Skip to content

Latest commit

 

History

History
35 lines (29 loc) · 3.39 KB

capyblanca.md

File metadata and controls

35 lines (29 loc) · 3.39 KB

CAPYBLANCA

Capyblanca has been developed in Delphi. As of May 7th, 2008, the codebase compiles and runs in free (as in beer) turbodelphi for windows.

Future and ongoing activities:

  • A review paper, containing the major findings in analogies in chess and of the Capyblanca architecture, is being written--perhaps for publication as a chapter in the sequel of Fluid Concepts and Creative Analogies.
  • Linhares plans to rewrite a number of AI/cogsci projects that may be of relevance to a future architecture, such as ENTROPICA, KT-Forms, and complex Clusterings.
  • A rewrite of Daniel Defays "Numbo" in Dr. Mahabal's Fluid Concepts framework is in the plans
  • A rewrite of Capyblanca in Dr. Mahabal's Fluid Concepts framework is in the plans
  • An original, Fluid-Concepts-based, re-write of Kemp and Tenembaum's Forms in Mahabal's Framework is in the plans
  • ...as is the idea of a book/(free e-book) on the computer science of building Fluid concepts architectures
  • The long-term goal is to create a computational architecture that, without knowing the rules or goals of a game, can come to understand and play combinatorics games like checkers, chess, or go.
  • Another long-term goal is to create a computational architecture that is able to solve Bongard problems and Raven's matrices. Extending Phaeaco beyond what is currently possible.
  • A paper on a "measure of human intuition" is in the plans, after Dr. Eric Nichols agreed that the idea and the methodology seems to make sense.
  • Some ideas on a technical, mathematical, definition of what is commonly called strong AI or AI-Complete are being sketched. They use mostly combinatorics and information theory... but they stem from the solving of Bongard problems.

If you are interested in joining these future activities, please get in touch with Alex Linhares.

Undergraduate exercises:

  1. Computer chess programs basically expand the decision tree. Explain why people play in a fundamentally different form. Cite the evidence for the human approach.
  2. Consider the model of chunking proposed by the computational cognitive science CHREST. What are the arguments that it falls short?
  3. How can Capyblanca "see abstractions"? Can CHREST do the same?
  4. Why is it claimed that Chase and Simon did a mistake in their classic paper "Perception in chess"? What is the evidence for that claim?
  5. A master chess player made loads of mistakes in reconstructing the board (after seeing it twice for 5 seconds only). Why does Linhares claim that Figure 4 of this paper is "remarkable"?
  6. What is meant by "strategically equivalent" chess positions?
  7. What is meant by "abstract representational invariance"?
  8. What is meant by "the abstraction is the most efficent level of description"?

Programming challenges

  1. Port to python 3. To be started soon. Progress will be posted on this repo.
  2. Afterwards, a port to Abhijit Mahabal's Fluid Concepts Framework.

Open research problems

  1. How can Capyblanca learn by observing? What kinds of new architectural elements are needed?