3 ms·

I care a lot about outcomes at the individual level. I worry that with these systems, generally health will improve because there will be a level of standardiza

I care a lot about outcomes at the individual level. I worry that with these systems, generally health will improve because there will be a level of standardization, but specifically it will get worse. An analogy for me is with furniture, but you could also substitute financial advice. In furniture we can get a 'quality' table from Ikea for a couple hundred bucks, and we can make 10s of thousands of them. But it's much much harder to get a reasonably priced custom table. 200 years ago when a lot more people spent time building wooden ships, it was much easier to get a carpenter to build a table/furniture at a relatively affordable price point. Now that comparatively orders of magnitude of fewer people are in the practice, the price for that furniture, or custom cabinets, etc. at the same quality level has skyrocketed. In financial advice, the rise of algorithms doing things like asset allocation has caused an entire generation of financial planners to never be hired, and therefore never trained. The old financial planners who do much more than asset allocation are becoming rarer and rarer and no one is replacing them. In 20 years, I firmly believe the specifics of my tax situation, etc will be more difficult to solve than today because I will have a harder time finding this kind of specialized expertise. I see the same in medicine and every other field which low-end machines doing the low-end work. We're not training the future high-level practitioners because they need those 10,000 hours or whatever to become super experts. How long will it take before the machine catches up with the best humans is really the question here.

ghkbrew 10y ago
That's an interesting argument and it certainly seems right, but I can't say if that's because there's truth to it or because I am as "algorithm averse" as the next guy. What's nice is it seems to offer a falsifiable prediction. Namely, that there are "super-experts" out there who consistently beat machine learning algorithms. Even if they aren't numerous enough to bring up the mean. Do the current studies show that?
- guimarin 10y ago
  To answer your question is complicated because of the existence of Centaurs[1]. I will say that the fact that entities like Renaissance [2] exist is proof that there are still some humans that are better than machine learning algorithms. Now whether Renaissance employs Centaurs is hard to tell. My statement though is that machines reduce the number of experts because they reduce the population looking at a problem and my suspicion is that experts are uniformly distributed. So in a population of 100, there might be 1 expert in a population of 10000, 100 experts and 1 super expert, etc. I chose the financial markets because you could say that it is the perfect competitive space for this kind of evaluation, though I wouldn't be surprised if Centaurs or humans existed in other areas as well which are better than machine learning algorithms. Still the point stands that for people in the 91st to 98th percentile a lot of value will probably be lost when humans start going away from practices in droves. Another acceleration for the 1%. 1. https://en.wikipedia.org/wiki/Advanced_Chess https://en.wikipedia.org/wiki/Advanced_Chess 2. https://en.wikipedia.org/wiki/Renaissance_Technologies https://en.wikipedia.org/wiki/Renaissance_Technologies
  - Symmetry 10y ago
    There was an era when human/computer combinations were better at chess than either alone but that era has passed and nowadays pure computer chess programs do better than human/computer teams. We should probably expect to see the same thing in other fields where as computers get better there's a period where centaurs work best before an era where pure AI dominates.
    - guimarin 10y ago
      Do you have a source/ranking I can take a look at here? Hadn't heard of that and it totally changes the game.(no pun intended)
      - Symmetry 10y ago
        Here's a discussion. The amount humans add does depend on how fast the chess game goes and it's 90 minute games where humans have become superfluous. For 1 day per move games humans can still provide useful input last I heard. http://marginalrevolution.com/marginalrevolution/2013/11/what-are-humans-still-good-for-the-turning-point-in-freestyle-chess-may-be-approaching.html http://marginalrevolution.com/marginalrevolution/2013/11/wha...
scott_s 10y ago
> 200 years ago when a lot more people spent time building wooden ships, it was much easier to get a carpenter to build a table/furniture at a relatively affordable price point. I'm going to make a call on that claim. Keep in mind that 200 years ago was 1817, well before the rise of the middle class. I bet that far fewer people could afford the luxury of getting a table even approaching the quality of an Ikea table, let alone what a few $100 extra will get you for a custom table today.
- rpazyaquian 10y ago
  The Industrial Revolution was sometime from 1760 to 1820, I thought?
  - scott_s 10y ago
    Yes. I realized the same thing a few minutes after posting, and edited that part out. The middle class part is what matters most: I think more people now have significantly more purchasing power than 200 years ago.
kwhitefoot 10y ago
> it was much easier to get a carpenter to build a table/furniture at a relatively affordable price What do you call an affordable price in relation to mean, or median, purchasing power? A quick web search finds a lot of one off dining tables in the USD700 to USD3k range in the US. Made to measure dining tables in the UK can be had for less than GBP800. Median household income in the UK 2014 was about GBP24k gross, roughly GBP20k net (https://www.google.no/url?sa=t&rct=j&q=&esrc=s&source=web&cd=3&cad=rja&uact=8&ved=0ahUKEwilr5DFjPrSAhVEnRoKHU9mAAEQFggkMAI&url=https%3A%2F%2Fen.wikipedia.org%2Fwiki%2FIncome_in_the_United_Kingdom&usg=AFQjCNH29KGBUFXHM6msEE0Z-SIsSNAUAQ https://www.google.no/url?sa=t&rct=j&q=&esrc=s&source=web&cd...). So a GBP800 table would be about two weeks of after tax income. Was it really so much less 200 hundred years ago? Perhaps someone can find the data for 200 years ago and do a more sophisticated analysis.

ghkbrew 10y ago

That's an interesting argument and it certainly seems right, but I can't say if that's because there's truth to it or because I am as "algorithm averse" as the next guy. What's nice is it seems to offer a falsifiable prediction. Namely, that there are "super-experts" out there who consistently beat machine learning algorithms. Even if they aren't numerous enough to bring up the mean. Do the current studies show that?

guimarin 10y ago
To answer your question is complicated because of the existence of Centaurs[1]. I will say that the fact that entities like Renaissance [2] exist is proof that there are still some humans that are better than machine learning algorithms. Now whether Renaissance employs Centaurs is hard to tell. My statement though is that machines reduce the number of experts because they reduce the population looking at a problem and my suspicion is that experts are uniformly distributed. So in a population of 100, there might be 1 expert in a population of 10000, 100 experts and 1 super expert, etc. I chose the financial markets because you could say that it is the perfect competitive space for this kind of evaluation, though I wouldn't be surprised if Centaurs or humans existed in other areas as well which are better than machine learning algorithms. Still the point stands that for people in the 91st to 98th percentile a lot of value will probably be lost when humans start going away from practices in droves. Another acceleration for the 1%. 1. https://en.wikipedia.org/wiki/Advanced_Chess https://en.wikipedia.org/wiki/Advanced_Chess 2. https://en.wikipedia.org/wiki/Renaissance_Technologies https://en.wikipedia.org/wiki/Renaissance_Technologies
- Symmetry 10y ago
  There was an era when human/computer combinations were better at chess than either alone but that era has passed and nowadays pure computer chess programs do better than human/computer teams. We should probably expect to see the same thing in other fields where as computers get better there's a period where centaurs work best before an era where pure AI dominates.
  - guimarin 10y ago
    Do you have a source/ranking I can take a look at here? Hadn't heard of that and it totally changes the game.(no pun intended)
    - Symmetry 10y ago
      Here's a discussion. The amount humans add does depend on how fast the chess game goes and it's 90 minute games where humans have become superfluous. For 1 day per move games humans can still provide useful input last I heard. http://marginalrevolution.com/marginalrevolution/2013/11/what-are-humans-still-good-for-the-turning-point-in-freestyle-chess-may-be-approaching.html http://marginalrevolution.com/marginalrevolution/2013/11/wha...

scott_s 10y ago

> 200 years ago when a lot more people spent time building wooden ships, it was much easier to get a carpenter to build a table/furniture at a relatively affordable price point. I'm going to make a call on that claim. Keep in mind that 200 years ago was 1817, well before the rise of the middle class. I bet that far fewer people could afford the luxury of getting a table even approaching the quality of an Ikea table, let alone what a few $100 extra will get you for a custom table today.

rpazyaquian 10y ago
The Industrial Revolution was sometime from 1760 to 1820, I thought?
- scott_s 10y ago
  Yes. I realized the same thing a few minutes after posting, and edited that part out. The middle class part is what matters most: I think more people now have significantly more purchasing power than 200 years ago.

kwhitefoot 10y ago

> it was much easier to get a carpenter to build a table/furniture at a relatively affordable price What do you call an affordable price in relation to mean, or median, purchasing power? A quick web search finds a lot of one off dining tables in the USD700 to USD3k range in the US. Made to measure dining tables in the UK can be had for less than GBP800. Median household income in the UK 2014 was about GBP24k gross, roughly GBP20k net (https://www.google.no/url?sa=t&rct=j&q=&esrc=s&source=web&cd=3&cad=rja&uact=8&ved=0ahUKEwilr5DFjPrSAhVEnRoKHU9mAAEQFggkMAI&url=https%3A%2F%2Fen.wikipedia.org%2Fwiki%2FIncome_in_the_United_Kingdom&usg=AFQjCNH29KGBUFXHM6msEE0Z-SIsSNAUAQ https://www.google.no/url?sa=t&rct=j&q=&esrc=s&source=web&cd...). So a GBP800 table would be about two weeks of after tax income. Was it really so much less 200 hundred years ago? Perhaps someone can find the data for 200 years ago and do a more sophisticated analysis.