Construal level theory (CLT) suggests that a person creates abstract mental representations, known as construals, in order to generate predictions, form counterfactuals, and guide behavior with respect to distal times, places, and actions . This paper takes a step towards implementing CLT in agent reasoning; the impact of abstraction level on an ability to scavenge for needed items and ingredients is investigated. Our approach was parametrically tested in a Minecraft environment. Results show that planning with construals increased trial success rate by 14.8% as compared to planning without construals. Our work lays the foundation for a family of cognitively-plausible models that would allow computational agents to generate predictions about future events and valuations of future plans based on very limited prior training.
|CEUR Workshop Proceedings
|Published - 2022
|2022 Thinking Fast and Slow and Other Cognitive Theories in AI, a AAAI 2022 Fall Symposium, TFSOCTAI 2022 - Arlington, United States
Duration: Nov 17 2022 → Nov 19 2022
All Science Journal Classification (ASJC) codes
- General Computer Science