Are you better value for money than AI? AI vendors are starting to say the quiet part out loud. As technology advances, it seems more about controlling costs and headcount. OpenAI chief financial officer (CFO) Sarah Friar was reported discussing the possibility of pricier tiers for the company's services and justifying four-figure tiers – quite a jump from the $200 …
Artificial Intelligence does currently exist, as that is computer programs that can carry out a specific task far more accurately and far faster than a human, but that has already existed for decades.
LLMs are rather artificial stupidity - while they can parrot stupidity far faster than a human ever could, they still cannot come up with anything more stupid than what a human has written previously.
I mean, the term itself was essentially coined by John McCarthy in the 1955 Dartmouth workshop, and basically was a marketing buzzword so that the Rockefellers and eventually the US military would buy the pitch. You couldn't sell a product if you had to rely on such nerdy terms like “automata theory” or “symbolic computation”. Cybernetics, maybe, but McCarthy didn't want to spend time deferring to Norbert Wiener, of all people.
I mean, it was always chosen for the vibes. The field itself does not — and cannot — be defined with any rigor.
Expert Systems - aka XPS - were quietly dissolved into online diagnostic systems and then mainly devolved into Vaguely Trained Systems as people couldn't be bothered actually working out what all that Bayesian nonsense is about. There are a few "proper" XPS around, diagnosing Important Things that Cost Money, but the term itself seems to be rarely heard[1]. Shame.
As for 4GLs - aside from the utterly, utterly stupid name[3], these were no more interesting than dBase - literally, just whatever "simple to use database language" someone wanted to flog you. What happened to them? Well, Sqlite allows everyone access[4] to the query language that actually *works* (not pretty, or easy, but in the end we all come back to it).
[1] which just means that people aren't looking up the info about old systems, MYCIN et al, and all the things we learnt from them. As we continually complain happens with - every other tech area, from basic coding onwards [2]
[2] lawn, off, now
[3] Sod what Wikipedia[5] says, the flurry of books and seminars etc about "4GL", spelt that way, in the 1980s was due solely to trying to jump on the hype wagon from the Japanese "Faith Generation Computer" research - this was for the fifth generation of *hardware*, not software: the intent was to build hardware that could run logic programming languages really fast, such as Prolog. Giving the slogan "Prolog, the Language of the Fifth Generation". Promptly misread by the flacks, who tried to find out what Prolog does ("oh, it uses a database of logical relations and resolves those to answer a query") - not understanding what a Horne Clause is nor what the Resolution algorithm actually does, they just heard "database". Ooh, we've got a database language but we can't claim it is Prolog, so, well, Prolog is a 5GL, what we have must be a 4GL! In the words of the great Welsh hard, "I know, because I was there!" and we pissed ourselves laughing at first then got pissed off trying to fend away the drivel merchants into the '90s. At which point, that AI bubble had burst and it was all very embarrassing, never mention it again.
[4] no, not Access, but I bet you can find some advert calling Access a 4GL
[5] that article is an incredible exercise in trying to force one specific viewpoint as "The One True Reality", just to boost the claim that the "4GL"s were actually anything interesting. The naming of generations of software languages is a bit convoluted, but back when we were studying languages, in the late 70s and early 80s before the "4GL" bubble, we were well past gen 4 languages, counting along multiple paths. For example, (1) machine code (2) assembler (3) macro assembler (4) simple compiled languages, e.g. FORTRAN (5) Functions as First Class Types, e.g LISP 1.2 (6) Object Oriented Languages, e.g. Simula-67. Which gets us all the way up to 1967, so from there to languages like Algol-W or even, gosh, Prolog and it's friends (let alone weighted logic languages, Fuzzy Logic languages...) we are *way* past "generation 4".[6]
[6] lookie here: if we follow *this* trail in the development of PLs, we find someone who came up with a novel way of doing things, so clearly this was the fourth generation of languages. But he was restrained by the filing system on the computer, which only allowed short file names, so he had to call the new language "FORTH".
Speak not truth or substance because it will just garner you a solitary down vote (at the time I read it anyway).
As one who has a passing acquaintance with expert systems, logic languages and what-have-you back in the day (i.e. back before Marketing discovered it) I reckon you've just about summed things up perfectly. Even to that bit about FORTH. Unfortunately there's never really been a study on the Herd Instinct in Programming (aka "How the likes of Microsoft made so much money") so the only course that seemed to work is to step aside and find a niche which is unfashionable (but lucrative) to lurk in.
The hype around 4th generation languages (counting them as "ones and zeros", "assembly language", "FORTRAN et al. compiled or interpreted", and 'whatever a 4GL is") was that they were supposed to be usable by 'ordinary people". I.e. managers and secretaries (instead of employing those weird "programmers").
And in some cases they sort of, kind of, were. For example, SQL can be used at a much higher level than previous means of storing and accessing large amounts of data.
However, they still need training to use. Or you end up with cr*p systems. See pretty much every set of SQL queries designed by a manager.
Which people eventually learned. So the hype died down, and they're on to the next thing.
Which now seems to be AI.
"Artificial Intelligence does currently exist, as that is computer programs that can carry out a specific task far more accurately and far faster than a human, but that has already existed for decades"
This is not ANY definition of AI I have been able to find on the 'interWebs' by anyone of recognized authority !!!
So I will ask ... are you sure ???
:)
If a [cough][cough] so called AI system in 2024/2025 can think for itself and then follow Asimov's laws of robotics then it is AI.
The short answer is that today, NO NO NO NO NO NO NO NO and NO.
AI is pure marketing hype.
Until an AI system can think like us, act like us and as Philip J Dick said "Do Androids dream of Electric Sheep" https://en.wikipedia.org/wiki/Do_Androids_Dream_of_Electric_Sheep
then I would regard anyone preaching AI as a Snake Oil Salesman (go to DC on 20th Jan 2025 and you will see their leader in chief and his cult out in force)
Current systems labelled AI are nothing more than glorified DSS. The regurgitate whatever garbage has been fed into them. GIGO rules!
I'll just remind the audience of the various lawyers who relied upon AI to write their legal briefs. The AI went off and happily made up citations and cases that never existed. Hilarity ensued when the judges discovered that and asked the lawyers for an explanation. Much backpedalling and apologetic noises were the result and several lawyers barely escaped with their licences intact.
Now, imagine if AI wrote the manuals for the operation of a nuclear power plant. Or, perhaps, prescriptions for medication. I suspect it will only be a matter of time before HR AI bots write a job description or Help Wanted posting that somehow violates the law...maybe explicitly saying "no old people* need apply"? And heaven help the AI bot who fires someone for an illegal reason (of course, the bot will suffer no consequences, nor will its overlords)
*insert your favourite protected class
I'll just remind the audience of the various lawyers who relied upon AI to write their legal briefs. The AI went off and happily made up citations and cases that never existed. Hilarity ensued when the judges discovered that and asked the lawyers for an explanation. Much backpedalling and apologetic noises were the result and several lawyers barely escaped with their licences intact.
The irony of your post is that the made up citation pointed to a specific peddler of legal data/software that has last year released its very own AI that it claims is "free of hallucinations". I make no such claim, of course.
Lout's law of AI states that "AI will take your job when someone above you in the hierarchy can be sold the idea that it can". Whether or not it can actually perform your role will not be a salient fact involved in the decision making process.
I think you're somewhat missing the point...AI might not be able to fix a paper jam (that's a problem that has arisen through human engineering) but it probably can design an unjammable printer.
Why not try it? Pit yourself against the best bleeding edge AI currently available and try and beat whatever jamless design it comes up with.
Whilst I agree that LLMs are not the full package...there has been AI technology around for quite a while that is objectively better than a human for the task it was built for...computer vision, for example, massively outperforms humans at pretty much everything (except maybe facial recognition).
For the most part, if an AI doesn't live up to it's promises...it's not because of the AI...it's because of the data it was trained on.
I've built a few AI based solutions over the years and all but one of them outperforms a human at the task it was designed for.
The main problem we've had for a long time is the interface between AI solutions and humans...LLMs kinda fill that gap and that is where the concern is I think...LLMs aren't by themselves an "AI Solution" but they do open the door to other AI solutions (that have been around for a long time) to be more accessible.
The main advantage that humans currently have over AI is that a human is capable of learning multiple skills and using them tandem, which makes for some very skilled humans (they aren't the norm though), the average human though does not possess a specialist set of skills that can be combined into a unique package...this is the benchmark that AI should be compared against and that is the benchmark that should be used to determine if AI has "arrived".
For example...can "Dave" in the warehouse pick up an AI tool and become more than "Dave in the Warehouse"?
That's the test...let's not forget, back in the stone age, building a house and making sticks with sharp stones on the end was the cutting edge...it was the "brain surgery" of it's time...our nerdy ancestors were burning sticks, making things sharp and figuring out materials in order to give "Dave in the Watchtower" the ability to do things he previously couldn't...like going out and killing loads of animals to keep everyone fed, making him look scary to scare off rival tribes, giving him sharp stones to cut trees down etc etc...
"Dave" has always been the one that we've sought to improve for the benefit of everyone. He's not the sharpest tool in the box, but he's big and dumb enough that if you put the right tool in his hands he's a capable unit...if AI tools can improve Dave some more, then we have a potential, revolutionary, shift on our hands.
Going back to the stone age, for a while, the forward scout was an important dude...he'd go off following the weather looking for nice caves to move to...as soon as the nerds put tools and fire in Daves hands, the scout was a nobody because we didn't need someone to constantly be out there finding new caves...I'm sure at the time, the Union of Cave Finders and Weather Followers were up in arms when the legions of Daves out there could start building houses...I bet they had all sorts of reasons for doubting Dave and his new found ability to build shit using tools the nerds had invented....the village can't move, rival tribes will know exactly where we are all the time, what about the seasons? It get's cold in winter...etc etc...
Anyway, my point is, nerds inventing things and technological progress has never been about creating or equaling specialists. It's always been about removing the abilities to Dave, to free ourselves up to invent more shit while he's off building, hunting, soldiering etc etc.
Dave, bless his daft face, is going to have new tools in the future and he's going to be able to do things that scare specialists and cause them to kick off...but it's all good...because big daft Dave, since the beginning of time, has always sought to look after his tribe using whatever is at his disposal...specialists since the beginning of time, have always sought to protect themselves and what they do...they're specialists, it's all they have.
Specialists come and go...but Dave is eternal...we need to keep giving him ever pointier sticks and hotter burning fire so that when we send him off to do the dangerous shit with that vacant confused look on his face, we know he'll be alright and get the job done...even if we get it wrong occasionally, it's all good...there are plenty of Daves, we can use the feedback from the ones that survive and better equip the other Daves.
Eventually Dave is going to be doing your brain surgery...he won't understand it...but the tools he has will enable him to do it without having to understand it.
In the future he'll be doing things at 3am that we can only dream of...like swapping the nuclear fuel rods in the basement because the hot water is out again and his missus is chewing his ear off. Or nipping to "Spoons on the Moon" for the darts tournament he's looking forward to after a hard week at his orbital warehouse maintaining space drones that have been mining asteroids...as a nerd, I want to avoid all that sort of shit, which is why I want to make sure Dave can do it...I want to be on Earth making shit for the Army of Intergalactic Daves who are planning an assault on Stroggos to stop them coming here...that's dangerous man...let's leave it to Dave, Baz and Al to fight the hordes, sit on space stations and work with fuel rods etc...It's why they've always existed.
We're the guys building and inventing machines...Dave is the guy sticking his head in to inspect it. When we finally invent Gundams and Space Mining kit...it won't be some middle class twat with a PhD in Ancient Alien Poetry & Literature up there defending us and bringing rocks back...It'll be Dave.
The PhDs will be down here with the engineers.
Engineer: Oi, PhD, what does this say?
PhD: Interesting, it seems to be an ancient poem about a great machine and vast destruction.
Engineer: Awesome, can you derive a schematic?
PhD: Sure here...this work does raise some interesting philosophical points and has quite a whimsical...
Engineer: Yeah great, I don't give a fuck about that, I'll just take the schematic...you can fuck off now...go and listen to some Martian Opera or something until I need you again...Here Dave, put this on, it's a bit radioactive, but I've added a DeWalt colour scheme to it and painted the words "MAX" and "XL" on it for you...give it a whirl.
Dave: Fucking hell, nice.
No, Mr (or Ms) AC. Why don't you?
"computer vision, for example, massively outperforms humans at pretty much everything (except maybe facial recognition)."
Oh yes. The "It's brilliant at this task (except when it's not)" argument. Especially amusing when the "its not" case is the one most people actually want it to do.
<Long rant which basically follows the trajectory of the human race in CM Kornbluth's "The Little Black Bag". TL:DR Dark tinged techno utopian)>
"I've built a few AI based solutions over the years and all but one of them outperforms a human at the task it was designed for."
And made a healthy profit, even on the one that didn't work out?
I guess cognitive dissonance is for other people, right?
TBH I smell a snake oil salesman who's been drinking their own product.
Well someone drank the AI Kool Aid.
I quote this from the possibly LLM generated verbal diarrhoea above
The main problem we've had for a long time is the interface between AI solutions and humans...LLMs kinda fill that gap
Between AI solutions and Humans? How can there be anything between AI Solutions (that do not exist) and humans (who definitely do) and therefore how can LLM's (which are sold a s AI fill a gap between two things then one does not exist?
LLMs would have value if you did not need to spend all the time you saved using them to check that they have given an accurate answer. All they do is add another layer to the problem solving and therefore add another opportunity for errors
My first thought when reading the article was "I'd like to see an 'AI' replace printer rollers. Or an industrial controller power supply." Etc. Lots of hands-on stuff we do. Then there's work that requires actual intelligence, like writing NEW code - stuff nobody's done before, so there's no StackOverflow examples to rip off use for training. Not to mention figuring out how to test the system to prove it actually does what we want it to.
This is kinda the point, nothing I've seen the claimed AIs do has been anything but derivative and often poorly, despite my having access to some rather expensive paid for services they're still not capable of creating new work, they're all just regurgitation or, at best, rewrites of training data.
That's not to say they're not useful for some tasks, but they're not intelligent or capable of anything other than being a useful natural language search interface
Having worked for a printer company, the "jamless printer" will never exist. It's always possible for the user to jam it by putting something in the output tray, or using extra cheap paper.
Much like the "paperless office", which still doesn't quite exist.
Au contraire.
I've now worked in two of them so far. Commercial grade copier/scanner/printers can handle hand written notes and spit out PDF's quite well.
I've even practised "Telecommuting," or as we now call it WFH.
Yes it really does feel as if the 90's have finally arrived.
Whatever next?
Happy New Year to all members of the Committee.
I guess we'll be in for an even bigger avalanche of even lower quality products and services from those companies that buy into this nonsense...
The difficulty about -e.g.- booking travel is not the booking travel part. It's the fitting it into all the other things that matter to the traveler and not all of which are expressed to, or even /expressable/ to the system. In the time it takes to go back and forth with the system ("prompt engineering", what a dumb term) and still not getting what you wanted, you could have booked it yourself and finished another task.
What these systems do is do a bad job at the so-called low hanging fruit, and do nothing for the actual hard bit, while selling it as the opposite of course...
I think many companies, especially those with a monopoly or major market share don't care anymore about quality.
Consumers are trapped to a high degree now, much more than in the past so they just have to suck it up (crap products, half backed products, crap services, products that become obsolete quickly etc) .
And these companies know it...so lets increase Cxx pay by killing 10,000 employees with AI... the companies end users don't matter any more.
Bluck
Why does she still have a job? Is her company's AI too shit to do the job of CFO? The people who keep saying this sort of nonsense need to eat their own dogfood and replace themselves with AI to prove that their claims are true. You can normally tell a good IT product by the fact that its creators use it themselves, so start demonstrating your claims by quitting and let the computer do the work.
You will have seen media coverage of the “AI” company advertising their wares with big controversial posters saying things like “stop hiring humans” etc.
The story is made all the more laughable because the company… has a careers/recruiting page.
Admittedly the page makes it sounds like a complete hellhole that only a mindless machine would want to work at. Tbh I feel bad for any of the “AIs” that do work there.
Friar seems to be saying that if you replace humans with "AI" then the company will save money. She then goes on to say that OpenAI will generously relieve you of the savings by charging you more for the product.
How kind.
Then there is the point that if everyone, bar a favoured few, are made redundant who will have the wherewithal to actually buy anything?
The likes of Musk might be able to take up a bit of slack but even they do not have the purchasing power of the rest of humanity.
This is bullshit, wishful thinking and special pleading. But then again so is much of so-called PR.
The CEO of OpenAI is Sam Altman, famous for a cryptocurrency. My expectations may be low but I am sure here can go lower.
Friar seems to be saying that if you replace humans with "AI" then the company will save money. She then goes on to say that OpenAI will generously relieve you of the savings by charging you more for the product.
There would still be many cost savings, even if the AI and the human had the same "dollar cost".
The AI can work 24/7. It won't go on strike. It won't leave for a better job. It won't get sick, or die. It won't start slacking and watching netflix instead of working. It won't get handsy at the office party. Its unlikely to ever need a pay rise - costs of these things tend to fall over time. It can be scaled with demand instantly. It doesn't need an office. Or training. Or even managing.
The simple fact is that a lot of the work done in an office actually isn't very complicated at all. Its not about automating 100% of 1 role, 10% of a role 10 people do is just as good.
You don't need to buy the hype. I certainly don't. But your manager or their manager or their manager will. And then your job will go anyway.
They will use AI for 'customer service' reducing it from poor to completely crap.
There is a strong likelihood that the use of AI will see companies hitting tipping point and starting to attract real hate from customers in the way that politicians have been doing. Companies that use real people, like those who have local call centres, will be more popular.
I would point out that AI may offer the best savings by replacing executives and those further up the pyramid. They cost the company more and tend to do less work, less well than those below them. So for real savings, keep your downtrodden, poorly paid workers and replace your overpaid executives with AI.
We're already not trusted to wire a plug on most household appliances. That said, at my first job I had to wire up RS232 leads for our data loggers until I was told by the building manager that it was far too dangerous for a mere mortal such as myself to do it given that electricity (albeit nothing that could do much more than frazzle a moth) was involved and that his electrician would henceforth be the only person allowed to do the job. Accordingly I send a few cables over for wiring with detailed pin plans and wait, and wait and ... After a month I get the cables back only to find that every single one of them didn't work because they were wired incorrectly but in a bizarrely different way each time. I then told said building manager that as I actually needed the cables to work, from now on I would do the soldering. Not that I was particularly good at it, but at least the cables worked.
Throughout my career as an EE in product development, the one constant task was making cables. Seems you never have the one you need when working in the lab, so the best you can do is to have a parts stock of the most commonly used pins and shells, the appropriate crimpers (and pin removal tools!) and a couple of reels of cable.
"They will use AI for 'customer service' reducing it from poor to completely crap."
Please do let us know which company is providing CS which is 'poor' as it would be a huge improvement on what is the norm nowadays !!!
[They obviously 'really care' about 'their' customers to stay at 'poor'. !!!]
:)
Spectrum: "Oh, you want to return this cable modem? We need proof you're allowed to do this from the original account payer, an account number, and did we mention proof?"
UPS store in same shopping center: "Oh, you want to return this cable modem to Spectrum? OK, here's a receipt showing what serial number you returned, keep it in case something goes wrong"
Spectrum a while later: "Thanks for returning the cable modem"
Welcome to the wonders of cable company "service"
"Companies that use real people, like those who have local call centres, will be more popular."
The pharmacy we had been using (CVS, name and shame!) has changed their phone system to force callers to either lie and claim they're a provider (and thus actually make the pharmacy phone ring, though they might not pick up), or leave a message and hope they call back. Eventually.
There's a local, non-chain pharmacy that answers calls promptly, actually notifies us of what's going on with prescriptions (including "we can fill it but your insurance probably won't cover it" instead of CVS's "computer says no" once arriving at the window), and has things ready when they say they will. Guess which one we're using now?
They will use AI for 'customer service' reducing it from poor to completely crap.
My local doctors surgery staff could be replaced by a pleasant sounding machine and it would definitely be an improvement. Even one that just treated you as benign, rather than actively considering you a nuisance because you need to book the appointment the doctor literally just told you to book.
https://archive.org/details/omnibookofcomput0000unse
"It's 2025 and they've closed down the museums. The works of the masters lie in the darkness of subterranean storage vaults. People don't need to see them anymore. Why should they? In the privacy of their homes, they can view the entire collections of the Louvre on 3-D entertainment modules or visit Michelangelo's Florence via electronic brain stimulation. No more hassles with crowds of tourists or with pompous tour guides.
"Home computer systems craft 400-page novels in hours, suited to the owner's personal taste, or duplicate the Mona Lisa from recyclable materials so that only a chemist could distinguish the copy from the original.
"In a robotic society we won't work. The robots will do most of the jobs more efficiently than we could do them. Such robots can usher in an age of super-abundant, very inexpensive goods. But there's a catch. If humans don't work, do they deserve to be paid? Where can they get the money to buy all the wonderful and inexpensive products the robots will produce?
"We can't have a society of unemployed people who can't afford the products made for them by the robots that took away their jobs. That's absurd. We must devise a scheme that will shift the manner in which we receive income without endangering either our standard of living or our self-respect.
"Robotics is the challenge of the future. The government could move us to accept it by forming a quasi-public agency to sell Victory Bonds for the Future. The money-gathering agency that issues the bonds might be called the National Mutual Fund (NMF). Here is how it might work.
"With the money it receives from public investors, along with money appropriated by Congress or gathered from other government sources, the NMF would build a significant supply of cash. The NMF can use this cash to finance companies that want to adopt robot technologies...
"Now let's look at the other side of robot economics: As the new technology expands, workers will be pushed out of jobs. We must plan to ease their discomfort and make up for their financial losses.
"One way would be to allow the workers to own the robots that replace them. As owners, they could lease the robots back to their former employers for use at their old jobs. The plan is highly speculative, of course...
I think there is a reason most main-stream Sci-fi from Star Trek, Dune, Foundation, Terminator, Battlestar Galactica etc have banned robots/androids/AI is this or some war/rising against machines and MegaCorp’s.
Literally who do these people think will fund society - and yes buy their products - if hardly anyone has a job. The 0.1% will continue to enrich themselves and will not trickle down anything.
Tech advances were going to reduce our working day down to 3-4 hours a day and allow us to live a life of leisure. Perhaps they did reduce the workload as it was then down to 3-4 hours a day, but then a load of bullshit jobs were invented to fill it back up again.
The placating retort from the proponents of robots (and now AI) to those luddites opposing progress is that the incoming technology will employ vast amounts of people to wrangle and otherwise maintain the technology, and, depending on the tech-boner of the responder, it would lead to even more people being employed than are employed now!
But anyway, I'm looking forward to the day when Microsoft Outlook will take my dot points, expand them into an impressive missive and then the receivers of said email then ask Microsoft to summarise this damned long winded diatribe into dot points dammit!!!1
"It's 2025 and they've closed down the museums. The works of the masters lie in the darkness of subterranean storage vaults. People don't need to see them anymore. Why should they? In the privacy of their homes, they can view the entire collections of the Louvre on 3-D entertainment modules or visit Michelangelo's Florence via electronic brain stimulation. No more hassles with crowds of tourists or with pompous tour guides."
But how are the Stop Oil! protesters going to get their publicity stunts now ????
One way would be to allow the workers to own the robots that replace them. As owners, they could lease the robots back to their former employers for use at their old jobs. The plan is highly speculative, of course...
Why would an employer lease a machine from a former worker rather than buy their own which would always be cheaper?
However the post work for money future plays out, the transition from here to there is incredibly unlikely to be peaceful, or enjoy evenly distributed challenges and rewards.
Do corporations need all that headcount?
There's always some fat in the system that they don't seem to recognise, mostly in back office departments doing jobs to make execs feel good such as generating endless pretty reports and dashboards that are rarely acted upon - data driven old boy! The only time they're acted on is when they're contrived to support a decision the exec has made based on emotion and personal experience - like consultancies. Also the endless internal marketing crap and endless 'troop rally' emails that are to show the middle manager is supporting his boss are truly pointless effort. Perhaps LLMs could both create and read them for us? That would save time.
Regarding the LLMs, if they are used to justify headcount reductions I think those corporations will suffer because the competition will use them to do more not the same. The other factor not accounted for, is that people sell to people and people are more complex than LLMs. The LLM's have sussed some of us, after all they feed on our data but they don't actually relate to people. There is nothing more annoying than being stuck with an unfeeling, narrowly confined chatbot because it's cheaper. If you make it as good as the Wayne and Tracy fleshy chatbots on the helpdesk it'll probably be just as expensive. Good LLMs are big, expensive and still are wanting when in complex contexts that humans intuitively understand, well some anyway. They are also insufficiently sceptical of what a human has told them be it directly or in their training data. LLMs supporting our lies is not good for humanity in the long term.
Don't believe everything you read in the papers (or Internet) and garbage in, garbage out still holds true.
But hear me out. What if we let the ChatBot write a legal argument. Then we could let a paralegal read it, decide to throw it all away, and write a good argument with real references!
This would give us the best of both worlds: the same argument could be written in only a slightly longer time, and Altman and Mush could get cash for selling an essential service, and credit for saving humanity!
There are some parts of some jobs that could be done by a chatbot. For example, support requests which often fall into two categories: the one where the user has to read the manual but didn't and the part where something weird is broken and we need to figure out what it is. A chatbot could, in theory, answer the former basic queries. Pretty much everyone would be happy if it could do that. The problem is that, when it isn't a basic query, it may try to answer it anyway with an answer likely to be unhelpful at best, and there's a chance that, while answering something basic, it will go off the rails and print something ridiculous nonetheless. The technology superficially looks like you could use it for some narrow use cases, but each unreliable example makes it harder and harder to do so.
It was only a matter of time before this was the latest mantra.
Like so many of the El Reg posters have already alluded to, there's no replacement for human beings.
Sure, I've enjoyed the LLM new and shiny and exciting, but when you delve deeper, could an LLM do my job or those of my colleagues? Unlikely.
AI/ML/LLM are a distraction in most corporations, it's new, and for some manglement, exciting, as they "think" they can lay off staff and make more money, because that's what the vendors are saying to them, whilst omitting to say that the cost of the LLM will be bundled into your deal and will no doubt not be as cheap as the manglement manglers thought it would be.
The thing is, corporations have found increasingly creative ways to hire and fire their meatsacks on demand.
I rather suspect they're going to find it a great deal harder to get out of their contracts with the AI megaliths - or indeed retrieve their data - with similar flexibility.
Still, I look forward to seeing the enthusiastic early adopters suddenly finding themselves to be the party with the least expensive lawyers for a change.
When AI can go out to site for me, look at a chemical plant or fuel installation and tell me what's right or wrong with it and how to fix it and write the report with graded risk levels and recommendations then I'll be out of a job.
Like the paperless office, it strikes me as more hype and hope [for M$ profits] than reality.
SWMBO works for a local council in social services and they still require many documents to be printed and wet signed.
And we all know what the result will be; Chinese whispers in the age of hallucinatory AI, autocorrect and probability of word adjaency:
Boss: book me a plane to Paris
His AI to travel company AI: Book a plain journey to Paris
Travel company AI to shipping company AI: I need a book about a simple trip to Paris, Texas
Shipping company AI: I have written you a book about films by Wim Wenders
Travel company: Thank you, I have received your essay about the City of Angels
Boss's AI: Thank you, I have your travel guide to Los Angeles
Boss: ???
Boss's boss: why have you missed the European meeting, you are useless, I have replaced you with an AI.
Boss's boss's boss: Error at line 153; rebooting.
The faster they can get everybody using it, the less of a problem this becomes. 1) Because there are probably cross licensing opportunities - you're using company X's IP but they're using yours.
But 2) because nobody is going to let the economy collapse under an epic sue ball (and the courts couldn't cope with all the cases). So if (1) doesn't cover it, rules will be changed and a accommodation arrived at. (Think about music sharing. How many individuals were unlucky enough to got threatened, let alone prosecuted, compared to those who shared?)
But that all hinges on getting lots of people using it quickly.
>This raises the question: why recruit somebody if an AI can assist lawyers as a virtual paralegal, help academics with their work, or do something as mundane as booking travel?
Can the 'AI' do that? To the point where you no longer need the assistant?
We'll see, I guess. So far, though, I don't think so. In any job that requires some exactness, LLM output needs to be very carefully double-checked, and it will sometimes be wrong in subtle ways. It takes longer to validate it than to do it myself.
Your example over-simplifies the paralegal's task. The human paralegal recognizes when "tweaking" an existing boilerplate document to fit the client's particular situation would become a major remodelling project, requiring review and input from a full attorney. LLMs cannot do that. So, the full attorney has to thoroughly review everything the LLM puts out, negating the economic reason for employing paralegals (or LLMs) in the first place.
"Now headcount is coming into focus."
Well that would be because headcount increased drastically a few years ago due to everyone been required to stay home. So internet usage increased and there was increased money flowing in.
But people are not required to be at home anymore, so the money river has dried up. And much of that increased headcount has not increased productivity, so it's just increased cost.
Overall, I think AI is more of an excuse than anything else. It's replacing people that didn't need to be hired in the first place, but it saves face for the boss.
The headcount barely needed to change in those companies- there was some scaling out and making do with what tech there was as supply dried up … but fundamentally in big tech’ a lot of piss poor strategic management and some Hype and some removal/denying employees from being available to others.
Where there was a need for bums on seat was in retail, warehouse and delivery. As I came back from furlough in July 2020, they were already reining this back in - Tesco warehouse where I worked as a key-worker for example.
Many corps monitor their employees' whereabouts via cellphone tracking, badge/keycard swipes/scans, video surveillance, and PC spyware. Your potential trip to the beach might be noticed bt TPTB.
Thus, use your tool to do the job in five minutes, use Auto-It on your work PC to stuff virtual keystrokes into your input stream (presuming your corp's spyware doesn't intercept keystrokes at a lower level) so it appears to the spyware you're editing documents, revising spreadsheets, writing code, or whatever. These revisions, of course, will be to bogus/harmless documents, and eventually backed-out by your "Slack-O-Matic" Auto-It code.
Meanwhile, sitting at your desk, using your personal laptop, connected to the Internet via your personal phone, you code your own projects, surf cat videos, and whatnot.
Be sure to have a "Boss Key" package running on your laptop, as well as having it fitted with one of those 3M "privacy" screen filters.
(Icon for "Escaping from your job while you're still in your office.")
So here's something I am seeing right now.
AI and automation being pushed. The code is often sub-standard (I've seen it - it is terrible). It covers currently deployed systems from a number of different vendors. The people who used to deliver/support the systems are moved onto other things or let go. New versions of software are released, things change, but the people who used to know what was going on and were able to look at problems as they arose during upgrades are no longer around. AI and automation code no longer works as originally designed. No-one is left who knows how to fix it or even where to begin. The backlog of requests to fix issues just grows exponentially as time goes by.
I've seen entire customer environments impacted by so called AI and automation. I've seen firsthand the delays in resolving issues as there are few people left who know what to look at. The people you do get (if you're lucky) all tend to be older as there is no attempt anymore to teach younger people those kinds of skills. As these older people leave the workforce, there is no-one left to replace their skills.
AI and automation are tools that are meant to be used to aid people in their jobs - not replace them. But, as usual, C-Suite have their heads so far up their arses they just seem to ignore this in favour of reducing headcount.
I'm waiting for AI to wade into the professions that require certification.
For a plain vanilla multiple choice exam I reckon an AI would pass with a 100% score unless there is a power failure (it would require an open book exam for a human to match perfect recollection), but I'd like it (a) develop an novel, independent thought and then (b) defend it at a thesis hearing.
That said, the way HR is going about recruiting people I'd say AI is already positioning itself to replace people as capable people whose real talent can only be read between the lines of a CV get filtered out already - by AI..
That apparent shortage of suitable candidates may already be more artificial than we think - the robots are already taking over.
Bonus question: if HR is already using AI to select people, why do we still need people in HR?
"Bonus question: if HR is already using AI to select people, why do we still need people in HR?"
Well, some one has to clear the paper jam in the paperless office printer.
Me, I'm still trying to figure out why there is a printer in the paperless office?
I'm just confused......!
Is a hallucinating psychopath that will make s**t up at the drop of a hat, be endlessly sorry if corrected and do exactly the same thing over again EG disregard gender preferences for answers.
The only good news is that unlike a real psychopath it's not actually after your job.
Remember it's not "AI," WTF that is, that's after your job, it's the dumbass C-Suite MBA types who believe this s**t and who onlyever see staff as a cost and who regularly demonstrate the Dunning-Kruger effect*
"AI" is like the "Personal Information Managers" for whole companies. It promises it will organise your life. But actually you have to be somewhat organised already to use it, with the possible exception of Lotus Agenda. Outlook appears to have (slightly, and very gradually) incorporated some of that tech in it's calendar function.
Part of the issue of "Customer Services" is the perception that a trained monkey could do it. If of course it could understand a full human language (in the UK that would be one of the 3000 distinct accents/dialects you could be talking to) and keep track of a dozen logins (which expire on a more-or-less monthly basis, but not simultaneously in best BOFH style) as some of my customer service friends do.
Reading between the lines integration between internal systems is pi** poor and wastes massiveamounts of their time in cutting and pasting the same information.
Another class of fu**ups are botched account start-ups. These could be eliminated by defining "assertions" on a database record and then running nightly sweeps looking for failures, IOW Looking for accounts with impossible settings EG A live account with no billing, paper or paperless.
This is not "AI," it's business improvement.
However this requires people who understand what the business does and how the business does it. IOW real management skills.
*Or as I like to think of it "The arrogance of ignorance." :-( Which could be summed up as "I don't need to know what this company does, or how it does it, because I'm an MBA"
Obviously the widget seller is expounding the alleged virtues of the widget. It’s as good (better sometimes) as a search engine - and that’s already amazing, but it isn’t what is now being touted. What is now being touted is (politely) wishful thinking.
In my work I see people doing complex tasks in series using a computer that could be automated using BASH/Python/tcl/whatever but the person doing the task (which requires a good level of domain knowledge) isn’t proficient at any computing scripting language sufficiently not to get tied up in proverbial knots and so they get the job done in the way they know how, in the time it takes.
Imagine I was sat next to them with full context awareness and could provide them with the scripting to do the job. Would it work? Yes. Would it be helpful? No. They still have no competence in the script/programme and now there’s a new artefact that needs looking after and maintaining.
As is pretty obvious to any non-sociopath business leader, the right thing to do is to invest in people to ensure you can keep up skilling whilst there are enough people to do the actual job. And keep enough people, or don’t do that job.
Once the pressure is on to get things out in a hurry and there isn’t enough time/people, then people can’t upskill or learn new things which could eventually allow them to become ‘faster’ in the future because the time between now and eventually is too great for the short termism and the output can’t be slowed down. Eventually this part of a business dies.
Are Excel Macros, possibly supported by VBA.
In some ways Excel is a whole separate paradigm, but that's a discussion for another time.
BTW that's a classic development question. Do you
a)Train domain experts as devs b)Teach devs about the subject domain c)Form a team of both to work the problem
Different organisations at different times have come to different conclusions. C is the optimal solution but there's usually some reason they can't spare X staff to do this so it's likely a or b.
Either option opens the chance of some breakthrough strategy that had not been considered before by the cross-fertilisation of ideas.
OTOH both offer the option of ham-fisted implementation using brute force methods when someone with deeper understanding (either of the domain or the language) would have done better.
Remember the first managers-can-write-code-now language was COBOL. How did that work out?
I believe (one of) the idea(s) of COBOL was not that non-technical managers could write programs, but that they could read and understand programs.
The problem was/is that non-technical managers do not want to read programs. So, they don't.
This will hit the public sector like a tsunami under whichever government comes next, in just 4.5 years time. Productivity of their homebased clerical staff is so low, while the unit cost because of the solid gold pensions, so very high. Everyone in the public sector needs to start spending serious time and effort in delivering much more value in order to make themselves harder to automate, because once it begins, and begin it shall, it won't stop and will become a time play for most of the lifers there.
The advantage is that having automated the lowest tier of workers, you then won't need any of the very deep hierarchy above them either, which is why the cost savings vs the more efficient and productive private sector will be so much greater.
The old "Yeah, version X was s**t but new super-duper X+1 solves all those problems and is a piece of p**s to roll-out"
Some of my UK customer service chums (with a business process outsourcer. So if the company don't compete, the company is done.) during lockdown went from office based to WFH literally overnight, using the tech built into Windows since the early '00s.
The business expanded and took on staff they had no hope of accommodating in the floors of the tower block they leased IRL.
And here's the big IT news from the pandemic lockdown.
Work continued to get done at the same level. Civilisation did not collapse because managers could no longer physically look over people's shoulders.
And govt systems (BTW that BPO had govt contracts) run on HW and SW all those I-know-all-about-MS-products developers know literally f**kall about (Fujitsu, formerly ICL mainframe running 1970's 4GL? Various flavours of actual Unix?) are much worse.
BTW You're assuming that all those jobs are actual Civil Service roles, and not being done by BPO staff on whatever contract they were hired on. So the benefits to HMG are much less than you seem to be imaging.
You are on safer ground that AI will grow at the rate at which AI vendors can con convince managers that it can do what they claim (at least in their oh-so-carefully-structured "realistic" demonstration).
IOW SOP since the first software product (probably an accounts package, like Horizon) was over-promised and under-delivered by the first actual SW Sales person.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
