I understand your point, but would still argue my own. Yes, machines are good at making certain predictions or classifications, such as credit loan assessments. However, I draw a distinction between analysis, and actually making a decision. Even in these relatively simple cases (little ambiguity, clear-cut decision rules, and plenty of historical data), assuming the same information was given to a human decision maker (ie. the classification results, and so on), the ability to apply common sense trumps the machine’s actual decision making abilities every time. In most cases, a competent decision maker would commit to the same course of action recommended by a computer. In a few cases, though, this should be able to be overridden, and there need to be processes in place to allow for this.

Indeed, this is exactly what happens in banks: at least in Australia, credit approval is always ultimately authorised by some person, even with the use of credit application analysis systems. This is required by legislation (I assume) as well as for sound business reasons.

I guess my position is that I have no problem with systems providing recommendations to decision-makers. However, I do have a problem with full automation of a decision-process (even in ATMs - my card’s been eaten enough times to justify that position!) without recourse to human oversight and approval or review of decisions. Full automation is risky, and as alluded to before, morally questionable. The question isn’t ‘can decision processes be fully automated?’, but rather ’should they be?’ My answer to the first is yes, but only for very limited, simple decision problems. My answer to the second is probably not.

Interesting stuff, though!