2 ms·

But you don't train the algorithm to make the same decision humans did. You train it to maximize your objective function. I algorithmically trade the stock mar

But you don't train the algorithm to make the same decision humans did. You train it to maximize your objective function. I algorithmically trade the stock market. I don't train my model by asking whether it trades the way I would. The whole point is for it to do a better job than me! I train the model to maximize my returns [1] in backtests and simulations - if it trades differently than I do, so much the better! [1] More precisely, volatility adjusted returns, and I also penalize model complexity to avoid overfitting.

lmm 10y ago
> But you don't train the algorithm to make the same decision humans did. You train it to maximize your objective function. In the idea case you would, sure. In practice mortgages take 25 years to deliver outcome data and the point isn't necessarily to get better outcomes, the mortgage provider would be content to get exactly the same outcomes (or even slightly worse outcomes) if it let them replace a large number of human employees with an automated system. In most fields of human endeavour you can't backtest, because you don't have access to the outcomes of the decisions you didn't make. And the data often aren't as clear-cut or objective as market prices. The stock market is great but it's atypical of modern "big data" in many ways.
- yummyfajitas 10y ago
  In most fields of human endeavor you do have an objective function better than "do what humans do". Mortgages are very specifically an area where you do. First of all, there is historical data. Second of all, you can backtest well before 25 years. A couple of weeks ago I wrote a blog post explaining specifically how to make measurements in the presence of delayed reactions - I'm discussing a situation involving sensor networks, literally the same mathematics would work for mortgage default or refinance: https://www.chrisstucchio.com/blog/2016/delayed_reactions.html https://www.chrisstucchio.com/blog/2016/delayed_reactions.ht... Third, mortgage lenders can often backtest alternate decisions because there are pretty straightforward relationships between decisions. Some of them are even mechanistic, e.g. refinance_risk(interest_rate) and default_prob(interest_rate) are monotonically increasing. You seem to think we are living in the exact specific dark age necessary to make your morality play poignant and relevant. That's about as silly as Star Trek landing on all sorts of alien planets, each one designed to highlight one specific social issue from USA 1966.
  - lmm 10y ago
    There's no historical data on the default rates of people who didn't get mortgages, because they didn't get mortgages. I'm willing to believe that theoretical solutions exist. I know from direct personal experience in the big data/lending industry that they are not always applied. If you are claiming that real-world lenders never train their models on human decisions then you are simply wrong. Nice zinger - I hope you weren't saving it for too long. But I'm not going to get into witticisms or personal arguments.
    - ves 10y ago
      To my knowledge, more sophisticated lenders will let through loans that have poor model scores to fill in that data for training future models.
    - yummyfajitas 10y ago
      Sure, but there is plenty of historical data on people who marginally got mortgages. If those people are solidly profitable, as opposed to marginally profitable, you try a marginally more expansive lending criteria. I.e., if I know f(0) = 10, f(1) = 9, f(2) = 8, but I don't have data on f(3), it's worth running a bit of an experiment to see if f(3) is actually 7. (I happen to know from experience I can't talk about that this analysis is regularly done, albeit keeping the Lucas Critique in mind.) I don't claim that real world use of ML is perfect. I claim that "bias" (in the sense of making wrong decisions due to race) is a statistical problem and the solution is simply better algorithms rather than Cathy O'Neil's statistical nihilism.
      - rlucas 10y ago
        You're right within the boundaries of what you're considering, meaning testing the profitability of marginal borrowers, etc. But you're missing the big picture question about the use of ML in making political decisions (and yes, the question of who gets mortgages is very much a political one). For example, in the earliest days of the creation of the US national mortgage market, conforming to FHA standards meant following more or less explicit racial guidelines (not to mention car-promoting platting and urban planning guidelines). That amounted to a political decision to reward people who invested (and/or resided) in white suburbia with greater liquidity and hence price appreciation. Focusing on the marginal profitability of individual loans ignore the bigger issue, which is that choosing the rules often encodes a much larger set of political preferences or goals. Then, the critique of using opaque algos is, for me, less that they hide the use of some verboten input variable like race or sex, but that by focusing on the algos we ignore the bigger question. (it's a subclass of what I call the scalar fallacy, where any real world outcome being reduced to a scalar always collapses a multidimensional space into a single dimension, and that projection contains a huge set of silent preferences)

lmm 10y ago

> But you don't train the algorithm to make the same decision humans did. You train it to maximize your objective function. In the idea case you would, sure. In practice mortgages take 25 years to deliver outcome data and the point isn't necessarily to get better outcomes, the mortgage provider would be content to get exactly the same outcomes (or even slightly worse outcomes) if it let them replace a large number of human employees with an automated system. In most fields of human endeavour you can't backtest, because you don't have access to the outcomes of the decisions you didn't make. And the data often aren't as clear-cut or objective as market prices. The stock market is great but it's atypical of modern "big data" in many ways.

yummyfajitas 10y ago
In most fields of human endeavor you do have an objective function better than "do what humans do". Mortgages are very specifically an area where you do. First of all, there is historical data. Second of all, you can backtest well before 25 years. A couple of weeks ago I wrote a blog post explaining specifically how to make measurements in the presence of delayed reactions - I'm discussing a situation involving sensor networks, literally the same mathematics would work for mortgage default or refinance: https://www.chrisstucchio.com/blog/2016/delayed_reactions.html https://www.chrisstucchio.com/blog/2016/delayed_reactions.ht... Third, mortgage lenders can often backtest alternate decisions because there are pretty straightforward relationships between decisions. Some of them are even mechanistic, e.g. refinance_risk(interest_rate) and default_prob(interest_rate) are monotonically increasing. You seem to think we are living in the exact specific dark age necessary to make your morality play poignant and relevant. That's about as silly as Star Trek landing on all sorts of alien planets, each one designed to highlight one specific social issue from USA 1966.
- lmm 10y ago
  There's no historical data on the default rates of people who didn't get mortgages, because they didn't get mortgages. I'm willing to believe that theoretical solutions exist. I know from direct personal experience in the big data/lending industry that they are not always applied. If you are claiming that real-world lenders never train their models on human decisions then you are simply wrong. Nice zinger - I hope you weren't saving it for too long. But I'm not going to get into witticisms or personal arguments.
  - ves 10y ago
    To my knowledge, more sophisticated lenders will let through loans that have poor model scores to fill in that data for training future models.
  - yummyfajitas 10y ago
    Sure, but there is plenty of historical data on people who marginally got mortgages. If those people are solidly profitable, as opposed to marginally profitable, you try a marginally more expansive lending criteria. I.e., if I know f(0) = 10, f(1) = 9, f(2) = 8, but I don't have data on f(3), it's worth running a bit of an experiment to see if f(3) is actually 7. (I happen to know from experience I can't talk about that this analysis is regularly done, albeit keeping the Lucas Critique in mind.) I don't claim that real world use of ML is perfect. I claim that "bias" (in the sense of making wrong decisions due to race) is a statistical problem and the solution is simply better algorithms rather than Cathy O'Neil's statistical nihilism.
    - rlucas 10y ago
      You're right within the boundaries of what you're considering, meaning testing the profitability of marginal borrowers, etc. But you're missing the big picture question about the use of ML in making political decisions (and yes, the question of who gets mortgages is very much a political one). For example, in the earliest days of the creation of the US national mortgage market, conforming to FHA standards meant following more or less explicit racial guidelines (not to mention car-promoting platting and urban planning guidelines). That amounted to a political decision to reward people who invested (and/or resided) in white suburbia with greater liquidity and hence price appreciation. Focusing on the marginal profitability of individual loans ignore the bigger issue, which is that choosing the rules often encodes a much larger set of political preferences or goals. Then, the critique of using opaque algos is, for me, less that they hide the use of some verboten input variable like race or sex, but that by focusing on the algos we ignore the bigger question. (it's a subclass of what I call the scalar fallacy, where any real world outcome being reduced to a scalar always collapses a multidimensional space into a single dimension, and that projection contains a huge set of silent preferences)