From Costanza's original thread (entire text):
This is for anyone in the LessWrong community who has made at least some effort to read the sequences and follow along, but is still confused on some point, and is perhaps feeling a bit embarrassed. Here, newbies and not-so-newbies are free to ask very basic but still relevant questions with the understanding that the answers are probably somewhere in the sequences. Similarly, LessWrong tends to presume a rather high threshold for understanding science and technology. Relevant questions in those areas are welcome as well. Anyone who chooses to respond should respectfully guide the questioner to a helpful resource, and questioners should be appropriately grateful. Good faith should be presumed on both sides, unless and until it is shown to be absent. If a questioner is not sure whether a question is relevant, ask it, and also ask if it's relevant.
Meta:
- How often should these be made? I think one every three months is the correct frequency.
- Costanza made the original thread, but I am OpenThreadGuy. I am therefore not only entitled but required to post this in his stead. But I got his permission anyway.
"Acausal" is used as a contrast to Causal Decision Theory (CDT). CDT states that decisions should be evaluated with respect to their causal consequences; ie if there's no way for a decision to have a causal impact on something, then it is ignored. (More precisely, in terms of Pearl's Causality, CDT is equivalent to having your decision conduct a counterfactual surgery on a Directed Acyclic Graph that represents the world, with the directions representing causality, then updating nodes affected by the decision.) However, there is a class of decisions for which your decision literally does have an acausal impact. The classic example is Newcomb's Problem, in which another agent uses a simulation of your decision to decide whether or not to put money in a box; however, the simulation took place before your actual decision, and so the money is already in the box or not by the time you're making your decision.
"Acausal" refers to anything falling in this category of decisions that have impacts that do not result causally from your decisions or actions. One example is, as above, Newcomb's Problem; other examples include:
There are a number of acausal decision theories: Evidential Decision Theory (EDT), Updateless Decision Theory (UDT), Timeless Decision Theory (TDT), and Ambient Decision Theory (ADT).
In EDT, which originates in academia, casuality is completely ignored, and only correlations are used. This leads to the correct answer on Newscomb's Problem, but fails on others- for example, the Smoking Lesion. UDT is essentially EDT, but with an agent that has access to its own code. (There's a video and transcript explaining this in more detail here).
TDT, like CDT, relies on causality instead of correlation; however, instead of having agents chose a decision that is implemented, it has agents first chose a platonic computation that is instantiated in, among other things, the actual decision maker; however, is is also instantiated in every other algorithm is equal, acausally, to the decision maker's algorithm, including simulations, other agents, etc. And, given all of these instantiations, the agent then choses the utility-maximizing algorithm.
ADT...I don't really know, although the wiki says that it is "variant of updateless decision theory that uses first order logic instead of mathematical intuition module (MIM), emphasizing the way an agent can control which mathematical structure a fixed definition defines, an aspect of UDT separate from its own emphasis on not making the mistake of updating away things one can still acausally control."