the covariation between parts of a causal system, or can intervene directly on the system in order to test and refine their hypotheses. While these studies point to impressive inferential abilities, both formal science and everyday reasoning also require us to make inferences about hidden generative processes even without any direct evidence. In this thesis, I aim to address scenarios in which young children can bootstrap new knowledge using 1) knowledge about their own knowledge, 2) knowledge about probable underlying generative processes, and 3) knowledge about high-level properties linking causal events. My approach includes a combination of computational modeling, and behavioral data from both adults and young children (ages 4-8 years).
The first set of experiments demonstrates that adults and children can metacognitively represent the amount of information they might need to solve a particular statistical reasoning problem, suggesting that young children have precise metacognitive access to their own knowledge. The second study demonstrates that adults and children can infer an agent’s mental state and goals based only on a trace left on the environment, suggesting that children can identify hidden underlying generative processes and use them as the basis for rich inferences. The third study demonstrates that children can use high-level properties in order to link causal events, using features that are preserved across simple causal functions in order to match effects to their candidate causes. Taken together, these findings suggest that even if children do not have access to covariation data necessary to establish a relationship through statistical evidence, they can rely on other subtle sources of information in order to bootstrap new knowledge in a variety of domains. |
| first_indexed | 2024-09-23T07:58:40Z |
| format | Thesis |
| id | mit-1721.1/140120 |
| institution | Massachusetts Institute of Technology |
| last_indexed | 2024-09-23T07:58:40Z |
| publishDate | 2022 |
| publisher | Massachusetts Institute of Technology |
| record_format | dspace |
| spelling | mit-1721.1/1401202022-02-08T04:08:23Z Bootstrapping new knowledge from abstract representations Pelz, Madeline C. Schulz, Laura E. Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences A long tradition in developmental psychology has used formal scientific inquiry as a basis for understanding learning in early childhood. But much of this work has focused on situations in which children observe firsthand the covariation between parts of a causal system, or can intervene directly on the system in order to test and refine their hypotheses. While these studies point to impressive inferential abilities, both formal science and everyday reasoning also require us to make inferences about hidden generative processes even without any direct evidence. In this thesis, I aim to address scenarios in which young children can bootstrap new knowledge using 1) knowledge about their own knowledge, 2) knowledge about probable underlying generative processes, and 3) knowledge about high-level properties linking causal events. My approach includes a combination of computational modeling, and behavioral data from both adults and young children (ages 4-8 years). The first set of experiments demonstrates that adults and children can metacognitively represent the amount of information they might need to solve a particular statistical reasoning problem, suggesting that young children have precise metacognitive access to their own knowledge. The second study demonstrates that adults and children can infer an agent’s mental state and goals based only on a trace left on the environment, suggesting that children can identify hidden underlying generative processes and use them as the basis for rich inferences. The third study demonstrates that children can use high-level properties in order to link causal events, using features that are preserved across simple causal functions in order to match effects to their candidate causes. Taken together, these findings suggest that even if children do not have access to covariation data necessary to establish a relationship through statistical evidence, they can rely on other subtle sources of information in order to bootstrap new knowledge in a variety of domains. Ph.D. 2022-02-07T15:25:13Z 2022-02-07T15:25:13Z 2021-09 2021-11-12T14:44:14.641Z Thesis https://hdl.handle.net/1721.1/140120 In Copyright - Educational Use Permitted Copyright MIT http://rightsstatements.org/page/InC-EDU/1.0/ application/pdf Massachusetts Institute of Technology |
| spellingShingle | Pelz, Madeline C. Bootstrapping new knowledge from abstract representations |
| title | Bootstrapping new knowledge from abstract representations |
| title_full | Bootstrapping new knowledge from abstract representations |
| title_fullStr | Bootstrapping new knowledge from abstract representations |
| title_full_unstemmed | Bootstrapping new knowledge from abstract representations |
| title_short | Bootstrapping new knowledge from abstract representations |
| title_sort | bootstrapping new knowledge from abstract representations |
| url | https://hdl.handle.net/1721.1/140120 |
| work_keys_str_mv | AT pelzmadelinec bootstrappingnewknowledgefromabstractrepresentations |