Writes Megan McArdle, in "Are we in an AI bubble that’s getting ready to pop? The promised AI revolution isn’t here yet. But it’s a smart bet that productivity gains will follow" (WaPo).
And this caught my eye: "A friend who is a lawyer... asked a chatbot to draft a document, and though the draft needed work, he estimated it had saved him two to four hours of typing. I asked him what he did with the extra time. He pleaded the fifth." Pleaded the Fifth, eh? That makes it sound as though he billed the client for the 2 to 4 hours it would have taken to do the work traditionally!
It's not just the "typing" that the AI did for him. It also composed material into solid standard English, wrote the citations in the required form, put the substance in some sort of order, and probably much more. It wasn't just "typing" he'd have been doing during those hours.
McArdle is using the word "typing" in a way that reminds me of Truman Capote's famous insult to Jack Kerouac: "That's not writing, that's typing." Oh, Jack wasn't just typing typing. He was typing typing.
पोस्ट्स
६७ टिप्पण्या:
By 2005 or so, it will become clear that the Internet’s impact on the economy has been no greater than the fax machine’s.' That was Nobel Prize winner Paul Krugman.... We’re now hearing similar questions about artificial intelligence...."
This is so stupid it hurts. Journalists are truly the dumbest professionals on the planet who aren't public educators.
Just because a lawyer fucked off for two hours in a stupid anecdote does not show there is no productivity gain from LLMs.
The stagnation of inflation-adjusted wages since 1975 or so seems to correlate pretty well with the widespread corporate adoption of computers.
The billing should be done in terawatts, terabytes, and land use as a measure of productivity.
Much of Trusts and Estates, Finance and Corporate law involves adapting boilerplate to individual cases. My 40 years of BigLaw practice suggests that AI could increase productivity significantly. (Virtually all of my finance work was fixed fee, so no billing fraud would be involved.)
I was recently switched from building C++ simulation models to using Matlab/Simulink. Matlab/Simulink is not particularly intuitive and it is made for Engineering majors rather than CS majors.
Normally I would have to be trained to use Simulink/Matlab for some period of time.
In the beginning I was working with the engineer who was moving to another project and taking up a lot of their time. But within a few weeks I found that the Conversational Agent we had access to explained things very well.
Additionally it is extremely good at finding bugs in code. The detailed explanations are very helpful and they are near instant. The old strategy of talking to the "rubber duck" has been replaced.
One thing about conversational agents that will be a massive boon for society is that they are extremely good as a teaching aid. I am running my kids through complete historical periods. This couldn't come at a better time. Public Schools and Universities have been consumed by parasites who are quickly turning into luddites.
The number of headlines where the numbers came in "unexpectedly" suggests there isn't a single competent economist in the United States.
It's possible that the lawyer's use of AI would be fine if he understood it was only a first draft. Back in the day when I was drafting legal documents or briefs I found it helpful to start with a rough draft--get something down on paper, and then edit and revise it. But there is a long way from rough draft to finished product. And as some lawyers have found to their dismay--and to an angry judge's ire those AI generated briefs contain significant legal errors. Like citing a case that is either nonexistent, or has a contrary holding. Letting the machine do it is a good way to risk your law license.
Twitter me this, and Facebook it, a Google electronic social diversions, breaking JournoLism to progress the blood pressure, and Yahoo!
My family had a legal issue that required a contract to be written and I prompted Grok with the details. It instantly produced a document covering all the various details perfectly. We then hired an attorney to write the contract (at some considerable expense), and he provided nearly the same contract. I wouldn't trust Grok with such an important task, but I'll just bet the lawyer used AI and then edited it a bit.
By the way, lawyers are still being sanctioned for trying to submit false - hallucinated - citations in their AI-written briefs that they falsely bill their clients for.
Of course, they're not being disbarred, despite perpetuating these frauds on the court.
They usually get small fines, which highly suggests that all lawyers are doing this knowing full-well that the penalties for getting caught are virtually non-existent and they can falsely bill the client anyway for the time.
Lawyers aren't people, in the traditional sense of that word. They are snakes.
FLC, do you ever feel worn down by all the anger and bile that seems to dominate your thoughts? You know that it is possible to express opinions without all the vituperation.
Just because a lawyer fucked off for two hours in a stupid anecdote does not show there is no productivity gain from LLMs.
-----------
McCardle has to add anecdotes as filler to the "column" AI generated for her.
Critical readers with old eyeballs that don't catch things keep going to "expert" opinion-makers like Meg. She reminds them of her children's generation, if they had ever brought home a girl to play with...
That epigram, issued by Robert Solow in 1987...
Back then, wasn't the "computer age" still finding its feet, mostly about finding equilibrium in hardware scale? There were major industry "shake-outs" in the transition from central to distributed network processing. Will the same be true of AI, which seems at this point to require centralized, energy-intensive processing?
Here's my summary that took 10 hours to type, I mean, write...
The 1980s witnessed significant upheaval in the mini-computer industry, largely driven by the rise of personal computers and changing market dynamics.
Here's a breakdown of the shakeouts:
1. Technological advancements
Mini to Micro: The advent of increasingly powerful microprocessors in the 1980s blurred the lines between mini and personal computers.
32-bit Processors: The emergence of 32-bit processors like the Motorola 68000 and Intel 80386 significantly boosted the performance of personal computers, allowing them to rival and even surpass the capabilities of many minicomputers, especially in the lower to mid-range.
RISC Architecture: The development of Reduced Instruction Set Computing (RISC) architectures offered even greater performance potential for personal computers, further challenging the minicomputer market.
2. Market pressures
Personal Computer Competition: The proliferation of personal computers, with their lower prices and increasing capabilities, began to erode the minicomputer market share.
Shift to Distributed Computing: The growing popularity of PC networks and distributed computing challenged the traditional centralized mini-computer model, according to CSUSB ScholarWorks.
Oversaturation and Price Wars: A surge in computer manufacturers in the mid-1980s led to an oversupply and intense price competition, impacting even major players.
3. Notable companies and their struggles
DEC (Digital Equipment Corporation): Once the second-largest computer manufacturer globally, DEC struggled to adapt to the changing landscape, particularly their continued reliance on proprietary systems (like their VAX machines and VMS operating system) while the industry shifted towards open architectures and commodity hardware, according to FedTech Magazine.
IBM: Even IBM, while initially successful with its personal computer, faced challenges as the market became saturated and PC clone manufacturers chipped away at its market dominance and profitability.
Other Minicomputer Companies: Companies like Data General, Wang, Prime, Computervision, and Honeywell either failed, merged, or were acquired as the mini-computer market dwindled.
In essence, the 1980s shakeout in the mini-computer industry was a multifaceted event driven by rapid technological advancements, intense market competition from personal computers, and a failure by many established companies to adapt to the new computing paradigm.
well lawyers don't need computers to lie, they do seem to be unincumbered with the truth, lawyers i mean,
At the Eighth Circuit Judicial Conference in KC last month, I saw a lawyer draft a plausible cert petition in ten minutes.
Justice Kavanaugh saw the demo and he assured us that SCOTUS doesn’t use AI.
I'd argue that if it was able to *actually* replace engineers then it would be so capable that it's achieved AGI and it would be worth enormous amount yes. That doesn't mean that the technology if it happens isn't just a commodity though. The tech industry has become used to building tech which has deep moats and extracts enormous rents.
I don't see any reason to think AI is or will follow this pattern. I flip between the different providers routinely, and if DeepSeek is being honest there's no reason to think I won't be self hosting it myself in a few years.
I also haven't seen any evidence that it's actually close to AGI. Those reasoning models still don't reason in a human manner and still don't have any understanding of context and real things. It doesn't mean they can't just I haven't seen any evidence that they are.
FLC:
A MS federal judge had to withdraw an opinion as his law clerks used AI and it was riddled with errors.
FLC, do you ever feel worn down by all the anger and bile ...
Found the shyster.
A computer engineer of mature years and great experience whom I know recently introduced me to a new type of engineering - "prompt engineering." Apparently some schools are already offering this specialty as either a degree or a cert of some kind. Basically the training is in how to compose prompts for AI that will result in the most accurate and fastest result.
He said he and his people do that themselves because they do have years of experience that the young folk don't, but he is less concerned than many about the effects of AI on jobs. The jobs, he believes, will just evolve. Even though AI seems unprecedented, I tend to agree with him, based on the history of technological advance. After all, I'll bet the plow seemed unprecedented too.
Lawyers billing hours for work done by AI in minutes. Another reason to hate lawyers.
↑ This is the issue. The industry is spending >$100bn/year on something which makes you slightly more efficient, without being able to charge much for the service due to competition.
I think psychologically the VC scene and adjacent are just completely used to land grabbing market share and haven't critically evaluated if the rentier model is actually possible with this kind of technology.
All of the big innovations and technical developments in the last 15 years or so (that I can think of) have led to winner-takes-all, or takes most, with few competitors. Its plain as day in the books and thinking of Peter Thiel, Marc Andreessen etc.
I'm very happy though! I get super useful tools and technology and don't imagine I'll ever be paying much beyond cost + minimum margin for it, all funded by those VC philanthropists. 😉
Begley noted: "A MS federal judge had to withdraw an opinion as his law clerks used AI and it was riddled with errors."
Yes, I'm aware of that one.
The thing to note is there are no negative consequences to this judge. In fact, he was allowed to remove the faked opinion from the public docket and replace it after having gotten caught. The opinion cited cases that don't exist. Quoted plaintiffs who were imaginary and misquoted the laws it was relying on.
The judge hasn't been fined, or sanctioned by the bar, or suffered any other negative consequences. He still sits on the court in the Southern District of Mississippi. This highly suggests that they are all doing it and that they know there won't be any negative consequences for getting caught doing it.
This phenomenon of no-consequences might have led to a race to the bottom. But lawyers are already on the bottom. Feeding.
I subscribe to 3 different AIs. Although I do not ask them to draft documents for me, I occasionally will ask Chat or Claude to draft a novel clause for me. I receive documents drafted by AI, from counterparties in the transactional part of my practice. They come mostly from non-attorneys in startups trying to save fees. Generally, they are a good first draft but have holes.
Where I do use AI frequently is in legal analysis. Sometimes it is to confirm my own judgement on issues I handle very infrequently, or which are far outside my core practice. It does an excellent job in most cases, because I have learned to give it a long prompt that provides a good deal of context about the issue at hand, why I am asking, and what format I want the answer in. I always ask for citations (which I always check) and have learned to ask for arguments for and against its conclusions. The latter is quite important, and has benefits in addition to avoiding a response that is solicitous to the point of being wrong.
I also use it for one other common legal task, for which I have hired a programmer to work with me to create a product that my clients can use when I finally retire, and which I hope has modest commercial potential. We are two months into the project and I'll share more later.
If I hadn’t retired, I would definitely use AI to write the first draft of patent applications. If it could prepare reasonable drafts I would likely get rid of the junior attorney in my department.
I worked in house, so there were no hours billed. AI would do the scut work allowing me to focus on more challenging tasks like freedom-to-operate analyses or determinations whether competitors were infringing our IP.
The computer saved me a lot of time at my job. But it didn't make me more productive. It allowed me to spend less time on mundane tasks and focus on more important things. Emails meant quicker and more efficent communitcation but not less communication.
AI will mean office workers will be able to spend less time drafting memos and reports. But they will just be spending more spending time on other things.
I can see AI drastically reducing the need for tax returns preparers and auditors. The AI could check all the numbers, fill out the tax return, and all the preparer would have to do is make sure it was correct. Same with audits of financial statements.
Judging by their terrible opinions, I suspect may US District Judges are already using AI. Sometimes I read them, and its just page after page of bullshit.
I don't think that anyone can seriously argue that the PC hasn't improved productivity. What is the metric?
And the internet's benefits aren't limited to productivity--spreading of knowledge and (at least theoretically) freedom of speech must count for something.
AI is not a panacea, but it does a lot of things a lot faster than we humans.
And Krugman is just wrong as usual.
In 2009 we were checking complicated chip designs automatically for sanity - namely things called liveness (chip never hangs up) and safety (certain things cannot ever happen). A task well beyond human capability.
The legal marketplace will fairly quickly cut out the ability of lawyers to get away with charging phantom hours for AI product. Because other lawyers are able to provide the same product and charge less. It's how competitive markets work. Many legal clients are not stupid and can figure this out. The clients pay the bills.
"Productive lawyers" is an oxymoron. In my fifty years of experience as an engineer, systems analyst and executive I can't think of a single significant productivity improvement that was made by a lawyer to any business with which I was associated. On the contrary, I have found lawyers to be impediments to progress so that anything that makes their work faster or easier is counterproductive to real world manufacturing and commerce. Their only useful contributions seemed to be countering the obstructionism of other lawyers. That's why the old joke about calling a busload of lawyers going over a cliff "a good start" is actually not a joke.
You may be able to get AI to write your brief in seconds, but it will still take hours to make sure that everything was right.
±
Ken Olson, the head of Digital Equipment said in 1977: "There is no reason for any individual to have a computer in their home." That accounts for Digital, a leading computer manufacturer in the Seventies, passing up the opportunity to enter the personal computer market and eventually going out of business. Some have defended Olson's comment, pointing out that the real big computers that we are interacting with aren't in our homes, but far away, and that we could make do with a terminal with minimal power and memory, but he was still wrong.
±
But what about the predictions of the leaders of the personal computer revolution that the machines would make us free and creative? Are those predictions coming true? Or are we more controlled, more surveilled, and more mindless?
There are plenty of discrete tasks that AI can do for lawyers, such as summarizing deposition transcripts and the like (which is optimally the work of a paralegal). AI is currently not to be solely trusted for legal research but could productively be used in conjunction with Westlaw, for example.
What will it mean when all "serious" work is written and consumed by AI, while the humans spend all of their time watching AI-generated cat videos? I have a feeling we're in for a very bumpy ride.
FLC:
I know from personal experience that the Nebraska federal judges are NOT using AI.
Most of these predictors don't really see the real impact of the technology. The electronics revolution was a bit slower, but in the last 40 years it has done a lot in the background, in the infrastructure, such as replacing motor controllers with far more capable programmable controllers. The large cabinets are still there, but the guts inside are tiny now and could be giving more efficient life to a motor wound a century ago.
Similarly, by 1987, the Programmable Logic Controller (PLC), the primary inventor refused to allow it to be called a computer, was just shy of 20 years old and was running most manufacturing lines replacing relay logic, at first in the auto industry.
The first half of the 20th century saw massive changes that were readily apparent in the home as technology transformed the US. But then the changes started happening inside the appliances or in out of sight infrastructure controllers.
An example of visible efficiencies that were seen.
"Our idea of roughing it is to walk to the end of the office away from the electric fan now and then, just to be able to appreciate modern comforts." The Decatur Herald, Illinois, July 20, 1935
Do you even know the move to ECM motors on fans and compressors with electronic expansion valves that have made your HVAC more efficient over the last 20 years?
It's not just doing things more efficiently, it's often doing them at all. I personally have accomplished tasks thanks to AI's help that I just wouldn't even have started. AI told me how it could be done and gave me enough info to make it suddenly doable, especially without all the old-style research time which might have failed due to fatigue or difficulty.
"Their only useful contributions seemed to be countering the obstructionism of other lawyers."
I assume there are exceptions, but in my 4 decades in business, this has been frustratingly true, and it makes the profession on balance a substantial negative . Even with something like a patent to protect your ideas, a lawyer is likely involved in the attempted theft, and without them, the writing of the patent would be within the ability of the average inventor.
The real reason why productivity hasnt increased is likely because the way we measure output of workers has changed drastically when you consider that. So much of our economy is now in in areas where there is no output that can be measured directly like healthcare
I think all first year engineering students should be required to learn how to use slide rules.
My daughter (ChE 98) and grandaughter (ChE 27) disagree. Granddaughter's roommate visiting from Notre Dame had never even heard of a slide rule.
Nuclear reactors requre a LOT of math to operate. In the 50s and 60s that math was done by HS grads using sliderules, pencil and paper. I spent 4 weeks in a navy prep school, 44 hours a week in classroom, relearning K-12 math and how to use a slide rule.
Most people plug some numbers into a computer, a cash register or excel and accept the result without question.
Using a slide rule we were taught to first guess the expected answer and write it down. Then we went step by step calculating the final answer writing everything down.
The thing with a slide rule is that you can't get an answer without understanding the process and and without having an idea of the answer before you start. Obviously the guess was seldom correct but it was in the very large ballpark. And if calculated answer and guess didn't align, we were expected to go back and figure out why. "Bad guess" was an acceptable answer but expected to be rare.
I still have my K&E slide rule from 1968, I tried to give it to my daughter in 93 when she went off to eng school. I told her $200 was too much to spend on a calculator when a slide rule would do all that did and the batteries would never run out. The case even has a handy belt clip. She looked at me like I had two heads. My argument that she would be the coolest kid in school for being the only one with a slide rule got me "The Look" and "Daaaad..." I bought her the fancy-schmancy calculator.
GPS is similar. Lots of stories about people getting into trouble because they blindly followed GPS directions. I love my GPS and rely on it to get me where I am going. But I've lost the signal and been completely lost. GPS tells me where to go but tells me nothing about where I am other than meatspace position. I always like to have a roadmap with me.
John Henry
It's much easier to measure tangible things like cars and airplanes. But how do you measure output like healthcare outcomes? All of these service industries are not showing up in the statistics in a way that the traditional economists can understand.
"I don't think that anyone can seriously argue that the PC hasn't improved productivity."
I know the Krugman quote was from years ago. But PK is the Jim Cramer of economists.
If the lawyer spent 1 hour using AI then 3 hours watching cat videos instead of 4 hours making the document, they have still increased their productivity.
We may not think they have made good use of the increase but is is way more productive to take 1 hour instead of 4.
John Henry
JK Brown
Great point about the PLC and how much better it has made machines.
I talk about PLCs and their use in packaging machinery in my Packaging Machinery Basics class (Next class Oct 8 in chicago, www.iopp.org for deets)
One of the slides I use is of a pouch forming machine. Think frozen veggies, or soup powder or the like. I have a picture of the control cabinet from the 70s. It is the size of a phone booth with 50-75 relays, timers, switches and so on. Also miles of wire and hundreds, maybe thousands of connections. I also have a pic of the control cabinet, same builder, same machine, same model. Only now it has a much smaller control cabinet because instead of all those electromechanical components it has a PLC, about 4"x6"x4" that does ALL those functions.
On the old machine it might take hours or days to figure out an electrical problem. If it could not be fixed inhouse, you might have to fly a factory tech in.
With the PLC, hook up a laptop and run a diagnostic program. Or connect to the tech in the factory. In a worst case replace the whole PC. Much less to break, what can break does so much less frequently and when it does is much easire to find and fix.
John Henry
It is 100% for certain a bubble in the AI related stocks but you never know what will burst it or when. The areas where it could, right now, offer enormous productivity improvements are exactly those areas where the lobbyists will prevent it from being used- the law, medicine, and education.
John Henry I had similar experience in Civil Engineering. (I kept my slide rule too. When we hired a young engineer tradition was to send them to old JAORE to show them what a slide rule was.) Professor Lee once told me he feared Garbage I, GOSPEL out. And the way he suggested we combat that was, as you note, having a grasp on what the result should be.
In my very first structures class an exam question was calculate the deflection of a truss structure under a point load. Keeping track of decimal points was critical with a slip-stick. I didn't. I proudly awaited my good grade. I got a zero for that problem. I'd put down a 3 foot deflection when the actual answer was 0.03 feet. I went to Professor Rolf and said I know I busted the decimal point and I knew WHERE I busted it. But I'd used the correct formula and otherwise applied it correctly. The one and only time I asked for partial credit.
Professor Rolf then gave me, perhaps, the greatest single "lesson" I received in engineering school. He said that anyone who calculated a 3 foot deflection in a bridge span and did not immediately ask, "Why is this wrong" should not be an engineer.
Mercifully I learned a bit and aced the class.
And the companies running these businesses will be forced to internalize the electrical power requirements- it is a political hot potato to expect voters to live with bearing the costs of electricity.
"That makes it sound as though he billed the client for the 2 to 4 hours it would have taken to do the work traditionally!"
This is a pretty common practice in a lot orf places. For example: At a car dealership, they have a list of how many hours any given task should be billed for. If the list says that the customer should get billed for 3 hours of work, the customer gets billed for 3 hours of work even if the work only took ten minutes.
Yancy,
Some of these companies are talking about building their own small nukes.
I've not looked into it but it would seem that self-generating using gas turbines with heat recovery (cogeneration) for heating and cooling or combined cycle, with heat recovery would be worthwhile too.
Cogen plant can get up to 75-80% efficiency if the waste heat can be used.
I think the REAL problem is forecasting. Spending all than money on a generating capacity makes sense if the need will last 20-25 years. But what if AI is a bubble? Who is going to build out the extra capacity if they can't be sure it won't be sitting idle in 5 years?
Or why build extra capacity is space based solar is coming in 5-15 years?
John Henry
If I hire a carpenter to build me a gazebo, should I pay him more if he uses a handsaw to cut the studs?
Or less if he uses a power saw?
John Henry
I am in agreement with John Henry for the most part...
I am with you regarding the needed to understand and grasp the basics, that is just good pedagogy. The slide rule itself is nothing more than a tool that enforces that idea, I would not say it is an indispensable part of that process. Keeping with the Navy theme, they emphasize that process very well with Radcon math (colloquialism for using simple tricks to arrive at a reasonable estimate using mental arithmetic), qualitative analysis (given this chain of equations, of this variable changes due to some physical process, here is how everything else changes), etc...
I used Matlab as your slide rule equivalent in my undergrad and graduate courses by developing little tools to perform basic functions of more complex algorithms rather than immediately using built-in libraries and skipping the entire learning process. The downside is that I can't whip it out at parties to impress chicks like you can with your slide rule. I forget where I first heard this idea, but the end goal should be that you can get to a point where the equations speak to you. Whether you are talking about astrodynamics or thermodynamics, if you can't take an equation and tease out the physical implications it is showing by just looking at it briefly, then you don't have that important and necessary grasp of the fundamentals
AI is a fantastic tool when used appropriately. As a non-SME, but with a heavy engineering background, I have been able to use LLM assistance to develop reasonably complex code to drive a telescope mount that is spread across C, C++, C#, and Python in various portions of the implementation in a fraction of the time I would have otherwise been able to. It is very clear when I have ventured into inappropriate usage when I find myself having spent the last 30 minutes cursing at my monitor because I am not being provided the artifact I thought I had been asking for.
100% agree that it generates rough initial drafts, however you can also use it to revise those drafts using recursive techniques. In the end, you still need s human eye to verify the final things. Where people get in trouble is when they don't follow the rigorous process. No different than not using a star pattern to tighten bolts and finding out the gasket is not properly sealing.
Who should get the benefit of extra productivity is always an interesting question.
Let's say I own a machine shop and Bob, a competent machinist, runs a typical manual lathe for $30/hour
I go out spend a hundred thousand on a CNC automated lathe. Spend more money installing it and even more money training bob how to use it.
Now, instead of 10 parts per hour, bob's productivy has doubled and he produces 20 parts per hour.
How much extra should Bob be paid for his increased productivity?
Should be be paid anything extra?
John Henry
In my senior year in H.S. I took the last class offered in slide rule. You were allowed to use a calculator to check your work if you could afford one. I had one, and wrote cheat notes on the back of the calculator on how to use a slide rule. A pretty strong set of signals sent that the slide rule was not the future, but I do find it almost magical that they work so well.
"How much extra should Bob be paid for his increased productivity?
Replace Bob with a robot, teach Bob to program the robot. Get quadruple productivity including making parts in the dark. Sell Bob the business and go fishing.
FormerLawClerk said...
The number of headlines where the numbers came in "unexpectedly" suggests there isn't a single competent economist in the United States.
Economists are the Academics of the business world.
They are the lowest 10% of business majors and fail at anything real that they try to do so they get jobs in think tanks, Universities, and media outlets.
John henry said...
Who should get the benefit of extra productivity is always an interesting question.
Let's say I own a machine shop and Bob, a competent machinist, runs a typical manual lathe for $30/hour
I go out spend a hundred thousand on a CNC automated lathe. Spend more money installing it and even more money training bob how to use it.
Now, instead of 10 parts per hour, bob's productivy has doubled and he produces 20 parts per hour.
How much extra should Bob be paid for his increased productivity?
Should be be paid anything extra?
John Henry
2 Components: The answer will come down to how expensive it is to replace Bob and what additional value Bob adds. This will be a pure supply and demand question on one side and extra value Bob adds to the other.
Any business owner will tell you that a Good Employee is worth their weight in gold. If they show up, work hard, don't take more than necessary amounts of your time they are worth a lot to a company. They are very much like good renters. This is the extra value Bob can add.
On the other side you need to judge how valuable the skill Bob has is by how many people like Bob you need and and many people like Bob there are.
What AI is doing is making Bob much less rare and it is making finding dependable consistent labor much easier. Robots don't call in sick on Friday to go fishing.
Lot's to go over on this one Ms. Althouse,
But right off the bat, it reminded me of someone's observation a few decades ago, that the most notable skill of an economist is "that they are very good at explaining why they were wrong".
I personally have accomplished tasks thanks to AI's help that I just wouldn't even have started.
I think this qualifies. My husband's grandmother passed away at least ten years ago; her daughter, my husband's mother, spent a good amount of time with her as she was failing, asking her to tell stories about her past. My mother-in-law recorded her on a microrecorder - about 8 little tapes.
My MIL then tried to transcribe those tapes. But she's not a fast typist and she wasn't very good with the sheer listening part. So she asked me to take a whack at it. I did - that is, I tried. There was so much white noise on the tapes that I could barely make out voices.
Fast forward maybe five years: she rebooted the project by asking if I could take another stab. I got those tapes transferred to digital media, with as much cleanup as the guy could do, and tried again. It was just... so... tedious. I'm a decently fast typist, but even with the cleanup I had to go back time and again to ensure that I was getting the transcription correct.
And fast forward again, to this spring. My MIL was about to turn 80, and my husband really wanted to get this transaction done for her. So, with an hour or so of research, he found an app that would auto-transcribe the digital files, complete with time stamps. Et voilá, the project was 90% done in the space of about half an hour - just editing remained.
A project that was, at first, not possible became simply tiresome and time-consuming, and finally easy, because of the march of technology.
…when your go-to economist is a NYT commentator the problem is with you…go read some Hayek or Adam Smith ffs…
“If I hadn’t retired, I would definitely use AI to write the first draft of patent applications. If it could prepare reasonable drafts I would likely get rid of the junior attorney in my department.”
But then, how do you train the next generation of patent attorneys? Claim drafting, in particular. Of course, I saw plenty of bad claim drafting. A lot of mediocre patent attorneys out there. But my view is that it is an art.
So, my thoughts are that a Background section is a prime candidate for AI. I used to unquote Wikipedia articles a lot. You knew that you weren’t accidentally admitting something was prior art. I expect that AI could do a better job there than I ever could. Maybe a first draft of the Detailed Description and drawings.
Let the AI take a first whack at the claims, if they lend themselves to that. My brother, an MechE, did a lot of connectivity claims. I did some connectivity claims in Circuits, but those were being phased out, as being narrow and impossible to enforce, as the scale of integration went into the billions of transistors on an IC. I tended to work in Systems and Software, where things were, maybe, fuzzier. The problem there, a lot of times, was which boxes to claim, in which order (in dependent claims) in order to avoid esp § 102 rejections. While trying to predict future technology and infringement/work arounds. Then sometimes set up for a family of patent applications sharing some/all of the Spec and Drawings.
In any case, after getting the Claims in decent shape, I would want the AI to make sure that everything in the Claims was fully supported in the Spec and Drawings, and they matched. That’s not that hard - I did much of that with VB macros, comparing Ref and Fig numbers in both. The Abstract then is easy, based on the broader claims.
I had an Indian sounding guy calling me up a couple days ago for, I think, some patent work. Could have been some sort of scam. Told him I was retired and hung up. Don’t know where he got my name. Maybe the USPTO OED Directory. Can’t afford the malpractice insurance to even think of doing any patent work anymore. My brother, with some long term clients (I was trained on one of his clients’ inventions 35 years ago)), doesn’t do much more than cover his malpractice insurance these days - he keeps his hands in to prevent boredom. I wouldn’t have minded, but my last job was with a large general practice firm, where those attys controlled the patent work. This year, he beat our father in how old he could still practice law.
Yep. Using AI to eliminate entry-level work may taste like efficiency right now… but long-term it’s just eating your seed corn. Future experts need to have the opportunity to learn how to think and do.
Using AI just yesterday I helped an elderly friend identify resources to (a) help him figure out where he could get the cheapest "season tickets: for watching college basketball and football on TV, and
(b) identify the organizations and contact numbers in his area helping seniors get appropriate in-home support and care .
He didn't know where to start, on either topic. It took me five minutes per AI conversation to find exactly what he was looking for. THAT's productivity!
Solow kind of became the Krugman of his era. It wasn’t the rock star propaganda position it became with Krugman. A more civilized era…
Second rate economist here, but measuring productivity isn't as difficult as some are claiming. Productivity is simply the value of total output divided by the sum of labor inputs. Both are measured as easily with healthcare and other service industries as with manufacturing or agriculture. It is only difficult if there are not market prices, as occurs with much of what government does.
Now have AI design a heat exchanger for a refinery.
Worked in IT and systems integration for a few decades. Implemented lots of systems to replace old systems. Swapping new for old but keeping processes mostly unchanged usually resulted in single digit percentage increases in productivity. Changing processes to take full advantage of the new technology typically changed productivity by 10’s or 100’s of percent.
Simple example. Worked with a Dermatologist as they started using tablets in the exam rooms to create the digital medical record of the visit. Problem was all of the words they had to transcribe to accurately describe the condition. Problem went away when I asked why they didn’t just take a picture and include it in the electronic medical record.
Mostly we haven’t figured out how to take full advantage of AI yet.
Economists need to focus more on incentives and less on convex hulls.
टिप्पणी पोस्ट करा
Please use the comments forum to respond to the post. Don't fight with each other. Be substantive... or interesting... or funny. Comments should go up immediately... unless you're commenting on a post older than 2 days. Then you have to wait for us to moderate you through. It's also possible to get shunted into spam by the machine. We try to keep an eye on that and release the miscaught good stuff. We do delete some comments, but not for viewpoint... for bad faith.