Artificial Intelligence and the Legal Industry Forum

[introversional]

Isn't the law, legal language, and it's constantly changing/developing interpretation far too nuanced a thing to rely on on AI, even in the limited capacity you're suggesting? (even for doc review - at least beyond keyword/phrase searching, which already exists)

Also, a set of programmed instructions can't be disbared - thus, an attny and firm will still be accountable. Will there be glitches? As a microsoft windows user, I suspect the probability of epic case fail is quite high due to "my fucking AI screwing up doc review again - lol." Can a software understand the sensibilities of a judge, cultural shift, and so on?

Anyways, as fallible as we humans are, I don't think we should outsource justice to a machine. Having said that, I do think there's a provacative idea here, but moreso related to the development of more sophisticated legal phrase/keyword/precedent searching, which probably already exists. (but as a 0L haven't yet played around w/Lexis or Westlaw)

Justice would probably be better served by a machine with algorithms to look at the nuances in cases. It would take arbitrariness out.

An algorithm would probably rely on a mathmatical probility to be reached (based on it's mechanized "interpretation" of nuanced data) before producing an answer, or result, one way or another. In this sense, the "arbitrariness" of repetition might be reduced, but the version of justice we'd have left would just be a system of probability based bias. Justice, IMO, almost has to involve some level of arbitrariness.

[flcath]

Con law should be.

[introversional]

Can a software understand the sensibilities of a judge, cultural shift, and so on?

Yes. Software exists that does this - in our brains.

You think our brains = software?

I can't be a part of this ideology. Corporations were already given the rights of natural persons - what's next, will a software be considered a "natural person" on account of it being able to "think" to a degree acceptably similar to our brains? I can think of a few dozen highly totalitarian-esque problems arising from this for humanity.

TL;DR - Lets cut to the chase, you're suggesting us plugging into the Matrix, amiright?

[AreJay711]

Isn't the law, legal language, and it's constantly changing/developing interpretation far too nuanced a thing to rely on on AI, even in the limited capacity you're suggesting? (even for doc review - at least beyond keyword/phrase searching, which already exists)

Also, a set of programmed instructions can't be disbared - thus, an attny and firm will still be accountable. Will there be glitches? As a microsoft windows user, I suspect the probability of epic case fail is quite high due to "my fucking AI screwing up doc review again - lol." Can a software understand the sensibilities of a judge, cultural shift, and so on?

Anyways, as fallible as we humans are, I don't think we should outsource justice to a machine. Having said that, I do think there's a provacative idea here, but moreso related to the development of more sophisticated legal phrase/keyword/precedent searching, which probably already exists. (but as a 0L haven't yet played around w/Lexis or Westlaw)

Justice would probably be better served by a machine with algorithms to look at the nuances in cases. It would take arbitrariness out.

Why? Use the averages to find a hard and fast equation and run with it. The answer would come out as indexes and certainty measurements which could be used for severity of punishment or damages. Of course some legal issues are just not clear so you would still need a human judge there but just to clarify issues the machine runs into. Appellate courts could handle it if something was way off base. It might help or it might not.


An algorithm would probably rely on a mathmatical probility to be reached (based on it's mechanized "interpretation" of nuanced data) before producing an answer, or result, one way or another. In this sense, the "arbitrariness" of repetition might be reduced, but the version of justice we'd have left would just be a system of probability based bias. Justice, IMO, almost has to involve some level of arbitrariness.

[Kohinoor]

Wouldn't the problem there be that, with millions or billions of dollars on the line, we are probably decades if not centuries from relying sufficiently on the results of an AI doc review to not have humans go over their results anyway?

No. http://ridethelightning.senseient.com/2 ... tware.html

http://www.nytimes.com/2011/03/05/science/05legal.html

That's today's technology. About a decade from now, it'll be even better.

I read those already. They do not address the realities of legal practice.

[deadpoetnsp]

Can a software understand the sensibilities of a judge, cultural shift, and so on?

Yes. Software exists that does this - in our brains.

You think our brains = software?

I can't be a part of this ideology. Corporations were already given the rights of natural persons - what's next, will a software be considered a "natural person" on account of it being able to "think" to a degree acceptably similar to our brains? I can think of a few dozen highly totalitarian-esque problems arising from this for humanity.

TL;DR - Lets cut to the chase, you're suggesting us plugging into the Matrix, amiright?

I don't think brains = software. Brains run software. Go to google.com/scholar, search for "biological neural networks" and you will see that there is a huge field of science that says so If you search for "artificial neural networks", you will find the software implementation of neural networks.

An algorithm would probably rely on a mathmatical probility to be reached ...

there exist AI techniques, called "soft" AI techniques, which I named earlier (neural nets, genetic algos, fuzzy logic, etc) which DO NOT rely on hard-coded probabilities. These techniques use supervised and unsupervised learning to analyze and find patterns in data - WITHOUT knowing a priori anything about the data. For instance, I have personally coded neural networks that analyzed pharmaceutical molecular structures to predict their absorption in our bodies.

[introversional]

And the analysis of those molecular structures to predict their absorption in our bodies must have relied on programmed
absorption rates, no? How were absorption rates normalized? How did you account for varying absorption rates in different bodies? Diet, sunlight, genetics, age, etc... all these would affect the absorption rate (and effects) of chemicals/compounds in the body. How did you account for the (potentially millions) of externalities which could affect rates of absorption? Every body and it's absorption of chemicals into the blood stream (and effects) is obviously different. If someone drank grapefruit juice (known to influence absorption rates) that might invalidate the prediction. How about drug interactions? (both known and unknown) How did the coded neural network account for that?

My point is AI helping w/pattern finding could be great, but human consideration/interpretation of the results will always be necessary.....

[deadpoetnsp]

My point is AI helping w/pattern finding could be great, but human consideration/interpretation of the results will always be necessary.....

If you read each of my previous posts on this topics carefully, you will realize that they agree with your last statement. I have always described scenarios in which AIs assist humans, not replace them

As for the details of the bioabsorption study, well, it's really too involved to get into. But these are research articles describing similar studies:
http://pubs.acs.org/doi/abs/10.1021/jm991030j
http://pubs.acs.org/doi/abs/10.1021/ci020013r
http://pubs.acs.org/doi/abs/10.1021/ci980029a
http://portal.acm.org/citation.cfm?id=1144042
http://pubs.acs.org/doi/abs/10.1021/ci0202741
--LinkRemoved--

And you can find hundreds of similar articles.

I am not talking about some "mumbo-jumbo plug-in molecular structures and get absorption by magic" Keep Arthur C. Clarke's quote in mind

[SOCRATiC]

Wouldn't the problem there be that, with millions or billions of dollars on the line, we are probably decades if not centuries from relying sufficiently on the results of an AI doc review to not have humans go over their results anyway?

No. http://ridethelightning.senseient.com/2 ... tware.html

http://www.nytimes.com/2011/03/05/science/05legal.html

That's today's technology. About a decade from now, it'll be even better.

I read those already. They do not address the realities of legal practice.

It answers your question.

[AreJay711]

I don't think brains = software. Brains run software. Go to google.com/scholar, search for "biological neural networks" and you will see that there is a huge field of science that says so If you search for "artificial neural networks", you will find the software implementation of neural networks.

An algorithm would probably rely on a mathmatical probility to be reached ...

there exist AI techniques, called "soft" AI techniques, which I named earlier (neural nets, genetic algos, fuzzy logic, etc) which DO NOT rely on hard-coded probabilities. These techniques use supervised and unsupervised learning to analyze and find patterns in data - WITHOUT knowing a priori anything about the data. For instance, I have personally coded neural networks that analyzed pharmaceutical molecular structures to predict their absorption in our bodies.

Nice, I'm gonna save this to read until I want to fall asleep tonight

[deadpoetnsp]

Nice, I'm gonna save this to read until I want to fall asleep tonight

Let me know if you felt enlightened in the morning