Logo Logo
Help
Contact
Switch language to German
Knowledge transfer in cognitive systems theory: models, computation, and explanation
Knowledge transfer in cognitive systems theory: models, computation, and explanation
Knowledge transfer in cognitive systems can be explicated in terms of structure mapping and control. The structure of an effective model enables adaptive control for the system's intended domain of application. Knowledge is transferred by a system when control of a new domain is enabled by mapping the structure of a previously effective model. I advocate for a model-based view of computation which recognizes effective structure mapping at a low level. Artificial neural network systems are furthermore viewed as model-based, where effective models are learned through feedback. Thus, many of the most popular artificial neural network systems are best understood in light of the cybernetic tradition as error-controlled regulators. Knowledge transfer with pre-trained networks (transfer learning) can, when automated like other machine learning methods, be seen as an advancement towards artificial general intelligence. I argue this is convincing because it is akin to automating a general systems methodology of knowledge transfer in scientific reasoning. Analogical reasoning is typical in such a methodology, and some accounts view analogical cognition as the core of cognition which provides adaptive benefits through efficient knowledge transfer. I then discuss one modern example of analogical reasoning in physics, and how an extended Bayesian view might model confirmation given a structural mapping between two systems. In light of my account of knowledge transfer, I finally assess the case of quantum-like models in cognition, and whether the transfer of quantum principles is appropriate. I conclude by throwing my support behind a general systems philosophy of science framework which emphasizes the importance of structure, and which rejects a controversial view of scientific explanation in favor of a view of explanation as enabling control.
Not available
Beebe, Cameron
2021
English
Universitätsbibliothek der Ludwig-Maximilians-Universität München
Beebe, Cameron (2021): Knowledge transfer in cognitive systems theory: models, computation, and explanation. Dissertation, LMU München: Graduate School of Systemic Neurosciences (GSN)
[thumbnail of Beebe_Cameron.pdf]
Preview
PDF
Beebe_Cameron.pdf

1MB

Abstract

Knowledge transfer in cognitive systems can be explicated in terms of structure mapping and control. The structure of an effective model enables adaptive control for the system's intended domain of application. Knowledge is transferred by a system when control of a new domain is enabled by mapping the structure of a previously effective model. I advocate for a model-based view of computation which recognizes effective structure mapping at a low level. Artificial neural network systems are furthermore viewed as model-based, where effective models are learned through feedback. Thus, many of the most popular artificial neural network systems are best understood in light of the cybernetic tradition as error-controlled regulators. Knowledge transfer with pre-trained networks (transfer learning) can, when automated like other machine learning methods, be seen as an advancement towards artificial general intelligence. I argue this is convincing because it is akin to automating a general systems methodology of knowledge transfer in scientific reasoning. Analogical reasoning is typical in such a methodology, and some accounts view analogical cognition as the core of cognition which provides adaptive benefits through efficient knowledge transfer. I then discuss one modern example of analogical reasoning in physics, and how an extended Bayesian view might model confirmation given a structural mapping between two systems. In light of my account of knowledge transfer, I finally assess the case of quantum-like models in cognition, and whether the transfer of quantum principles is appropriate. I conclude by throwing my support behind a general systems philosophy of science framework which emphasizes the importance of structure, and which rejects a controversial view of scientific explanation in favor of a view of explanation as enabling control.