Generative AI won't steal your job, just change it, says UN
- Reference: 1692784926
- News link: https://www.theregister.co.uk/2023/08/23/ilo_ai_jobs_impact_study/
- Source link:
The report, titled "Generative AI and Jobs: A global analysis of potential effects on job quantity and quality," observes that the launch of OpenAI's [1]ChatGPT has created concerns about job losses, as machines capable of creating text and images, or analyzing data, perform tasks many humans are paid to perform.
But those concerns may be overblown, the report suggests, because the ILO's researchers [2]found most workers are not at high risk of being replaced by AI.
[3]
Instead, the [4]report's [PDF] main finding is that "most jobs and industries are only partially exposed to automation and are thus more likely to be complemented rather than substituted by AI."
[5]
[6]
Anyone who's used generative AI and seen its, shall we say, imaginative output will likely have come to the same conclusion.
The report finds that while tools like GPT-4 can do some of the tasks humans perform in fields such as administration, customer service, data management, and providing information, AI models cannot do all the work required in most roles.
[7]
"As a result, the most important impact of the technology is likely to be of augmenting work – automating some tasks within an occupation while leaving time for other duties – as opposed to fully automating occupations," the report states.
But clerical workers will feel more impact.
Generative AI could mean certain clerical jobs never emerge in lower-income countries
"We find that only the broad occupation of clerical work is highly exposed to the technology with 24 percent of clerical tasks considered highly exposed and an additional 58 percent with medium-level exposure," the report states. "For the other occupational groups, the greatest share of highly exposed tasks oscillates between one and four percent, and medium exposed tasks do not exceed 25 percent."
That analysis is bad news for women, as the report finds they are over-represented in clerical work – especially in high and middle-income countries.
"Since clerical jobs have traditionally been an important source of female employment as countries develop economically, one result of Generative AI could be that certain clerical jobs may never emerge in lower-income countries."
[8]
Adoption of generative AI in developed countries means workers in those nations are more likely to feel the impact of the technology. The report estimates 5.5 percent of total employment in high-income countries may be at risk of being partly automated, compared to just 0.4 percent in low-income countries.
[9]Google reportedly designing chatbots to do all sorts of jobs – including life coach
[10]TV and film extras fear generative AI will copy their faces and bodies to take their jobs
[11]Netflix offers up to $900,000 for AI product manager while actors strike for protection
[12]OECD finds 27% of jobs are under threat from AI
"We focused on the potential of task automation as of today, without speculating on the numbers of new jobs that might emerge. This approach might have been expected to generate alarming estimates of net job loss – but it did not. Rather, our global estimates point to a future in which work is transformed, but still very much in existence," the researchers said.
They warned, however, that countries need to define policies to ensure that workers' rights are still protected as industries adjust to generative AI.
"Without proper policies in place, there is a risk that only some of the well-positioned countries and market participants will be able to harness the benefits of the transition, while the costs to affected workers could be brutal," the ILO report concluded – as can be expected from an organization that exists to develop and promote policies that make work fair and dignified. ®
Get our [13]Tech Resources
[1] https://www.theregister.com/Tag/ChatGPT
[2] https://www.ilo.org/global/about-the-ilo/newsroom/news/WCMS_890740/lang--en/index.htm
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZOYtKElE8pEjAHplsiCKTQAAA5U&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://www.ilo.org/wcmsp5/groups/public/---dgreports/---inst/documents/publication/wcms_890761.pdf
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZOYtKElE8pEjAHplsiCKTQAAA5U&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZOYtKElE8pEjAHplsiCKTQAAA5U&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZOYtKElE8pEjAHplsiCKTQAAA5U&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZOYtKElE8pEjAHplsiCKTQAAA5U&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[9] https://www.theregister.com/2023/08/17/google_designing_life_coach_chatbots/
[10] https://www.theregister.com/2023/08/07/tv_and_film_extras_are/
[11] https://www.theregister.com/2023/07/26/netflix_ai_manager/
[12] https://www.theregister.com/2023/07/12/oecd_finds_27_percent_of/
[13] https://whitepapers.theregister.com/
We just don't know.
The thing about predictions is they are very hard, especially when you are predicting the future.
Probably in the early 1970s people building cars in factories would have mocked the robots being developed, as they did a pretty crap job. Much like people are currently mocking the output of LLMs. However, after years of refinement we are now in a position where no sane company would consider mass producing cars without using robots. A couple of things though, 1) Today's cars have way more stuff in them compared to cars in the 1970s 2) Many of the jobs have moved up the value chain into things like design.
We now have a consistently better product, which does far more than the previous generation of cars could. Cars in the 1970s used to regularly break down / overheat or just plain fall apart from the rust. They would not start on a cold morning. A bit like software today.
So, perhaps in 20 years time we'll be writing software which does what it's supposed to do, in a much more secure and reliable way? We probably won't spend as many hours writing code or tests as much of it will be done by AI?
And probably the same thing for accountants, lawyers, cleaners and hundreds of other roles. So, yeah I guess I kind of agree with the UN on this one. Although try telling that to the thousands of car workers in Detroit and many other places who suddenly found themselves unemployable.
On the other hand I could be completely wrong, as I said predictions are hard.
Re: We just don't know.
There are only three hard things in Computer Science: cache invalidation, naming things and predicting what AI will do to your job.
Re: We just don't know.
I thought it was TWO hard things in Computer Science? Cache invalidation, naming things and off by one errors.
LLMs will create jobs
Companies need extra staff to correct the mistakes made by LLMs.
Re: LLMs will create jobs
They won’t need staff to correct the mistakes, but rather just to assume liability.
AI models cannot do all the work required in most roles
Of course, what they fail to consider is that an AI model can do part of several roles, so they'll simply consolidate the roles, lay off half the staff, and the rest can work harder to fill in the gaps that AI doesn't do (and harder yet because ultimately AI will be a lousy replacement for a person).
You can see evidence of this sort of thing already - just try to contact pretty much any large company for assistance. They run small client help lines these days because most of it is an automatic system that's designed to maliciously interfere with any hope you had of talking to anybody with a functioning brain. Moreso if after the machine you have to wrestle with a support droid reading cue cards.
ISCO classifications are the weakness
The paper is well researched in general, but the ISCO occupation classifications it uses leave a lot out (e.g. there are no codes for anything explicitly related to risk management). As a result there are serious holes in the profiles (p. 21) and consequently in the overall analysis, because it's just such occupations that are most sensitive to AI-takeover.
I'd love to throw our policies and guidance into a generative AI; and then start asking it questions about it.
Though knowing full well that authoring practises do not enforce cross checking of cross references between documents (just like the ISO-standards) the outcome is somewhat inevitable.
Garbage in, garbage out. The best use for generative AI might be to identify and highlight those errors in the underlying sources.