Structuring an AGI research

History / Edit / PDF / EPUB / BIB
Created: September 4, 2015 / Updated: December 14, 2017 / Status: in progress / 5 min read (~950 words)

  • Extract references automatically based on what is said in a text

  • Create a directory on my computer where I store everything related to my research
  • For each topic I find interesting, create a directory within that directory and start collecting information about the topic. It might be ideas, questions, notes, url of relevant websites, book titles, authors, etc.
  • Frequently go through the list of directories in the main directory in order to review the state of each "project" and add new ideas, questions and information when appropriate
  • From time to time a project may take a good amount of my time. For instance, if I become interested in handwriting recognition, I might do a lot of research on the topic and thus update the content of the directory more frequently than for other topics which I might only have had a cursory interest

I have two systems I use to approach research:

  • reading about theory and then devising projects
  • devising projects and reading the appropriate theory

The most general questions every AGI researcher needs to answer include:

  • What is AGI, accurately specified?
  • Is it possible to build the AGI as specified?
  • If AGI is possible, what is the most plausible way to achieve it?
  • Even if we know how to achieve AGI, should we really do it?

A complete AGI work normally includes:

  • a theory of intelligence,
  • a formal model of the theory,
  • a computational implementation of the model.

Source: https://sites.google.com/site/narswang/home/agi-introduction#TOC-AGI-Basics

Define the problems and questions you are trying to solve.

  • This is what I am solving
  • This is what I am building
  • This is how I am solving it
  • Here is how we can think about it now

Source: ?

  • State the problem you want to solve in general terms
  • Collect previous work related to your research
  • Understand the material
  • Implement the ideas found in previous work
  • Specify your problem in details
  • Think (and use PĆ³lya How to Solve It)
  • Implement your idea and compare it/Review
  • Report/Publish
  • Discuss topic with colleagues

Source: http://www.umiacs.umd.edu/~knkim/HowToPhd.htm

  • Good research should be novel:
    • Describe state-of-the-art (SotA)
    • Describe state-of-the-practice (SotP)
    • Describe how your work is different from them
  • Good research should be relevant:
    • Consider what problems we face today or are likely to face within some time span
    • Consider what other researchers and practitioners problems are and what they consider important
  • Good research should present generalities/principles
  • Good research is often systematic and structured
    • Systematic: You have a clear idea of what to do and that this will clearly "cover" the most likely relevant aspects
    • Structured: There is good "logic" and "flow" in what you are trying to do and how you describe it
  • Good research claims something and validates those claims

  • Focus on a part of your subject area that is limited so that you can go deep in 40-150 papers
  • Create a taxonomy of the papers you find

Source: http://www.robertfeldt.net/advice/feldt_guide_to_starting_a_phd.pdf