Topic: US Magistrate Judge Andrew J. Peck (SDNY) Endorses Algorithmic Discovery (Read 3346 times)
May 08, 2012, 01:22:10 AM
We still can't seem to teach computers to do something that comes very naturally to an infant, namely identify human faces well. Search for cvdazzle and you'll come up with a website dedicated to 'playing peekaboo' with the public camera systems in London. Fashionablely.
May 08, 2012, 12:51:21 AM
So throwing more computing power at the problem will not help. The problems for AI are not that the processing power is insufficient. It is that the known algorithms are inefficient. Human brainpower is needed, not computer megaflops. And only genius mathematician computer programmers in blue sky research programs tend to invent breakthrough new algorithms for AI. There's at least 100 programmers working on mundane business software and games for each 1 working on making a computer understand how to see like a human.
May 08, 2012, 12:46:02 AM
The sweatshop document review is going to be taken over by AI, for sure, probably by 2020-25ish.
If you don't know much about AI and are like "wtf is taking so long?" Then read about Moravecs paradox http://en.wikipedia.org/wiki/Moravec's_paradox
The main sticking points have been, like most AI research, the ones that you would think are not sticky at all. But the hard part is actually taking all these disparate documents, the handwriting, the pictures, the fonts, the grainy fuzz, the titling, the from and to, and all the little things that need to be taken into account when digging for the smoking gun, and putting that into a searchable data structure. In other words, a human still has to actually read most documents because current programs would miss critical information.
for instance, in an IP case, you might get things like this:
that is hard for a computer to read at all, letter for letter. And even harder to take in the spatial relationship aspect. Also when unwinding an e-mail chain, a computer is going to have a hard time understanding the social interaction element. Like who is responsible for saying what. And what is big time and what is small time. Those social cues that go back to the pack animal in us, the alpha, beta, omega roles. And further, the herd, the school of fish, the bacterial colony.
Other than that, yeah another essay to stack on top of thousands of others about how seeking a law degree is a fool's errand, especially if you are not in the top schools. Some say only top 14 is worth it anymore. I'm more inclined to think top 100. But I've met many talented and successful attorneys from the bottom tier schools, so don't count anybody out just yet.
I'm just going to leave this here:
If you don't know much about AI and are like "wtf is taking so long?" Then read about Moravecs paradox http://en.wikipedia.org/wiki/Moravec's_paradox
The main sticking points have been, like most AI research, the ones that you would think are not sticky at all. But the hard part is actually taking all these disparate documents, the handwriting, the pictures, the fonts, the grainy fuzz, the titling, the from and to, and all the little things that need to be taken into account when digging for the smoking gun, and putting that into a searchable data structure. In other words, a human still has to actually read most documents because current programs would miss critical information.
for instance, in an IP case, you might get things like this:
that is hard for a computer to read at all, letter for letter. And even harder to take in the spatial relationship aspect. Also when unwinding an e-mail chain, a computer is going to have a hard time understanding the social interaction element. Like who is responsible for saying what. And what is big time and what is small time. Those social cues that go back to the pack animal in us, the alpha, beta, omega roles. And further, the herd, the school of fish, the bacterial colony.
Other than that, yeah another essay to stack on top of thousands of others about how seeking a law degree is a fool's errand, especially if you are not in the top schools. Some say only top 14 is worth it anymore. I'm more inclined to think top 100. But I've met many talented and successful attorneys from the bottom tier schools, so don't count anybody out just yet.
I'm just going to leave this here:
May 07, 2012, 07:09:50 PM
Maybe further context is warranted.
I was thinking along these lines... https://bitcointalksearch.org/topic/m.744541
I was thinking along these lines... https://bitcointalksearch.org/topic/m.744541
May 05, 2012, 07:52:18 PM
http://volokh.com/2012/05/05/is-contract-lawyering-doomed-by-algorithm/
Discovery is far and away the most expensive part of litigation. A highly efficient system could go great lengths towards enhancing judicial access.
In the future, jurisdictions may be forced to choose between competing closed source "algorithmic discovery" options, and (if ever developed) an open source massively distributed system (within a given jurisdiction) a la Bitcoin. Are there specific reasons other procedural functions couldn't be done in a similar way(service of process, notice, etc.)?
Discovery is far and away the most expensive part of litigation. A highly efficient system could go great lengths towards enhancing judicial access.
In the future, jurisdictions may be forced to choose between competing closed source "algorithmic discovery" options, and (if ever developed) an open source massively distributed system (within a given jurisdiction) a la Bitcoin. Are there specific reasons other procedural functions couldn't be done in a similar way(service of process, notice, etc.)?
Jump to: