Recommended Posts





I'll try to organize some of my posts into categories here.


My Approach

These are posts which provide some degree of exposition, explaining what approaches I think are most worthwhile.

Why Relational Methods: I lay out my basic referential generality requirement (a general machine intelligence system should be able to represent any kind of pattern that a human can) and its most important consequence (relational methods are needed for general intelligence).
Why Logic: I explain why I prefer to think about logical knowledge representations rather than procedural knowledge representations.
Beliefs: I briefly review what we know about how local probabilistic beliefs can usefully interact.


Questioning My Approach

Although laying out my views on my blog has long-term value to me, I get more short-term value by trying to disprove my views, hopefully to improve them.

Truth and AI: Reasons to doubt my referential generality requirement.
Bayesianism?: A continuation, with a more direct questioning of Bayesianism.
My Unhealthy Obsession with Message Passing: I really like the idea of local beliefs which interact via message-passing algorithms. Is this healthy?
Losing Faith in Factor Graphs: I outline why I've given up thinking of factor graphs as the unifying formalism for probabilistic reasoning.

The Procedural Logic Sequence

This is a sequence of posts trying to find an answer to the vague question "How can we make a logic which exists inside time?". What I mean by this is a logic which includes a representation of its own inference process. The goal of the line of thinking shifts over time, though.

Self-Modifying Logic: Not really starting to talk about procedural logic yet, but it's where the sequence started.
More on Self-Modifying Logic: Here I start to talk about (self-modifying) inference control.
Action Logic: Starts the topic in earnest, giving an overview of existing systems from a "logic of action" perspective.
Procedural Logic: Re-naming the project because "action logic" is already taken, I continue the line of thought.
Procedural Logic Once Again: A year later, a quick summary of the thoughts I've had in that time. Enthusiasm about Dyna begins.
Procedural Logic Answers: More Dyna enthusiasm, as well as some thoughts about generalizations.
Useful Meta-Analysis: A few more things to think about concerning self-guidance.

Truth, Finally

This is a sequence in which I examine a possible route to a satisfying theory of truth.

Old Demons: I try to convince myself not to worry about it.
Impredicative Truth: A solution is proposed.
New Truth: Intuition: I examine the solution a bit.
Problems I Didn't Solve: I outline some stuff this theory doesn't do.