The Busiest Man On The Internet Forum

[09042014]

http://www.forbes.com/sites/kashmirhill ... -internet/

[FSK]

.

[09042014]

This dude is awesome......

I wanna know how he just did law school as a gag and then got DPW?

[iliketurtles123]

Genuinely curious...

Would his software actually threaten biglaw jobs? The article implies that the reason he was efficient was due to the software.
His sole purpose of going to law school was to destroy biglaw jobs (so I've heard).
Is this an actual threat?

[papercut]

Genuinely curious...

Would his software actually threaten biglaw jobs? The article implies that the reason he was efficient was due to the software.
His sole purpose of going to law school was to destroy biglaw jobs (so I've heard).
Is this an actual threat?

The kind of AI you'd need to replace lawyers would be so advanced we'd have to have give it vacation days and a dental plan.

So, no.

[09042014]

Genuinely curious...

Would his software actually threaten biglaw jobs? The article implies that the reason he was efficient was due to the software.
His sole purpose of going to law school was to destroy biglaw jobs (so I've heard).
Is this an actual threat?

Some jobs, but not nearly as many as he probably thinks. I've only done maybe 100 hours of doc review in my 8 months in big law and that's because I volunteered for two projects. Most cases use external contract attorneys for doc reviewing. Some even use "predictive coding" aka robots to do doc review.

My understanding is that top tier big law firms actually do more doc review than the little guys because their clients are willing to pay top dollar for the best no matter what.

I can't speak to transactional work.

[Ohiobumpkin]

Last time I checked, computer code cannot sit for the bar. Doubt you will ever take the human out of legal services.

[FSK]

Practicing Attorneys ITT: How much of your work do you think a computer system could plausibly do? Would your clients be willing to pay for a computerized version of these services?

[papercut]

Practicing Attorneys with CS degrees ITT: How much of your work do you think a computer system could plausibly do? Would your clients be willing to pay for a computerized version of these services?

[brazleton]

.

[haus]

Last time I checked, computer code cannot sit for the bar. Doubt you will ever take the human out of legal services.

I suspect that 20 years ago, many travel agents felt the same way about computer solutions in their industry.

[09042014]

Last time I checked, computer code cannot sit for the bar. Doubt you will ever take the human out of legal services.

I suspect that 20 years ago, many travel agents felt the same way about computer solutions in their industry.

Maybe legal jobs will be replaced in 20 years, but if you can replace law jobs, you'll have already replaced most other jobs. Legal jobs aren't low hanging fruit.

[rayiner]

Practicing Attorneys ITT: How much of your work do you think a computer system could plausibly do? Would your clients be willing to pay for a computerized version of these services?

My degree is in AE, but I used to be a software developer, if that counts.

How predictive coding works is that a human attorney sits down and classifies an initial corpus of documents that are "interesting." Usually, you need reference set of fifty to a couple of hundred documents. The software then looks through all your other documents to find ones that are similar (and hopefully interesting). At no point does the software really understand what it's doing--it's looking for feature similarity between the reference set and the whole database.

When you actually understand how Google, etc, works, I think you have to be more skeptical about the potential of existing technology to replace human knowledge workers. AI hasn't advanced much since the 1980's. Most of what seems like "AI" is actually machine learning + crowd sourced data. Google does a good job giving you relevant search results because you're the thousandth person to search that particular term today. But with legal services, you can't crowd-source data. And without that big corpus of data, the machine learning algorithms aren't very useful.

Look at Google Scholar, for example. It's total shit for searching for cases you don't already have the citation of. Why? Because it works the same way as Google's search engine: it measures relevance using search terms and the number of references to a document. The net result is that you'll get a high-profile SCOTUS case like Brown v. Board at the top of your results when you're looking for a totally un-sexy educational law issue.

Heck, take something exceedingly simple and rote: replace references to "X" in a form motion with references to "X, Y, and Z." How long before we have software that can do this simple first-year job without the partner having to carefully re-read it and check for errors? I think we'd need a fundamental breakthrough in AI to get to that point.

[papercut]

Practicing Attorneys ITT: How much of your work do you think a computer system could plausibly do? Would your clients be willing to pay for a computerized version of these services?

My degree is in AE, but I used to be a software developer, if that counts.

How predictive coding works is that a human attorney sits down and classifies an initial corpus of documents that are "interesting." Usually, you need reference set of fifty to a couple of hundred documents. The software then looks through all your other documents to find ones that are similar (and hopefully interesting). At no point does the software really understand what it's doing--it's looking for feature similarity between the reference set and the whole database.

When you actually understand how Google, etc, works, I think you have to be more skeptical about the potential of existing technology to replace human knowledge workers. AI hasn't advanced much since the 1980's. Most of what seems like "AI" is actually machine learning + crowd sourced data. Google does a good job giving you relevant search results because you're the thousandth person to search that particular term today. But with legal services, you can't crowd-source data. And without that big corpus of data, the machine learning algorithms aren't very useful.

Look at Google Scholar, for example. It's total shit for searching for cases you don't already have the citation of. Why? Because it works the same way as Google's search engine: it measures relevance using search terms and the number of references to a document. The net result is that you'll get a high-profile SCOTUS case like Brown v. Board at the top of your results when you're looking for a totally un-sexy educational law issue.

Heck, take something exceedingly simple and rote: replace references to "X" in a form motion with references to "X, Y, and Z." How long before we have software that can do this simple first-year job without the partner having to carefully re-read it and check for errors? I think we'd need a fundamental breakthrough in AI to get to that point.

This is right on. Natural language is a helluva lot more complicated than most people realize.

[09042014]

It's not even natural language, it's critical thinking. AI isn't even a little bit close that.

I'd love to get my hands on the software and give it a whirl on my case to see what it came up with. But I wouldn't rely on it at all.

[papercut]

It's not even natural language, it's critical thinking. AI isn't even a little bit close that.

I'd love to get my hands on the software and give it a whirl on my case to see what it came up with. But I wouldn't rely on it at all.

Yeah, I was just trying to set the bar lower than full blow critical thinking, but the point is the same.

[Anonymous User]

I just hope that I can pay off my loans before being replaced by AI. At that point tho, it's a win win. Cheaper bills foir businesses and I get out of being a lawyer

[haus]

It's not even natural language, it's critical thinking. AI isn't even a little bit close that.

I'd love to get my hands on the software and give it a whirl on my case to see what it came up with. But I wouldn't rely on it at all.

A lot of what gets billed for falls well short of critical thinking.

[09042014]

It's not even natural language, it's critical thinking. AI isn't even a little bit close that.

I'd love to get my hands on the software and give it a whirl on my case to see what it came up with. But I wouldn't rely on it at all.

A lot of what gets billed for falls well short of critical thinking.

Yea but's usually wrapped up in critical thinking. I don't think I could really automated anything I do unless I had AI that could think like a person.

We can't even get paralegals to do shit correctly.

[brazleton]

.

[09042014]

Well, indeed there is much a first year can do that software is not close to replacing. But if you think that machines can't currently beat humans in relevance review of a large document set, saving much time and money, then you simply don't understand the current state of the technology. The fact that it has not become popular in more firms and cases has more to do with inertia and structural barriers than with the limitations of the machines. This is not debatable stuff.

I think it is pretty popular right?

[FSK]

I doubt that a lot of what partners do could be replaced by computers. Deciding on strategy negotiating key terms, etc.

[rayiner]

Well, indeed there is much a first year can do that software is not close to replacing. But if you think that machines can't currently beat humans in relevance review of a large document set, saving much time and money, then you simply don't understand the current state of the technology. The fact that it has not become popular in more firms and cases has more to do with inertia and structural barriers than with the limitations of the machines. This is not debatable stuff.

Predictive coding is very good for large document sets, but the training period makes it impractical to use for smaller ones. That said, most of the territory that predicative coding covers was already ceded to contract attorney shops over the 2000's. If a first year at a large firm is still doing relevance review, it's either because the document set is too small to make it cost-effective to outsource (in which case there's a decent chance it's too small for predictive coding to be worthwhile), or because the client wants a $300/hour lawyer looking at each document.

I think over time technology will make more experienced associates more efficient, and reduce the need for large entry-level classes. This will reduce the number of jobs, but in a way will really be a step back towards the era before electronic discovery created the need for huge first-year classes in the first place. I don't think it's going to categorically replace any particular class of legal job, though the increased pressure might cause second review work and the like to shift from associates down to contract attorneys as predictive coding makes relevance coders obsolete.

[brazleton]

.

[brazleton]

.