Permalink
Switch branches/tags
Nothing to show
Find file
Fetching contributors…
Cannot retrieve contributors at this time
46 lines (26 sloc) 6.36 KB
I'd like to see researchers and commentators focus on more practical stuff. For example, I've been reading OB since the start, and although I find the more theoretical stuff (such as Eliezer's posts) very interesting, I seem to have profited little despite reading thousands upon thousands of words. While it's fine to cover all sorts of interesting little biases and ideas, there seems to be more theory than praxis; another way I might put it is that by focusing on low probability issues, we ignore realizing the payoffs.
# The lowdown
What real-life recommendations have we come up with? By real-life, I mean it is something one could start doing today, not a general recommendation like 'respect majority opinions'. An example: if I wanted to memorize material such as foreign language vocabulary, a real-life recommendation would be to use a flashcard program that exploits the [spacing effect](Spaced repetition) (such as SuperMemo or Mnemosyne).
Here's a few I can think of:
- Invest in index funds; you can't beat the market.
- Don't spend too much on medicine; it's not as useful as you think.[^medicine]
- By default, believe what the prediction markets say about politics and events; you are more partial and less accurate than you think, as is are the media and your favorite pundits.
-- If you still think the markets are wrong, prove it by making money off them.
For that matter, aren't we neglecting an entire field? For example, there seem to me to be a lot of possibilities to come up with algorithms or software to correct for human frailties.
# Mental prosthetics
Here's a suggestion. Just occasionally dabbling in prediction markets isn't going to help us increase our predictions much. We're only going to put our money at risk when we're certain, and we'll only risk a little. Only risking a little, we get very little feedback. Sometimes this feedback can be [very useful](Prediction markets) - I learned quite a bit from losing money on Kerry in 2004, and making a mint on McCain and Obama for nominees in 2008. But most of the time it isn't that useful. What does one really learn from selling out and profiting a few percentage points once or twice a month? Not a whole lot.
## Gauging our politics
So why not write a program which sucks down all the existing bets from Intrade, IEM, Betfair etc. and has the user make a bet on `x` markets a day. Throw in some crude confidences ('I think McCain will win the popular vote, with 70% confidence')[^ui]. Then it tracks the market until resolution.
Given that it's September, many markets will close relatively soon, at which point the user can review her performance and find out whether they were overconfident, underconfident etc. Further refinements: periodically ask the user for a fresh assessment; the software could check against Intrade, say, and if the user is out of whack with market prices and very confident, then ask the user to either buy shares or revise their estimates; various other statistical tests could be run (did the user assess as the markets did, but lagged them? Or did they exhibit foresight?) and so on.
## Home economics
That's an example which I think is clearly OB-y, but I'll admit it's a little dicey.
Looking around an average middle-class American living room, and you'll often find an enormous collection of DVDs - a collection so enormous that it's unlikely that the owner has watched the entire collection twice, much less once! It's pretty easy to accumulate a collection larger than one can watch - just TV show box sets can fill an astonishing amount of time. I have a large black binder that contains all my DVDs (and much else besides); 3 or 4 of its 50 pages contains the Monty Python DVD collection. At a runtime of just over 24 full hours, those pages could use up more than a month, assuming I watched an hour each and every day.
But these DVDs can be awfully expensive; at 20 or 30$ a piece[^python], that swiftly adds up. An owner would often be better off just renting a copy or downloading something she's only going to watch once, instead of buying it. I think a major reason people buy instead of some cheaper approach is not just the convenience, but as insurance against the unlikely event they'll love the film. Their thinking goes, 'If I like it, then by buying I'll have it for later; but if I rent/download and like it, then I will need to buy it in addition to the money I already spent to rent! That would be outrageous! So I'll buy just to be safe. Plus, then I don't have to return it or watch it immediately.'
They buy because of the uncertainty about whether they'll like it. But it's not like this is impossible to estimate! If one's TV periodically asked how much one liked a particular show or movie, and also kept track of watches, it would be easy for it to then tell the user that she is buying too many things, or that a particular show is a rent and not a buy for her. (A MythTV plugin could probably do all this.)
Just the basic statistical data ('Buying is 10x more expensive than renting; you would break even buying instead of renting with only 5% of your collection. Please take this into consideration.') would be a big help. --We can't really blame people for [base rate neglect](!Wikipedia) if they don't know the base rate.
The plugin could be even smarter than that. There's nothing stopping such a plugin from integrating with NetFlix: storing user ratings, querying it about how recommended something is[^netflix], etc.
There are all sorts of ways our faint memories of the past and fallible powers of inference & induction can be improved upon by software, but so far precious few of them have wormed their way into our daily life where they could be so helpful. This is too bad...
[^ui]: Confidence levels are actually pretty easy from a UI perspective, I think. Provide 10 buttons, for 0% to 100% in 10% intervals; you'll note that this doesn't leave a button for 50%, which is a good thing as it makes the confidence level a [forced choice](!Wikipedia "Ipsative"). A user has to make some sort of specific commitment; this prevents users from chickening out and always picking 50%.
[^medicine]: Apologies to Dr. Hanson if I've misinterpreted his postings.
[^netflix]: Obviously the more highly recommended for a user an item is, the greater the odds it is a 'buy' and not a 'rent' item.
[^python]: _Monty Python's Flying Circus_ was a very large box-set, so the per-disc costs works out to something very reasonable like 10$/disc.