[This meetup structured by Ernest, who takes all blame CREDIT for memeposting]

So, you want to maxx your gainz? But wait, what do we mean by "gainz"? What do we mean by "maxx"? What do we mean by "want"? (What do we mean by "you"??)

In this meetup, let's go back to one of ye olde foundations of LW: the notions of payoff or utility maximization. How have perspectives on this matter evolved over time? Have we moved on from expected-utility maximization or have we learned nothing all these years?

Readings

We'll organize this in a partitioned reading format. First read all 3 of these articles to establish mutual knowledge between all of us:

• VNM utility theorem: The traditional explanation for why we supposedly "must" maximize expectation of some utility function (if you want the mathier version, just read the wikipedia article's description of the axioms and theorem statement)
• Kelly criterion: One argument of several for the Kelly criterion, suggesting that we shouldn't just maximize expected wealth (although, while the post doesn't specifically say this, note that the result is equivalent to maximizing expectation of a utility function logarithmic in wealth)
• Allais Malaise: To emphasize that a logarithmic utility function does not resolve some classic challenges like the Allais paradox (if you're unfamiliar with that paradox, pick the post on it as your reading from the list below)

Then pick one of the following articles to read, indicating your choice on the signup sheet here. There's a huge range of different background levels here, so pick one that you find interesting and/or feel comfortable with. The focus is on "prescriptive" theories that recommend what people should do, rather than "descriptive" theories that explain what people actually do. For some of these articles, I've indicated questions you should consider while reading them.

• EY's presentation of the Allais paradox: If you're not familiar with the Allais paradox, pick this (it's okay to have multiple signups on this)
• Kelly isn't just about logarithmic utility
• Kelly is just about logarithmic utility surprise_pikachu.jpg
• Maximizing minimum payoff (under Knightian uncertainty)
• Risk-weighted expected utility: [Slightly mathy] Consider: do you find it valid to call this approach "rational", or is it begging the question by defining it to be rational by fiat? (Optionally, if you like, you can additionally read part 2 and/or part 3)
• Coherent risk measures: [Slightly mathy] Given a choice between portfolios with the same monetary EV, this approach picks the one with lower risk under some measure. Consider: does it violate the VNM theorem postulates, and if so, is it satisfactory to you?
• On the Independence Axiom: [Warning: longpost] For those who're already veterans of the LW discourse around expected utility, and want to see yet another extended counterargument against it
• Write-in vote: if your preferred decision theory isn't one of these, feel free to list it in the signup sheet, but do link some introductory material for it! (+100 points if you know timeless decision theory or updateless decision theory, but that may have to be its own meetup)
  

Discussion

Each person will briefly summarize the contents of their article, and bring up questions they have about it. Other general questions:

• How convincing do you find the postulates of the VNM theorem? If you reject the postulates, do you accept that you can be money-pumped?
• Do you find any alternative decision theories convincing in addressing the above point? (Or more radically, do we get somewhere by rejecting the entire framework, e.g. via Knightian uncertainty? Though note that there's also Savage's utility theorem imposing some barriers.)
• Do you find risk aversion or ambiguity aversion "justifiably" rational under some viewpoint? If so, which, and does it overcome the objections based on the VNM theorem?
• The motivation of Kelly betting via risk-of-ruin or growth rate has some subtle limitations, e.g. it is restricted to "money-like" quantities that are additive, non-negative, can be "re-invested", etc. Are there ways to expand its scope, or do these restrictions make it too limited? More generally, what do you think of counterarguments against Kelly betting?
• Are the LW and/or EA communities stuck in a suboptimal viewpoint by mostly focusing on expected utility maximization? Do you find any alternative proposals convincing?

Location

Meet at Holland Village MRT Exit B (below ground part) at 3 pm. We will wait for everyone to gather for 5-10 mins, and then head over to my nearby apartment. If you've been here before, feel free to come directly to the apartment (though you may need to wait to be let in).

Some snacks will be provided. There are many restaurants and a hawker centre nearby, if you want to buy your own food.