DeepMind Co-Founder: AI Is Fundamentally a "Labor Replacing Tool"
gizmodo.comI consulted for three different large organisations over the last 14 months. All of them chose to let their workforce know that the management are "looking into using AI". In all three cases one of the first questions during all-hands town halls was "are we getting replaced by AI?" People are freaking out and managers are loving it, because they have shifted people's attention from how little they are getting paid to fearing being replaced by a hallucinating algorithm. So, based on my limited experience, I would argue that AI is a wage rise suppression tool, not a labour replacing tool.
My townhalls have been very different. My company introduced a harsh policy against AI, even local LLMs. The main fears were leaking IP & confidential information and contamination of code bases with poor or copyrighted code. That policy may loosen in future but I think it was the right move.
I think in the short term that definitely makes sense, but long term… I think we’ll see more and more jobs be automated away. Look at what Amazon is doing with their new wheelhouse robots. In 2 to 5 years that’s going to become mainstream. Maybe sooner.
Wheelhouse robots? Could you be more specific?
Amazon has been using robots in their fulfillment centers for years now, and there are plenty of public videos out there on those processes. But I don't see how this would replace the use of humans in other critical aspects of fulfillment center operations.
Sorry autocorrect. i meant warehouse. Have you seen these? https://youtu.be/ZWonAz7Kczs?si=a0yF-YRRmXYsYj-Q
They are not the robots I assume you’re referring to that move the crates of orders around that roll on the ground.
I had not seen those, no. I had seen the robots shown at https://youtu.be/0MCMB53jlIk?si=QwWegssGG_d6utJL and https://youtu.be/G-WdDeQ4TKw?si=b3IJF2OMcFYKEIRt which I think are likely to be much more useful and practical in a warehouse setting.
IMO, human-style robots are not a good solution for automating a warehouse. They're too limited in terms of how much they can pick up and move, and they're too small. They can't do the highly dexterous operations that currently require real humans, and they can't replace what the big robots can do.
IMO, anyone who is pitching human type robots today is just spinning a bunch of of smoke and mirrors. The real success will come from completely redesigning the whole system around the new robots that will be much more effective and efficient.
Their new robots are dexterous and able to automate much more complex tasks than the existing bots.
Any technology is fundamentally a labor replacing tool. Isn't that the whole point of "technology"?
Society will adapt. As it always has. Most Americans no longer work on farms or factories either.
Also, demographics and other changes (in many developed countries at least) mean we really need more labor replacing tools, as there’s simply a lack of human labor supply in many areas.
Of course, if the labor replacing process is too rapid, then that could become a problem.
As I hear about AI replacing workers, I think back to the short story “Manna”
I think about this all the time.
Economies are human constructs and can be changed with political will.
For those lazy of reading there's the Wikipedia summary https://en.wikipedia.org/wiki/Manna_(novel)#Plot_summary
Thanks! But it really is worth reading the whole thing.
Great read. Thx for that!
I really worry that we're discussing these topics like idiots. I totally understand why you would look at a peice of technology that can enable 1 worker to do the work of 5 and panic. But... we're not living in Dickensian Britain. We know what happens when technology enables massive productivity improvements and the answer isn't mass unemployment. We're not living in some dystopia where typewriters caused mass unemployment.
I actually think to a large extent the corporate world is putting the cart before the horse. Unemployment is at an historic low, wages are growing fast and so this is a great story to tell as to how you're going to continue to exploit labour to enrich capital.
We've gone from a single person being able to buy a house and feed a family, to two salaries in many cases not being sufficient in the past 50 years. While you're right that mass unemployment didn't happen, wage stagnation certainly did.
This seems like a very American perspective. So may I suggest that wage stagnation came from no longer having a underclass of women and people of color to be underpaid for their labor, and frozen out from the general market? Additionally, the US went from being one of the few industrial nations to not have bombed our factories and streets to having competition from other countries. Americas golden age is not really a time with conditions that are easy to replicate.
>So may I suggest that wage stagnation came from no longer having a underclass of women and people of color to be underpaid for their labor, and frozen out from the general market?
Wage stagnation is everywhere in the west, even more than in the US, and Europe didn't have an underclass of women and people of color 50 years ago in the prosperous times when a factory worker could buy an apartment and support a family, so let's not point the fingers there.
Wage stagnation exists because all the fruits of increased productivity gains in workers' labor from the last decades have been skimmed at the top instead of trickled down to the workers in the trenches.
Secondly, it exists because of the last 30+ years of corporate mergers and acquisitions have a concentrated a large piece of the pie of certain industries into the hands of only a few ultra-large players who get to dictate the global market in terms of wages and prices, due to the lack of competition, meaning workers have less bargaining power for wages and consumers less bargaining power for prices.
Thirdly, globalization and lifting of trade barriers, meant that those few ultra-large players left after mergers and acquisitions, could tap into places that offered the lowest wages, lowest taxes, and lowest regulations in terms of environment and labor, enjoying all those benefits simultaneously instead of having to compromise, meaning they could have their cake and eat it too, resulting in massive more profits for them and a global race to the bottom for most of those working for them.
>Americas golden age is not really a time with conditions that are easy to replicate.
America is still living a golden age, due to being the world reserve currency, dominating the finance sector, the software sector (the only one with major growth and the most profitable), being land, energy and resource abundant, having no enemies at its borders, having the worlds biggest and most power-projecting military, and its citizens enjoy one of the highest purchasing power in the world.
Is that not a golden age or what? Most countries would be happy to tick just one of those boxes, let alone all of them combined. I think some Americans often forget just how privileged they are compared to the vast majority of the world.
When the demand for labor continuously drops but the supply increases, what happens to the ability of labor to price higher?
That view shows how drenched our society is with laissez-faire capitalism as if it's the only thing in existence.
'Free' markets, supply & demand, governments interfering in those markets here & there, is just 1 of several options.
Some kind of UBI, "social credits" that everyone simply receives (as depicted in 2nd part of Manna) + some practical limits on resource consumption, is another option.
There are other options. And like another poster said: choice between those is one of political will. This relies on some conditions though:
1) A significant fraction (if not majority) of the population sharing a similar vision.
2) A functioning democracy.
3) Elites, corporations etc not powerful enough to subvert politics enough to prevent that vision from becoming implemented.
And
4) Absense of external forces big enough to undo progress. Effects of climate change, a huge meteor hitting Earth, aliens showing up that are out to exterminate us, another pandemic, etc, etc, etc.
So don't hold your breath. Much is possible but our odds of reaching some kind of utopia aren't great.
>wages are growing fast
We clearly live in different countries.
Wages may be growing… but buying power is definitely declining.
AI fact-checkers of the world unite!
When anecdotes disagree with data, double check the data.
Let me guess which co-founder. Yep.
What's your point? Is he wrong?
My point is that he has a history of making strong, public declarations, that happen to align well with what he's working on at the time.
Tractor company CEO: Tractors are fundamentally a “labor replacing tool
More than a year into ChatGPT and… unemployment is at historic lows.
I don’t remotely see the present generation of LLMs as being “labor replacing tools.” I’m aware of empirical work which shows them to be useful complements to human judgment, not substitutes for the same. I’m also aware of one study showing (iirc) less work for illustrators on an online job board. That latter case is the only case im aware of where workers have been replaced though if you have citations to actual empirical evidence for job losses please comment!
An alternative theory: Part of what AI CEO’s are doing is marketing their products. If you want to get investors and customers to pay your very expensive compute bill you’d better sell your product. You can sell your product by claiming it can replace workers. That doesn’t mean it can do it. If the current generation of LLMs can replace workers, that is not obvious.
Perhaps some future generation might be, but that will require something fundamentally different than what’s currently available. (And no I don’t think “more data” is enough.)
I'm always wondering when people speak about (un)employment. Like ok non-farm payroll numbers may look good but what kind of jobs, for what kind of pay?
Logically, employment may increase while people's standard of living may get worse at the same time?
Is there somewhere more specific data?
Other than that, I think I remember an article claiming that the introduction of thr personnal computers led companies to be able to generate the same revenue with less people. Like it used to be 8 employee per million USD and it got to 5. Then again, depends of the size of the average enterprise and the market size (which increases with population increase)
>Logically, employment may increase while people's standard of living may get worde at the same time?
Yes it can and yes it is. I'm also baffled by people keep pointing out low unemployment like it's the be all end all of all arguments for economic prosperity.
Obviously sky high unemployment like in the great depression isn't great either but just because people are forced to accept any one of the abundant gig economy jobs, no matter how shitty, or even more than one job, due to the high CoL and low welfare, it definitely reduces unemployment numbers on paper, but is that ideal?
Maybe some jobs are so shit that people would rather bum around on minimum welfare than get dirty, tired and aching bodies just to only to make peanuts over welfare.
Maybe the median inflation adjusted take home wage of the employed person after subtracting essentials like rent, utilities, bills and food, would be a better metric to measure economic prosperity of the working class than collectively clapping at record low unemployment numbers.
So true. And unemployment doesn’t take into account disenfranchised workers, which is significantly higher than it has been in recent history.
What's the measure of disenfranchised workers? I couldn't find any good source.
I believe the technical economic term is “discouraged workers”. Someone correct me if I’m wrong.
This is a useful resource on aspects not included in the unemployment metric: https://www.investopedia.com/financial-edge/0609/what-the-un....
Ah, thank you. Found some useful links here: https://www.bls.gov/cps/lfcharacteristics.htm#discouraged
Seems like this measure of discouraged workers is down over the past 10 years: https://data.bls.gov/timeseries/LNS15026645
I now remember what metric I was thinking of… participation rate. There was a charter data storytelling post about it a while ago that I’m not able to find. But apparently participation rate on the workforce are down. I don’t remember what time scale. But I believe that metric includes discouraged workers as well as people who have retired. But I believe the angle the article I read was talking was a post pandemic look… like lots of people retired early. I believe this has those numbers.. https://www.bls.gov/news.release/empsit.t01.htm
Here’s a chart: https://www.bls.gov/charts/employment-situation/civilian-lab...
I think this metric, like unemployment, is obviously nuanced and also could be showing an aging demographic and other such subtleties.
Thanks for sharing. The major disruption to the labor market caused by covid (and other long term trends) defies simple characterization. Weird times.
Absolutely! Happy to share, and totally agree. Very strange times.
That would be the basis measurement of the population of working age people not requesting and or not eligible for unemployment benefits, champ.
Unemployment mostly has to do with matching and coordination problems.
AI or increases in labour supply doesn't typically result in reduced unemployment, rather it acts to push down wages. An increase in labour supply could even lead to people bidding over each other with time, so that the result is that people work more rather than less.
Here are a few jobs that I'm aware of, which are already well on their way out:
- Language translators.
- Video annotation, summary, transcription.
- Voiceover artists.
- Graphic designers.
- Warehouse inventory work.
- Personal tutors.
- Junior software engineers.
There will still be openings for these roles in the future, just as there are still openings for Personal Assistants (secretaries, which were once as abundant as teachers and truck drivers), and maybe even the occasional Draftsman.
For all of these jobs, AI systems may be able to solve the easy 80% of the lowest level tasks, they can't solve the hard 20%, and that's where humans will prevail. But AI plus humans will do even better.
I think what's going to happen is that AI will enable these kinds of jobs to be done much more often and in many more situations where they could not be feasibly be performed before, and the actual human job losses will be minimal in these areas.
At least for the near future, and it's hard to predict what may happen ten or twenty years from now.
So true. And when autonomous driving finally get to the safety levels where society accepts it… that will be a shocking number of jobs taken. Also Amazon is going to start selling their self checkout technology to stores… so most cashiers will definitely be out of work in the next decade if not sooner.
What we know of the current Amazon technology is bad enough that Amazon keeps closing stores. It's just not profitable enough to keep them open.
Now, maybe what they're thinking is that these stores were just a training exercise and known to be unprofitable from Day One, and now those training exercises are over so they can start shutting them down. Dunno -- I have no inside information on this.
But I wouldn't be looking to Amazon to solve all these problems for all the retailers. At least, not until they can come out with better technology.
Fair point, but then again… self checkout sucks!! And it’s everywhere. point being maybe it doesn’t have to be very good if it saves companies money. I hope that I’m wrong, but I honestly wouldn’t be surprised
Self checkout as we know it today sucks. Imagine if you could do it based on RFID tags and just walk past with everything still in your cart. Of course, you'd have to make sure that you don't get double billed for anything, and you don't get billed for anything that someone else has nearby.
But if you got properly billed for only the things in your own cart, and that RFID scanning process was quick and easy, then I'm not sure it would suck so bad.
Totally agree. I’d actually prefer this type of system if it’s in fact faster than checkout with a cashier. Especially now that it seems many markets have dramatically cut down on staff and it can take shocking amounts of time to checkout.
It seems like personal tutors will actually be in higher demand if AI learning takes over. I mean kids have had an ever increasing array of ways to learn and supplement materials over the past 20-30 years and if anything education has been getting worse in recent years with many benchmarks for education reaching decades low levels.
Which tool isn’t?
A union signup sheet.
> The tech and media industries—which are uniquely exposed to the threat of AI-related job losses—saw huge layoffs last year, right as AI was “coming online.”
This is one of those cases where correlation isn't causation. The actually meaningful thing that came online were interests rates which people discovered can actually be above zero, and one thing that went offline was the pandemic.
AI as it currently exists is not a job replacing tool, it's a task automation tool and in that capacity no different from any other tool. And companies don't actually fire people when tasks get automated, they give the more productive employees more tasks so they can make the company more money.