Is Mandating AI Use in the Workplace Backfiring? (msn.com) 58
Some workplaces are holding mandatory AI training sessions, while others "are requiring employees to use a specific number of tokens," reports the Washington Post, "or log a certain amount of time on AI platforms."
The article's headline? "Why mandating AI use in the workplace can backfire." [A] lot of employers are going about it the wrong way, said Arvind Karunakaran, an assistant professor of management science and engineering at Stanford University. "You can consume a lot of tokens without creating a lot of value," Karunakaran said....
The adoption of AI is happening faster than the adoption of the internet did, according to a new report from Stanford's Institute for Human-Centered Artificial Intelligence. And many users report that AI tools have really helped — whether saving time and costs or increasing innovation. The same report, though, found that one-third of organizations surveyed expected AI-related staff reductions in the coming year. Especially in that context, simply mandating usage won't spark excitement about the technology, Karunakaran said, and neither will insisting that employees use AI to cut turnaround times and ramp up productivity. "People start asking questions like 'What am I as an employee getting out of it?'" he said.
The report he co-authored found that in the short-term, mandating AI tools may ramp up productivity. But in the long-to-medium term, employees tend to feel devalued and disengaged. From there, people find ways to mindlessly check the box.
When employers present these tools and their use as an opportunity for employees to expand their skills and grow their careers, people are much likelier to use them, Karunakaran said. He suggests that rather than requiring employees use a specific number of tokens, companies should encourage workers to experiment and share their own use cases and success stories. Most workplaces are still shutting employees out of the conversation. More than half of employees said they have less of a say than they'd want on adopting new tech at work, according to a report from Gallup and Jobs for the Future, a work and education nonprofit.
Any Slashdot readers want to share your own stories on how AI was introduced at your workplace?
The article's headline? "Why mandating AI use in the workplace can backfire." [A] lot of employers are going about it the wrong way, said Arvind Karunakaran, an assistant professor of management science and engineering at Stanford University. "You can consume a lot of tokens without creating a lot of value," Karunakaran said....
The adoption of AI is happening faster than the adoption of the internet did, according to a new report from Stanford's Institute for Human-Centered Artificial Intelligence. And many users report that AI tools have really helped — whether saving time and costs or increasing innovation. The same report, though, found that one-third of organizations surveyed expected AI-related staff reductions in the coming year. Especially in that context, simply mandating usage won't spark excitement about the technology, Karunakaran said, and neither will insisting that employees use AI to cut turnaround times and ramp up productivity. "People start asking questions like 'What am I as an employee getting out of it?'" he said.
The report he co-authored found that in the short-term, mandating AI tools may ramp up productivity. But in the long-to-medium term, employees tend to feel devalued and disengaged. From there, people find ways to mindlessly check the box.
When employers present these tools and their use as an opportunity for employees to expand their skills and grow their careers, people are much likelier to use them, Karunakaran said. He suggests that rather than requiring employees use a specific number of tokens, companies should encourage workers to experiment and share their own use cases and success stories. Most workplaces are still shutting employees out of the conversation. More than half of employees said they have less of a say than they'd want on adopting new tech at work, according to a report from Gallup and Jobs for the Future, a work and education nonprofit.
Any Slashdot readers want to share your own stories on how AI was introduced at your workplace?
Re: (Score:2)
requiring employees to use a specific number of tokens
Leaking your Anthropic credentials via your Github repo: It's no longer an accident, it's a great career move!
Re: Bettridge's law... (Score:3)
Every use of an AI for work related purposes exposes your workplace.
"present as an opportunity" (Score:5, Insightful)
When employers present these tools and their use as an opportunity for employees to expand their skills and grow their careers...
... those employers would just be lying, and many employees would see through this.
There is little "skill" in asking some AI agent to do something, and even if some employee meticulously formulates his prompts to optimize the effective use of the AI, this will go unnoticed by his supervisors, while at the same time a lot of general thinking skills shrink among the entirety of the population due to the outsourcing of thinking to AIs. The only growth their careers will see is a growth in free time due to unemployment or a growth in unpleasant physical work at minimum wage for those filling that gap until the robots get good enough also in physical work.
Working as intended (Score:3, Interesting)
Humans invented machines to do their work so they would have more free time. Unemployment will not be a problem when everything is being taken care of by the machines and programs.
Re:Working as intended (Score:5, Insightful)
Humans invented machines to do their work so they would have more free time.
Yes, and that is a good thing as long as humans own the machines that do their work. This time, however, humans do not own the machines that do their work, and their unemployment is a problem, because the machine owners have no reason to keep the no longer needed humans around (or even alive).
Re: (Score:3)
Indeed. And the second factor is that the machines are not helping, they are replacing. That means work stops being a valid way for wealth distribution. And we currently do not have other ways that work on society-level.
Now, whether that will work out on the tech-side looks more questionable very day. This time we will probably get lucky and LLMs and generative AI will do very limited damage in the longer run due to sheer incapability and massive drawbacks of a no-insight technology to replace humans with (
Re: (Score:2)
Yeah, there have already been AI booms and busts before. I think this will be another, but next time, as you say, maybe in 10-20 years, they might be too good and all the dreams of the people who want to replace humans may come true.
Re: (Score:2)
This is just gonna be a big crash when the bubble pops.
Re: (Score:3)
Personally, I have been through several AI hypes. They always promise a golden age and then deliver very little. This one is no different. What may happen in 10-20 years that non-AI automation with some help from not very capable LLMs may cause that job catastrophe.
Re: (Score:2)
Excuse me?
It doesn't take much in the way of hardware to run a very competent open source model.
I own my machine, the code is open sourced. If every commercial LLM was shutdown, I could still run my model...
But I also think that concentrated ownership in the very top tier models which are moving very fast towards AGI and beyond is problematic.
Re: (Score:2)
If every commercial LLM was shutdown, I could still run my model...
Are there any open source models that generate 3D models without a 3D card that costs a fortune? I have an RTX 3060, and not only do there appear to be no open source AI models that will generate 3D models, the closest ones (which still suck) require a $3000 video card. Honestly, I don't know the current prices for qualifying video cards, but I know I don't have the money to buy any of them. Meshy.ai costs $20/month for the time being, and it is VERY good for most uses.
Re: (Score:2)
I should also point out that Meshy.ai is completely useless for a wide range of model generation (firearms and human bodies topping my list so far).
Re: (Score:2)
Re: (Score:2)
There is little "skill" in asking some AI agent to do something
I disagree. Since the dawn of computers, GIGO was a thing: garbage in, garbage out. AI doesn't change that principle.
If you prompt AI to do the wrong thing, AI will do the wrong thing, much faster and more completely wrong than a human.
burning tokens (Score:2)
"rather than requiring employees use a specific number of tokens, companies should encourage workers to experiment and share their own use cases and success stories"
It's real easy to type in some crappy prompts that throw the LLM's into loops that burn tokens in exchange for nothing useful. If you are dead set against using AI to help you get things done you will not learn how to work with it and will be much less productive than people who will.
You have to actually know what you want to accomplish, be able
Re: (Score:2)
You have to actually know what you want to accomplish, be able to articulate it clearly even if it takes a few tries, and cause the AI to come up with an incremental plan that will help it get you there.
That sounds very much like you need to do 90% or more of the work yourself anyways. Might explain why nobody sees revenue gains from LLM usage.
This is a learnable skill. If you don't learn it you will waste a lot of time and money, and your employer will notice it.
I disagree. This is a skill that needs real insight and it is also more abstract than solving the problem directly. That means fewer can do it. As far as we know, insight and abstract reasoning cannot be taught, you can only qualify the people that already have an aptitude for it. And those are not many and no amount of education increases their number. Hence what yo
Re: burning tokens (Score:2)
Re: (Score:2)
"what you are doing with LLM use is decreasing the pool of people that can use the tool competently"
Utter bullshit as usual. You haven't learned the skill and you don't even work in the industry.
You have to actually know what you want to accomplish in order to write software manually. Working with AI assistance brings this right out front where it belongs. If you can't articulate the requirements you sure won't be able to write the software.
Re: (Score:2)
Au contraire, I have a respective engineering PhD. But that is so far outside of what you can grasp, it is not even funny anymore. All your comment shows is that you failed to understand mine. And that you are unfamiliar with the research on software engineering.
Re: (Score:2)
"a respective engineering PhD", lol.
Re: (Score:2)
Utter bullshit as usual. You haven't learned the skill and you don't even work in the industry.
I work in the industry. Also happen to be an luser of AI and I don't have the slightest clue what "skill" you believe is involved with using an LLM. One could argue navigating latent space (e.g. prompt engineering) was a skill yet this has gone the way of the dodo.
You have to actually know what you want to accomplish in order to write software manually. Working with AI assistance brings this right out front where it belongs.
Who doesn't know what they want to accomplish in the first place?
Re: burning tokens (Score:1)
Re: (Score:2)
Nice fallacy you have there. Yes, I get that you do not know what a fallacy is.
Re: (Score:2)
You have to actually know what you want to accomplish, be able to articulate it clearly even if it takes a few tries, and cause the AI to come up with an incremental plan that will help it get you there. This is a learnable skill. If you don't learn it you will waste a lot of time and money, and your employer will notice it.
There is no "skill" involved with using an LLM.
Re: burning tokens (Score:2)
A waste of time (Score:4, Interesting)
Re: (Score:3)
Stupid complicates, genius simplifies.
Re: (Score:2)
Yes. And LLM-type AI is very complicated indeed. You essentially add a layer of complexity to doing anything and have to "trick" the "AI" into doing it right.
Copyright risk (Score:3)
Re: (Score:2)
anything that comes out of your standard LLM is illegal and unsellable
They won't decide that.
all the conflicting licenses that need fulfilling.
They might decide you need to pay a little extra to cover licensing.
Re: Copyright risk (Score:2)
UK courts look to maybe be coming close. They had a recent ruling that suggests training on copyright materials is infringement.
Re: (Score:2)
Re: (Score:2)
At least in countries that aren't ruled by CEO's.
Jetsons (Score:3)
George Jetson worked at a computer panel a couple of days a week for a few hours. In the center of the panel was a large red button. Mostly George slept on the job, but occasionally something happened that required him to push the button. I am not sure if at the time this job was supposed to be a fantasy or a nightmare. Now that it is a reality, I see it very clearly as more on the nightmare side.
Using AI is very depressing and also very unavoidable.
Adoption (Score:2)
The adoption of AI is happening faster than the adoption of the internet did ...
Apples and oranges, used for different things. In addition, to a greater extent, utilizing the Internet required getting hardware and infrastructure, which AI uses and is already in place. Lastly, the Internet is many things, more diversified, and AI is a lot fewer, more consolidated.
My take on AI being forced everywhere (Score:2)
Is mandating hammer use in workplace backfiring? (Score:2)
You can bash a lot of things without creating a lot of value.
Tragic lessons (Score:2)
Just the Hawthorne Effect? (Score:4, Interesting)
The report he co-authored found that in the short-term, mandating AI tools may ramp up productivity.
The Hawthorne Effect [wikipedia.org] is based on a series of experiments in the 1920s and 30s. They tried varying lots of aspects of a work environment to see how it affected productivity, like lighting levels, break schedules, and whether workers were given snacks. They found that just about any change led to an increase in productivity, but only for a short time. Then it went back to where it started. The particular change didn't really matter, only that you changed something, and that people knew they were being studied.
This report is likely something similar. If you make people work differently, for a short time they become more engaged. They may also be focused on proving to their manager they can use the new technology. But it doesn't last. Productivity drops again. It's a mistake to assume the effect had anything to do with the particular change you made.
Is Mandating AI Use in the Workplace Backfiring? (Score:2)
I don't know. My team uses Claude to deploy all work they do on their personal machines without any kind of reproducibility. In my experience, I try to use the most expensive model to blow through money and at least make me feel a little better that half of my time each day is waiting for the output of Claude. I assume all other workspaces are like this?
Is there value to putting in your best effort at a company determined to enforce the use of tools that are designed to kill humanity? I mean... the AI the
Re: (Score:2)
Re: (Score:2)
I don't see the point in using slop generators while 95% of my work is already analyzing legacy code and thinking how to fix it without tumbling the fragile stack of cards over. And to try to figure out what the code actually was supposed to do.
What is the point of trying to accelerate something that is not even like 3% of my workload?
Don't get me wrong. I use tools like debuggers, automated testing and code inspectors.
The big difference: if you run them 100 times, they w
Re: (Score:2)
95% of my work is already analyzing legacy code and thinking how to fix it without tumbling the fragile stack of cards over.
Some people use their agent to do this analysis for them, and report that it goes much faster. You might try that.
Re: Is Mandating AI Use in the Workplace Backfirin (Score:2)
It isn't perfect. It isn't replacing you. But it will help you do your job better.
stupid metrics (Score:3)
The only thing worse than not having metrics... ... is having metrics.
Big companies are almost always stupid.
Organizations measure what is easy for them to measure, not what actually matters. The drift between what matters and what gets measured always produces dumb behaviors.
Measuring token usage is perhaps a proxy for "is this person trying to use AI?". That question is only interesting if you are quite confident that at some point, your staff will be able to grow the business value they produce as a result of using AI.
What you actually should measure is business value delivered. Companies are bad at that (Because its difficult), So instead they Measure what is easy to measure, and they don't test the assumption that AI helps staff deliver more BV.
At my small software company, having developers use Claude is delivering tremendous BV. We are getting higher quality PRs, code reviews, and built software, and we are getting it faster. We are getting MUCH faster bug investigations. Overall time to fix is down. We are shipping faster and we are paying down technical debt.
(we may be accumulating new technical debt we don't see yet. time will tell)
Re: (Score:2)
Absolutely. Before AI came along, these companies were already using stupid metrics. Now, they not only have new "metrics"--they can get AI to invent them even more creatively.
My previous employer just laid off about 10% of their workforce. Not because things are going so well because of AI, but, as the COO told the remaining employees, because they aren't using "enough" AI. And he said that if they don't start using AI more, there are going to be more layoffs.
This is the kind of nonsense that's going on. B
Re: (Score:2)
Metrics aren't necessarily a bad thing. You can't iterate on what you can't measure. As Galileo famously opined, measure what is measurable, and make measurable what is not so.
But measuring AI productivity by the number of tokens used is just as retarded as measuring developer productivity by the number of lines of code written. It encourages inefficiency and waste.
At my company (Score:2)
The CEO stopped just short of demanding everyone in the IT dept spend time at home vibing apps. Even Ops, QA, network engineers, ... everyone. He phrased it like "if it were me, I would be doing this every chance I get." I did not feel comfortable objecting.
I recall an article... (Score:2)
There was an article not too long ago, might have been here on /.
It was discovered that, in several situations where employees had a token quota, the employees simply instructed the ai to periodically use a token or made a script to do it and just went about business as usual without ai assistance.
The pendulum swung way too far into hype (Score:2)
Re: (Score:2)
Like imagine an OS or web stack written entirely in Brainfuck.
It's not that bad, if you have a half-way decent macro system tacked on top. M4 is good enough.
Incompetent companies before AI remain incompetent (Score:2)
My previous employer was one of these companies that monitored token usage. As a manager, I was instructed to question anyone on my team, who wasn't using "enough" tokens. They just laid off 10% of their workforce, not because they were so efficient due to AI, but because they were losing money hand over fist. The COO announced to everyone in an all-company meeting, that if they don't start using more AI, they will have to lay off more people.
But the stupidity predated AI by a long shot. Before AI, their pr
Depends on the field (Score:2)
Mandating someone uses a powerful tool instead of refusing to learn something new, where the tool has real benefits: Do it.
Prescribing a workflow for someone who knows better: Don't.
In software development you usually leave the choice of tools to the developers, because each has an own workflow they are productive with. A software developer using vim won't be faster if you force them to use vscode. If they struggle with AI, they won't get faster when you mandate them to use AI. On the other hand will a good
Go full malicious compliance (Score:2)