@909 wrote:
Folks here like to say that action A correlates to risk factors that should result in a lower score because that's what the algorithm says should happen. But get 10 people in a room, evaluate the person who made A happen, e.g. refi a mortgage but has 20 years of excellent credit, and 9 may say the algorithm is making an incorrect conclusion by lowering that person's credit score.
Sure, credit scores change because the algorithm says they should because that's how humans designed the algorithm but that doesn't mean the algorithm isn't being capricious.
Today's credit scoring algorithms are rudimentary at best: they may deliver reasonable results in general but when it comes down to specific humans they can fail badly.
The threat of algorithms truly harming individuals is real. The OP makes a valid point that we need to guard against bad algorithms making bad decisions that affect people badly. But that'll never happen so just be sure y'all are on the right side of the algorithms.
@909 first and foremost, I totally agree with the bolded portion.
in reference to what proceeded that: People inherently take into account more than just the relevant data when coming to a conclusion, you're right. That's one of the advantages of using computers to make a lot of decisions, because they make the decisions based on the criteria set forth. Computers are not capable of capricious actions, like humans, that's one of the advantages of using them, imho.
Would those 10 people in the room be making that conclusion based on their feelings & experience, or only based on the data and the correlations shown by data research, or some mixture?
because I don't think there's anyone who can legitimatly argue against the fact of the correlation between opening new accounts and increased risk. That doesn't mean everyone that opens up a new account is going to default obviously, but you've got to admit the numbers are increased compared to those who do not.
Humans may feel 20 years of perfect history warrants approval. I'd be willing to bet the algorithm still has the score quite high as well, which would likewise assist in approval. But does that mean a perfect score? Could there be no better risk?
So, I have to respectfully disagree regarding the algorithm being capricious. There is nothing erratic, unpredictable or whimsical about it. It follows established rules it's been programmed to follow. Anyone with basic credit knowledge could've predicted a score decrease for the refi which in of itself refutes its being capricious.
Now, if you feed it the same data and come up with two different results, then yeah it's capricious, but that's not possible.
If it's following rules set forth to come to its answer, then there's nothing capricious or whimsical about it, imho; I would argue it is the antithesis of caprice. We may not like the answer, we may feel differently, but we like to have data confirm our conclusions and we actively try to conform it to do so.
Confirmation bias is real. That's a great thing about
computers, they can be programmed to do something without bias. By the same token, they can be programmed to do something with bias, so your warning is on point.
These algorithms were created and designed to predict risk based on the strength of correlations, without bias, something nearly impossible for a human to do, imho.
does that mean some people get inappropriate scores? Sure nothings perfect and in implementing this system they have just divided us in the groups and that doesn't necessarily mean that we will perform like others in that group. But, it does mean that we had multiple correlations that seemed indicate we should be in that group. And for now that's as far as they have progressed. But don't worry, pretty soon AI will take over. Caprice may become ubiquitous then?
