Moral dilemmas engender conflicts between two traditions: consequentialism, which evaluates actions based on their outcomes, and deontology, which evaluates actions themselves. These strikingly resemble two distinct decision-making architectures: a model-based system that selects actions based on inferences about their consequences; and a model-free system that selects actions based on their reinforcement history. Here, I consider how these systems, along with a Pavlovian system that responds reflexively to rewards and punishments, can illuminate puzzles in moral psychology.

I find this (short) paper appealing because it:

  • Makes connections across domains
  • Unifies competing explanations within an overarching framework
  • Offers an elegant explanation of otherwise confusing phenomena e.g. the means/side-effect distinction

That said, it does still seem a bit speculative.




Sorted by Click to highlight new comments since: Today at 7:59 AM

Cf Fiery Cushman:

Dual-system approaches to psychology explain the fundamental properties of human judgment, decision making, and behavior across diverse domains. Yet, the appropriate characterization of each system is a source of debate. For instance, a large body of research on moral psychology makes use of the contrast between “emotional” and “rational/cognitive” processes, yet even the chief proponents of this division recognize its shortcomings. Largely independently, research in the computational neurosciences has identified a broad division between two algorithms for learning and choice derived from formal models of reinforcement learning. One assigns value to actions intrinsically based on past experience, while another derives representations of value from an internally represented causal model of the world. This division between action- and outcome-based value representation provides an ideal framework for a dual-system theory in the moral domain.

+1 and to generalize I think a bunch of philosophical debates are basically reifications of different sorts of ways different pattern matching cognitive systems operate. We let the urge to compress for efficiency reasons get a bit out of hand and try to build perverse monisms out of everything.

This is somewhat unrelated, but I once did a little research on the problem of how DNA gets translated into proteins---I (and some others) viewed DNA as a 4 letter code (nucleotides) or 'syntax' , while proteins were words using a 26 letter (amino acids) 'word' with a 'semantics'---it meant something or had a functional use. I took what is perjoratively called a 'fact free science' approach (associated with SFI/complexity approach) , which meant the idea was to see if one could figure out if there were any patterns in the DNA code (using as little biochemical information as possible --experimentalists dealt with that detail) which could be used to predict which ones might be 'coding regions' for proteins . This is analagous to trying to figure out from some randomly selected 'text' whether its just a randomly generated set of (nonsensical) 'words' , or actually is a meaningful 'book' (maybe shakespeare).

It was assumed that the 'reinforcement history' was actually hidden in the DNA code --i.e. there were dependencies between the 'letters' (nucleotides) so they were not randomly distributed (any more than letters in a book are).