4 ms·
But at least there's a way to test it: - Give some subjects the advice from this program, and some subjects advice from a knowledgeable co-worker of the person
by SilasX 12y ago
But at least there's a way to test it:
- Give some subjects the advice from this program, and some subjects advice from a knowledgeable co-worker of the person in question (PiQ). (Make sure they don't know which the advice is coming from.)
- Find some way to measure the "success" of the subjects in question -- maybe have them and the PiQ give a subjective rating of the interactions, which should be good enough.
- See if there's a significant difference between the two.
If all such advice is unhelpful, or this service is no better than astrology, you'll see no difference.
- easong 12y agoAn even easier test - give subjects the advice for a different person than the one they're writing to. If there's a similar success rate, then the personality advice is meaningless. Crystal does have a couple cool things going for it, though - for instance, they're in an excellent position to improve user writing quality by presenting standard stylistic advice (don't use passive voice, don't re-use words, etc) as advanced AI clairvoyance.
- analog31 12y agoThis is similar to a criterion that I once proposed when a personality test was given to the employees at my workplace: Would I recognize myself in the report if I found it on the mailroom floor without my name on it? You could take the reports, remove the names, and let people try to guess which person belonged to which report. To make an analogy, imagine finding a room full of un-labeled bottles of chemicals, and an inventory of chemical names. Could you match the chemicals to their names?
- thaumasiotes 12y ago> give subjects the advice for a different person than the one they're writing to. If there's a similar success rate, then the personality advice is meaningless Not quite. This would mean that the personality advice is unrelated to the personality of the target. But it could still be good advice! It's quite possible that a machine that always gave you the same advice, no matter who you were writing to, would still improve your results. (What if the advice was "Are you drunk? Are you sure you want to send this?")