Mechanical Turkers paid $1 are better at predicting recidivism than secret, private-sector algorithms


#1

Originally published at: https://boingboing.net/2018/01/18/mechanical-turkers-paid-1-are.html


#2

For anyone curious, that silver gavel sculpture is here in Columbus, Ohio outside the state’s supreme court building. It’s the backdrop for pretty much every attorney in town’s commercial/professional headshot/website etc.


#3

I’m surprised that these scientists are surprised; it’s been common knowledge that these algorithms are horrid and corrupt for ages now.


#4

“Julia and I are sitting there thinking: How can this be?” Farid says. “How can it be that this software that is commercially available and being used broadly across the country has the same accuracy as mechanical turk users?”

Well, see, humans aren’t very good at AI yet. So, having a human brain counts for a lot in making sense of real world data. I’d think the scientists involved would not be so surprised at that.


#5

Like the complicated personnel system that we have at work, these algorithms aren’t supposed to be BETTER and more accurate. They’re just supposed to be more difficult to challenge.


#6

How can this be? I mean, if I am working in the justice system and am looking at the results I am going to look at the cost of the system analyzing the same 400 people. Is it less than $400? There, that’s how this can be.


#7

So wisdom of the crowds wins again?


#8

Going by scientific definition, what’s the threshold (statistically speaking) to consider something being better than random chance? Tried searching it but i wasn’t sure how to properly word my question.

I would assume a coin flip is 50-50, so a piece of software being 65% could arguably say that it’s better than a coin flip. I’m not making that argument myself, but i was just curious why a 65% accuracy is acceptable… maybe it isn’t but then what percentage would be the threshold for acceptable?


#9

Why the the government is allowed to buy proprietary, closed-source software, I’ll never understand.


#10

Devil’s advocate here:

Private enterprise buys proprietary closed source software, should the government really be unable to do so? It would mean skills are less likely to be transferable (for example private: Excel, or rarely Numbers, government: well none of the open source spreadsheets actually do the full role of what Excel does, but some are pretty close, I guess one of those, plus some extra funding to get pivot tables and such; now get 6 years experience in private or government pushing numbers and try to transfer out…).

I personally like Open Source software, but that doesn’t mean there is always a profit incentive to write it. So if the government isn’t allowed to buy the stuff, they will have to contract out for someone to write it just for them (or have that talent in-house). For things that are sufficiently different from all existing solutions this can be cost effective, but for things that more or less exist (say written for another government, or for large enterprises) it will cost a lot more to write a new one from scratch.

I mean I personally like writing code, but more the kind that controls a laser cutter then evaluates criminal records and predicts recidivism. So if nobody is going to pay me, I’m controlling the laser. If they are going to pay me, I would rather work on something I can sell more then once (say, to each of 50 states) then to something only one state will buy, and then they will give to all the other states. It’s not just greed here, I do have a mortgage to pay for, and food to put on the table. Plus I need to keep buying wood for the laser cutter…


#11

I’m a web dev for a non-profit, and I write apps all day that manage educational resources and track professional development program work and outcomes. All for (usually) government contracts. They are welcome to every bit of source I write, as I’m paid a salary to write it and maintain it. I’m not a great dev, I couldn’t hack it in silicon valley, but I’m damn good at the combination of skills it takes to understand the issues surrounding my work (Sociology undergrad, a master’s in teaching and decades of development experience). For most government, public-benefiting development, you don’t need to blow anyone’s mind, you’re just helping bureaucrats be a little more efficient. On the face of it, it’s dull work, but I love the environment, our professional mission and my colleagues. I am a weekend warrior game dev, and love that more, but you’d have to drag me kicking and screaming into a for-profit game development environment.


#12

Does it even exist?

(See current results of elections and referendums, success of Facebook, viewer rates and circulation for anything out of Mr Burn’s Murdoch’s diverse “news” outlets, … you get the gist.)


#13

This topic was automatically closed after 5 days. New replies are no longer allowed.