2 ms·
> Abstraction tends to lossy and context-insensitive, which is why an algorithm will say "no" to a loan request based on a credit score that ignores personal co
by franga2000 4y ago
> Abstraction tends to lossy and context-insensitive, which is why an algorithm will say "no" to a loan request based on a credit score that ignores personal context, where a human loan manager who knows the people involved might say yes - for good and valid reasons.
A human loan manager, given only a credit score, will approve the loan based on only the credit score. The problem isn't human vs machine, it's context-awareness and context-sensitivity vs a lack of one or both.
All large systems, human or machine, tend to be implemented using abstractions. Layers of context-stripping abstractions were a part of buerocracy far before machines took it over. If a loan request form doesn't have an "if I don't get this loan my family will literally starve to death" box that you can check, the evaluator will not know about that context couldn't possibly act based on it. The form can be digital or paper and the evaluator can be a program or a human - it doesn't change a thing.
Similarly, context-sensitivity (the willingness to let the context sway your decision) was strictly limited even back in the day when decision makers were close enough to the source to be aware of the context. A bank clerk had to stick to the rules and couldn't just decide to give a loan to someone that didn't qualify under them. If they did, they'd be risking losing their job or worse, so they didn't. This discretion was only given to the people in charge - and that hasn't changed with computers. A person in a high enough position can override what the computer says, just as easily as they can override a lower clerk's decision.
Yes, making machines context-aware takes some effort, and making them context-sensitive takes even more effort. But even if machines magically came with both, their operators would do the best to get rod of them, just like they did with their employees. From their point of view, empathy is inefficiency. And that's the real problem.