I do not use AI in my legal work.
Some of my clients use AI.
I saw an AI output this morning, purporting to give information about data protection law (whether an IP address is personal data).
It included two cases, which do not exist. It did not include any case law which does exist. It gave no warning that the output was fiction.
Even if they can get over the ethical concerns (which, currently, I cannot), it would be cheaper, probably more fun, and no less inaccurate to ask a passing toddler.
Ethical concerns butter no parsnips, it seems.
And getting the message across that the output of AI is unreliable, and not "better than nothing", is difficult.
So I asked for another small experiment, with another, ostensibly simple, non-legal, query.
The problem was incredibly obvious. This wasn't legal nuance, or a mistyped citation. Just utter nonsense.
"Fancy relying on this for anything now", I wondered...
@elexia Yes!
I think I saw something, years ago, in the context of flying, and how it was better for a pilot's instruments to fail off, than fail wrong.
@neil My philosophy has been IF we ignore the ethical concerns then what is the up side of using AI?
Currently AI is a crap shoot on how right it is and even IF AI was accurate 100% of the time the act of using it is making it less likely for people to learn for themselves making it a net negative. I feel like I have noticed similar with people of all ages not being able to navigate without GPS anymore.
So currently there is little to no good reason to use GenAI.
Yeah. I saw this just this morning, and thought: this is how we're being conditioned to accept AI killer robots.
BBC News - Ukraine using AI drones to strike vital convoys supplying Russian troops
https://www.bbc.co.uk/news/articles/cdjp0n7rn41o
@neil The other thing about the "better than nothing" argument that some people make is the alternative isn't nothing. We already have established methods and resources for all these things!
(Including ascii donkeys - this took me literally 2 seconds to find: https://ascii.co.uk/art/donkey )
But people throughout the developed world, including me, are now being told to use AI or lose our jobs. At my age, no one's going to give me another software development job, I have no other marketable skills, and I can't afford to retire. So I have to choose between using AI, with all its practical and ethical problems, and failing to provide for Mrs Wife in our old age.
Yeah, there are occasions when AI code-completion saves significant amounts of typing, and you do get a buzz from that. But they're outnumbered by the times when it goes off in the wrong direction, or produces buggy code, code that won't compile, or duplicated code that ought to be factored out. I'm not yet convinced that it makes me any more productive, and I'm certain that AI fanboys don't take into account the long-term costs of buggier and less maintainable code.
I've started to wonder whether there's such a thing as AI offsetting, in much the way that some people who still fly offset their carbon emissions. Is there a way I can keep my job but do something to counterbalance the evils of AI?
> But people throughout the developed world, including me, are now being told to use AI or lose our jobs.
Yes. That was at the heart of one of my toots earlier in the week: will there come a tipping point where people feel they have no realistic choice, because of their employers. That may have already happened for some / many.
But there is a distinction between those forced to adopt AI because of their reliance on their employer, and those uncritically promoting it.
(And, yes, someone will make a "following orders is no defence" argument.)
Agreed, it's an important distinction. The motives for promoting or opposing AI are important, too.
I can't believe the naïveté of my colleagues, who built AI to automate everyone else out of a job and thought it would never come for them. And I can't believe the gall of senior managers who use AI to replace everyone else's job but never, ever their own.
Artificial Intelligence and legal work - a crucial case to read There was a routine court application which would usually go through on nod, but the judge suspected something was not right and asked questions, and this then came out. If you must use AI, be careful. https://www.bailii.org/ew/cases/EWHC/Ch/2026/1199.html
There are some genuinely fascinating parallels there with interactions in online worlds, and the extent to which offences against a person can apply in an online environment.
This book is now a little dated, but it is well worth reading:
@manchestermelly @neil let me preface this with "not an LLM fan":
A colleague recently claimed that there's a huge difference in how well they work now vs. 6 months ago.
Therefore out of curiosity (and definitely not claiming that you're wrong!); how recent was your experience?
@damien With all due respect for your colleague:
My wife calls AI "Drunk Uncle as a Service"
Neil Brown
Stuart Gibson
Christopher M0YNG
jamesb (MW1CGG)
refraction 
Dr. Bruce Simpson
kabel42
Reinald Kirchner
Enola Knezevic
Su_G
surfsup31
VANTABlack
skua
Edward L Platt
Shadedlady
C++ Wage Slave
Impudent Strumpet
Robotistry
Andrew Deacon
Dr Ro Smith
2xfo
datordžeks 🇱🇻
Ivor Hewitt
Steve Hill 🏴🇪🇺
scmbradley
Melissa
decibyte
David Cantrell 🏏
Damien
Isocat
Linker3000
Poul-Henning Kamp
Blamellors