Thank you so much for your answers - this is all very helpful! Especially the fact that people are choosing different things for different reasons
I’m going to explain some of the reasoning behind the questions, but in a spoiler box so that anyone else who feels moved to answer can do it without seeing the man behind the curtain
One of the factors I’m interested in is local versus global explanations - an explanation that works for your own personal situation versus one that has wide coverage. In Bob’s case the simplest ‘local’ explanation is that he knows PHP and nobody else does - in general it looks like most people aren’t finding themselves satisfied with just saying that - they want to bring in factors that are good for both hirees. Many people aren’t thinking of PHP as a factor, presumably because it doesn’t work for Aya. This is interesting stuff, and goes against some theory in some papers I may be citing
The smallest necessary-and-sufficient (characteristic of all the people who are hired, not characteristic of any people not hired) explanation in the programming case is “Doesn’t know Ruby”. I was expecting people to completely discount this because of prior knowledge of the world - nobody’s going to bomb out of an interview because of unneccessary knowledge they *do *have. So I was interested to see a couple of people tangentially mention it. The next-smallest is “Javascript+SQL” which was - as expected - highly popular
I’m not sure if anyone noticed, but the rental table is the same data as the programmers’ table, relabelled and with the columns shifted about. There are two big differences between the situations. In the programmer case, we know which is the ‘good’ value and which the ‘bad’ - in the rental case it’s more on personal preference. And in the programmer case, the decisions for “success” are all being made by one institution, whereas for rentals it’s different people with different possible criteria. I must confess I was only thinking about the first of these differences when I made up the table but I think it may be the second which is responsible for the fact that large numbers of people zeroed in on location (the ‘local’ explanation) as the reason for #2’s success (though it could also be that location is higher up on our personal ordering of ‘reasons why people choose dwelling places’ - it was a lot harder to balance that factor in the second table than the first). The equivalent to the first table’s “Javascript+SQL” was “expensive and near trains” - that was a very unpopular explanation. The equivalent of “no Ruby” in the first was “2br” in the second, and people do seem to be more likely to consider this - some in an “it seems real weird but…” kind of way
1a) They needed people competent in Javascript and SQL.
1b) They needed people competent in Javascript and SQL.
2a) The resident wanted to live in Carlton.
2b) The resident that wanted to live in Fitzroy wanted to live in a house.
In looking at the second table I out-of-hand rejected the idea that anybody would prefer a more expensive place or a smaller place on those ‘merits’, due to duh, and presumed that location was always going to be of paramount importance, because duh. Making the (unwarranted) presumption that the data in the table was relevant to the problem, that left House/Unit as the only difference between 1 and 5, so house it was.
I’m not sure if anyone noticed, but the rental table is the same data as the programmers’ table, relabelled and with the columns shifted about. There are two big differences between the situations. In the programmer case, we know which is the ‘good’ value and which the ‘bad’ - in the rental case it’s more on personal preference.
I’m not sure I agree.
[spoiler]I feel that while you have mapped out the lists in the same places, there two lists are not equivalent. For example, the location of a rental unit (Carlton or Fitzroy) can be a positive or a negative. People might avoid the Fitzroy neighborhood because it’s a high crime area. There’s no equivalent negative value in knowing computer languages. Somebody might get hired because they know Java and SQL. But they wouldn’t get turned down for the job if they know Java, SQL, and Python. Knowing that additional language may not have helped but it wouldn’t have hurt.
There’s a similar issue with the bedrooms. Any rental unit with three bedrooms is also a rental unit with two bedrooms just as any rental unit with two bedrooms is also a rental unit with one bedroom. Again, this is not the same as programming languages, where knowing one language does not include knowledge of a different language as a subset.[/spoiler]
Maybe I should also use spoilers, but in general expanding on what Little Nemo said:
[spoiler] Everything in the apartment descriptions, except probably the amount of rent, has the same problem: it’s either an advantage or a disadvantage depending on the particular renter. “Close to trains” is an advantage to somebody who wants to take the train frequently, but not to anyone who doesn’t; it’s a disadvantage to somebody who likes quiet. Some people prefer smaller places, which are easier to clean – or, as Nava said, fewer but larger rooms. It’s very likely that some people prefer Fitzroy to Carlton, and others prefer Carlton to Fitzroy – even if one neighborhood is generally considered “better” than the other, which is information that we aren’t given, someone might prefer the “worse” neighborhood because they want to live close to someone else who lives there, or because they work on the same block, or because they feel more comfortable with its particular mix of people, or because it allows easy access to some particular amenity most people don’t care about but they do. Some people prefer gardens, others don’t, because gardens take time and care. Some people prefer apartments because they require less care, others prefer houses because they’re more private.
Even the rent could go both ways, because some people see status value in having more expensive things.
I can’t think of any disadvantage of knowing an additional computer language that the employer doesn’t use; at least, unless the employer thinks that knowing it will affect the employee’s frame of mind in a fashion the employer doesn’t want, but if the employee’s competent in the languages desired that doesn’t seem very likely.[/spoiler]
“We know which is the good values and which is the bad” if we know Ruby as anything other than a shiny crystal thingee one puts on rings and if we know what each of those languages is used for. I wouldn’t be able to tell PHP from PAP, so that’s knowledge that I can’t take into account.
My understanding of this thread was: What would be a human-acceptable way to justify these decisions, that a well-tuned A.I. could come up with?
For the employment question, if an A.I. gave an answer such as “Bob was hired because he knows PHP” to an average person, and they could see the raw data, they would say it’s incompetent.
For the rental question, any answer claiming to find a clear pattern based on that table would earn the “incompetent” label.
Of course, in reality, for an A.I. to be well-tuned, you need much larger data sets for the training. If we’re supposing that the A.I. was trained on a larger data set and then writes a justification for the small subset we’re shown in the tables, then an answer would be more acceptable if it included a phrase such as “based on analysis of 300 recent candidates at ACME”, etc. That adds credibility, like a white lab coat.
Also, something that’s been glossed over so far: in the employment situation, we’re explicitly told that ACME bases its decision on knowledge of programming languages. For the rental situation, there’s no such hint (and indeed, in real life, there wouldn’t be).
