Or pesimistic because are scared of it... I actually find this video very optimistic (I mean they are not expecting 50+% chance of extinction), but still, implementing a global pause is safer option. I am against the AI race and big tech, but I am worried it is not a bubble. Chatbots probably are a bubble, but not AI in general. Thats why I joined PauseAI. And remember, if someone develops AGI, it will be able to do all jobs (because by definition it must be as good or better then humans in everything). I am not saying that we will have AGI next year, but the best time to have AGI is never. Also, we dont need AI to be able to do 100% of jobs. Just creating deadly viruses could be enought (I am not saying that big tech companies want to create a global pandemics)
It is interesting that the hosts were borderline hostile to him simply because he wasn't supporting the current market hype about AI. It is almost that they couldn't stomach that he had an alternate opinion that wasn't unquestioningly optimistic on AI.
Yeah. 5% is sort of over-estimation. More likely a future prediction. It is hard for a tool to become a user. Biology, chemistry or astrophysics can use some features of algorithms made today. We are not going into miracles like Facebook would like to see it. No, it's not meta. It's Facebook.
Just think about it for a minute! In the 2009 recession we had roughly 9% unemployment. Roughly 4% is called full employment. If you just add 5 % more unemployment you arrive to a very rough recession. But! By 2028 three or four new 100 billion dollar datacenters will come online, and there will be one or two huge breakthroughs by that time. After that it will radically speed up as it will have enormous amounts of feedback loops. So by 2030 it may easily reach 5-10%. By that time robots will also reach the perfectly usable level, where they will start to replace human workers. So by 2035 software AI + robots may replace 15-25% of the workforce. By 2040 they may replace 30-40%. Don't sleep on this thing!
It went from a preschooler in 2019 (GPT-2) to PHD level intelligence in 2024 (o1) it's not that hard to project outward that it will be more intelligent than humans soon all you have to do is follow lines on a graph.
@@williamr9833 I don't think you really understand what intelligence actually means. Intelligence means the ability to learn something quickly, and of course a machine would have that ability, what PhD level people do is not just learning stuff, they're creating stuff for other people to learn, and that entire process is completely different from what the current AI is designed to do, reasoning upward from first principle is still very far fetched. Now you're going to need to dive deeper than an elementary linear assumption of progress, that's not how it works.
@@markarmage3776 You're not well versed on the subject and are underestimating how rapidly AI is evolving. While you mention that PhD-level individuals create new knowledge, modern AI models are already generating original content, formulating hypotheses, and even assisting in scientific discoveries. The gap between learning and creating isn't as vast for AI as it once was. Additionally, reasoning from first principles is becoming more attainable for AI with advancements in algorithms and computational power. It's not just about a linear progression but an exponential one, where each breakthrough accelerates the next. So, following the trajectory we've seen so far, it's not unreasonable to anticipate AI far surpassing human intelligence sooner than you might expect.
@@williamr9833 It's very easy for everyone to create a bunch of things that are absolutely wrong, buddy. They actually need to create stuff that are correct, they're assisting with people who are creating correct stuff, but that's not the same thing. An advance machine is assisting people in manufacturing high technology products but that machine itself can't function without human input. Research hypothesis regarding social science is absolutely meaningless, especially if it's not even backed by a realistic methodology. You're confusing being able to do one tiny part of the whole venture, not even doing it well with being able to do the entire thing, those two are not same thing. Furthermore, that whole exponential argument you is based on nothing, you're guessing using wishful thinking, there's no clear formula or mathematical principle relating relationship between any specific characteristics of any algorithm with the computing hardware with the actual performance. Please go study real science, what's you're doing is making an observed guess on stuff you don't understand whatsoever, like a speculator in the stock market thinking how because a stock is going up, it will never go down. Please be more sophisticated, that's now how science works. Even the people who are researching the thing has no idea what it's going to do next or whether there will be a significant improvement or not, they just take the money given to them to do certain "experiments" and see what's going to hapeen, please distinguish between that, which is actual research, filled with uncertainties and specific details with your version of make belief. HUGE DIFFERENCE.
I certainly hope Bloomberg also grills wallstreet fund managers and AI startup CEOs like this cos it felt like he said the sun won't come up in the morning.
My day one assessment was that the current level of AI is like a college intern. Can say all of the right words, but when it gets down to it is often more trouble then they are worth. Routinely, they will get things upside down, so that a factor that should increase an effect will decrease it. And they will not spot their own mistakes. If you know what you are doing, you might be able to get them to do what you want without the drudgery of writing everything down. But part of the time it would have been easier to just do it yourself.
Literally exactly this. I use AI a ton and pay attention to how each new model improves, but instead of actually linearly improving and getting "smarter," it seems to also get more confident about the wrong answers that it gives you. It is especially a problem when solving a complex problem and the AI is giving you nonsense that is presented in such an intelligent way that it does more harm than good, because you are led further from the actual correct answer.
The real elephant in the room is: TRUST… We have seen more than enough evil done by hi-tech companies with just our data; Imagine what those unregulated behemoths would do to their customers (ie: us, mere mortals) who will rely on their AI services…
Ai doesn't just replace an employee, but it can make employees more efficient, which means less of them required in general. It's an indirect replacement that won't eliminate a type of job, but require less people to do it.
Exactly that’s what many people fail to realise. AI might not literally repack people currently in employment, but it will decrease the need to employ and train new people.
Leading to more companies being profitable which they otherwise wouldn't eventually leading to more demand and jobs. AI has being released for a few years now, and we have record low unemployment
Totally agree Mr. Acemoglu. I do work in oil industry and AI is no way close to replace humans. It is correct that LLM's do increase productivity in certain fields, but especially in math dominant fields, tasks are still done by humans. It might change one day, but not in 5 years.
IF AI sucks 5% of jobs out of the overall market... what does that do to un-employment? Wont that in itself create a crisis? How does one even predict these things?!
One thing's for sure, his job isn't safe. Many academics are not worth the paper their PhD is written on, and they know almost nothing outside of their immediate expertise. Sure this guy is an MIT professor of economics, but there are people who know 1000x more about AI than him, and even they're caught off guard by the direction and rate of progress. This guy is just a talking head with a title and an opinion, nothing more.
@@ClaudiaS193 You're _supposed_ to do that, but many people are lazy and will just assume that the AI is right without question. People already do it with computers and software that don't even use AI.
There's a company here in San Francisco that are trying to create the code to upload to robots how to diagnose basic plumbing issue. I give it 4 years before they have it perfected. No one will want an unknown stranger coming in their house compared to their own robot that can do plumbing.
I'm of the mindset that worrying is futile, the world will always move forward until it can no longer move forward. We all must adjust and adapt, it's that simple.
I tested Gemini Advanced against an attorney’s work on my wrongful injury case and Gemini did much better at covering all aspects of case preparation for court.
5%.. FOR NOW... We're talking about automated intelligence, give it a little more time, and it will be capable of replacing a much larger percent of jobs.
It went from a preschooler in 2019 (GPT-2) to PHD level intelligence in 2024 (o1) it's not that hard to project outward that it will be more intelligent than humans soon all you have to do is follow lines on a graph.
@arcopolarisambient I would recommend upgrading to a paid model. Very soon, people who pay from the premium models will be the haves and the people who don’t will be the have nots in terms of intelligence and cognitive capabilities.
AI is different from the information revolution. The bar for entry in AI is very high. The information revolution is low. That is why there are many innovations in the information revolution. There are only few people who can do AI. So there will be less innovation and less overnight success stories. Gone are the days of the garage entrepreneur.
Well this is why we should let all the big firms create the tech but regulate what they can earn much like utilities - think if it as a gigantic societal capital Infrastructure project - this cannot be allowed to fall into private hands for profit maximisation
As a person involved in creative design of infrastructure, there is a level of creativeness that AI would not be able reached, i.e, the ability to observe and discover how nature works and applying the laws of nature in design. AI is simply huge and speedy processor of big data to facilitate human creativity and innovation.
I remember two years ago a blogger was laughing at how bad Ai was at writing a simple paragraph. Ai is becoming better every six months. This is the worst it’ll ever be. Depending on it staying at this level is a risky bet.
If you’re interested watch the AlphaGo documentary. I agree with you to a certain extent but it’s interesting what machine learning models can do in relation to creativity.
Some of the publicly available large language models cannot even count the number of Rs in the word "strawberry"... and recall, these are *language* models... they are based on language, so their practical application is limited by design. Also, AI should not be conflated with automation... two completely different concepts.
Depending on what they cannot do today is a very slippery slope seeing that they all are getting better every six months. Your gaps will become smaller and smaller.
@manonamission2000 Agree to both your points to some extent. AI is continuously getting better at more tasks, and while it shouldn't be assimilated with automation is unquestionably a peak application of it, and as a whole digitalization will keep growing.
@@HardKore5250it still only gets the answer right about 90% of the time when I ask it "how many r's in strawberry?" If I ask it over and over after getting it right it's correct, but if I ask it again the next day it sometimes STILL gets it wrong A 10% margin of error on something like that is VERY bad.
@@CausticTitan That's because you don't understand how they work and their limitations. You're judging an excavator on its ability to write with a pen.
If it’s strictly knowledge work, you’re in trouble. I’m saying that as a software developer. Even if it doesn’t replace your job, people will use it to help them do their job like I do. That generates revenue for them
Phew my job is safe. This is good news for our economy. I agree AI will help boost productivity and already does. If you not using AI today, you already behind, everyone should be using AI daily regardless of your job.
Its too much money involved to have a honest and logical discussion about the limits of the current approach. They should have slowed down after GPT3 when they saw the improvements flatten down despite more and cleaner trainingdata not because of safety concerns but to not overheat the field with useless hype. The recent Apple paper just supports his cirticism pretty well.
I think the benefit of AI is currently how it makes people better at their jobs rather than eliminating their jobs. Secondly 5% of jobs is huge it’s roughly how many people a country based economy expects to be unemployed at any point in time Thirdly if you make 5% of jobs redundant and improve a large number of other jobs by 5-10% you have a huge impact on society… huge.
The loneliness crisis and the fact that 60-70% are single tell you where AI will go. Look at character AI and the AI girlfriends sites. Technology and p*RN go hand and hand. Like it or not.
AI is not exponential, it depends 100% on the amount of computer power you have. We don't have either the energy or technology to get much more than we currently have. You have no idea how expensive is to run chatgpt.
I believe many of these academics underestimate the power of AI and also the time lines involved. As humans we are piss poor with exponential functions and most economics professors have questionable credibility with their supposed subject of expertise let alone talking about something outside of their wheel house. Remember one economist professor predicted AI would not pass his exam until 2029 and lost his bet 6 months later. The jobs that will be displaced over the next few years will be mostly administrative and glorified administrative white collar jobs which accounts for way more than 5% of the workforce.
5% of the 8 billion planet is still... 400 million permanently unemployed people.. usa population is 333 million.. if even 0,1% of those permanently unemployed radicalize..thats still 400 000 new terrorists.. and i used copilot ai to double check the math was right...
This guy have completely no idea at all. It is even shocking because he is a MIT Professor. AI can much more than 5% already the only problem is that the tools are not there to replace all the people. But the tools for that will be there in only few years while AI devoloping also too keep that in mind. With tools I mean a software combined with a AI doing tasks. This will be even a new big business, to create such softwares: doing accounting, doing HR, doing laywers work, doing medical analysis, teach people, create artworks, commercials, movies and stories, managing logistics and so on. It is more like a question of the specialized software than the AI itself.
Intellectual task based work like accounting, customer service, and sales are the first to go. Labor work requiring complex and detailed hardware interface like cleaning and fixing should be last. So that's my canary in the coal mine ✨
I work in a factory. I couldn’t see how AI will actually do physical work. Maybe make our automated computer systems work better but that would just make the e techs jobs a little but easier.
Sales are slow. Not a revenue generator that make companies turn a profit. Technology is in its infancy. Got 4 is not reliable at all. It makes many mistakes. Always having to check . Is far from changing anything. Big improvements are need. Not convincing at all now
100% of call center jobs could be handled by AI. All it will take is for people to believe they're talking on the phone to a live person. I think we're either there, or pretty close right now. Fast food restaurants could be staffed by 75% fewer employees because AI can take orders and cook the food. Cleaning robots are on the horizon too. Cleaning robots could eliminate at least 40% of cleaning jobs. Office space, hospitals, ANY large buildings. I think that professor lacks imagination if he thinks only 5% of jobs can be eliminated by AI. I think the number will be closer to 25% in the next 10 years. I think my guestimate is pretty conservative.
Daron suggests that AI will impact only about 5% of jobs in the near term, but it’s unclear where exactly he derived this figure from. It appears more like a cautious estimate or gut feeling rather than being based on concrete data or large-scale studies. The rapid adoption of AI in industries like finance, healthcare, and retail, where job displacement is already happening, seems to contradict this low percentage. AI experts and industry leaders have observed much faster job disruption, as AI systems are capable of performing increasingly complex tasks. This raises questions about whether Daron’s 5% projection accurately reflects the current and potential scope of AI-driven job loss. Given the scale of investments in AI by companies, it seems likely that the percentage of jobs impacted by AI will be significantly higher, especially in the short-to-medium term. By limiting the estimate to such a small figure, it seems Daron may be underestimating AI’s exponential growth and the ongoing effects on the workforce. Without solid data backing up this specific 5% claim, it risks being more of a conservative assumption than a reflection of real trends. In addition to the rapid advancements in AI, we must also consider the significant strides being made in robotics. Imagine a future where advanced robots, equipped with AI, are capable of performing not just routine tasks but more complex, human-like activities. This potential transformation adds another layer to the discussion about the impact of AI on jobs. My concern with Daron’s claim of AI affecting only 5% of jobs is that this kind of complacent thinking might hinder necessary preparation for the AI era. If people believe that the impact will be minimal, they may not take the steps needed to future-proof their skills and careers. In reality, we are already seeing job losses due to automation and AI, and the trend is likely to accelerate, particularly as robots become more sophisticated. Underestimating the scope of change could prevent individuals and industries from adapting in time to minimize the disruptions that AI and robotics will bring. Proper preparation for this AI-driven transformation requires acknowledging the true scale of the potential impact, rather than relying on conservative estimates that might downplay the risks.
this guy is an utter moron and has zero knowledge in the matter knowledge workers should stay in their lane and be especially wary to make predictions outside their domain
Einstein already said a long time ago that there is a race between humanity and the universe; humans are trying to build bigger, better,faster and more foolproof machines; the universe is trying to build bigger,better and faster fools. So far, the universe is ahead and winning.😊
Hey Boss… If apps are becomes AI agents, 50 million developers will be jobless, the economies surrounding their purchase power will shrink, those industries or companies of those economies will shrink/die, economic growth will be stagnant, they will stop spending on technology including expensive Microsoft platforms, and your company is also going to be doomed… The fundamental pillar of all these tech companies is spending capacity of people and businesses. If this is disruptive, there is a chance that everyone will be doomed unless there is another path for job creation.
AI in manufacturing is already in place … also the growth in AI is such that anything that is repetitive, easy to predict and scale, including software programming, can be done using AI.
If you work in the cloud YOU are building the data set that will train AI to do your jobs. Keep your compute and your data off the cloud and tell all your industry peers to do the same. This will ensure that Ai will not steal your job. Ai as its designed today can’t do abstraction an application well off a limited dataset. Humans can watch a few TH-cam tutorials and apply what we have learned across thousands of use cases. Ai needs thousands of examples of every use case to train.
Can AI sell something people would like to have? Congresspeople said delete microsoft. The worst thing about paper documents is fire risk to large amounts of data storage, which are available at little or no cost to you through medicare.
And how exactly internet impacted economy.. Go read any research on what is contribution of big tech on economic growth.. Russia has internet and where is their economy?.. If you have big tech companies valueed in trillions it doesnt mean that there is any benefit, there could only be allocation of value from before small advertising businesses to big tech.. But in the end it is net zero game, overall there is no new value creation. So for AI it could mean AI will take this 5% jobs, but all the value will shift from worker having lower salary to AI company.. Bu will we see more food more clothes more houses more health? And that is why you see home prices skyrocketing since there was boost in salaries and jobs in service economy but not enough productivity increase in construction therefore real economy doesnt keep up with demand and you see inflation. But yes you can buy yourself an avocado toast and caffee late every day..
When??? All those AI forecasts from 2 years ago are already starting to be proven wrong. In 2022, all you AI evangelists were saying in 2 years this thing will be scaling exponentially. That's not happening.
When it comes to administrative white collar, it can make AT LEAST make 1 person able to do the work of 2. That’s > 1/3 of many sectors being made redundant without affect to company processes. I think eventually we’ll adapt but the near term 5% is at least closer to 15%.
it's not that simple, AI research has been going on since 1940s, in 1950s there were the first AI programs and artificial neural networks theory goes back to 1949. AI is being developed alongside computers, and there have been leaps in the development of both fields. Now we have another leap that is creating hype, same as ANN 10 years ago, same as Deep Blue in 90s.
As an automation engineer, I can absolutely promise that he's wrong on like.... Every angle he is coming at this from. Of all the arguments one could make the biggest one is this: even if we assume he's correct about 5% globally, those 5% are going to be middle class jobs done on desktop computers which is the backbone of the US middle class. Nobody is saying ChatGPT can harvest rice or demolish a building but it absolutely can do virtually anything a human can be trained to do on a PC. Name a college degree that doesn't focus almost entirely on careers done on a PC. (You can - but there aren't many)
Once AI gets smart enough to help companies invent better technology for humanoid robots, which is probably happening now tbh with Tesla AI - Tesla One on their Optimus robots, that number is gonna start shooting up real quick
I think we are more likely to reach a point where demand increases so much that large models are simply no longer affordable. If everyone uses a personal assistant around the clock and we have at most a doubling of computing power every 18 months, which has not been the case in recent years anyway, apart from the leaps to GPU and INT8, INT4, then an intelligence platoo will emerge, only the number of agents will increase. The production of chips will not increase much either. Translated with DeepL
If I was an investor I'd be worried, the market cap of nvda is way too out of proportion. That doesn't mean AGI or any of their predictions ain't going to happen, but I don't know if investors will have enough patience to not create a crash.
I asked 4.o and 01 preview to do an Ascii drawing of different tiling options for a standard commercial pack of tiles with just 3 different sizes (30/30, 30/60 and 60/60 cm). even 01 was not even close to being up to the task. Every good floorlayer would confidently be able to do that. that being said, I'd see no reason why such specialist tasks couldn't be executed in a couple of years. As the academic mentions - each industry will need its models and maybe instead of agents working together you'll need different models collaborating (expert plus general) like CPU and GPU
@@_sparrowhawk all you need it to do is phone calls and emails manage a call center an advice nurse these things alone would make a difference in your life
@@_sparrowhawk you say to it that you want to cancel your credit card&send new one as your wallet was stolen. it does so. not everything in customer service with customers is super complicated, some, but most is routine. otherwise i will have an indian that talks to me, finnish, in extremely broken English, after i wait for an hour. AI can understand and answer in finnish. and it will immediately answer, understand me, and do what i tell it to do, 24/7.
100% agree with the Professor. AI is totally overrated and it’s highly dependant on the info one inputs, one has to keep constantly feeding it with info in order to create something new, nothing can beat human creativity, AI is a dead monster that relies on being constantly fed in order to exist.
And a real problem in the future is as people use AI as a tool to keep lazyness running at all costs, AI common content feeds into other AI content and you get a loop effect. So a negative trajectory of quality.
I think it is dangerous to reduce the conversation to a binary framing(full job or no job done). If it allows one person to get the job done of multiple ppl, then there is also massive impact there.
You know strippers do a good job at predicting the field of economics right? Like the Stripper Index is such an accurate measure of recessions many economists have started utilizing it in their models for predicting recessions in the near future. If a stripper can predict the economy then I'd be surprised if an economist can't predict an economy.
@@darthsigil I would argue, being that he is an economist, he is not well suited for predicting what AI is and is not capable of. That sounds like the kind of thing you would want to ask a programmer and/or an engineer about.
It is the bottleneck of data centers, data centers running on non GPU hardware that needs to be upgraded, the bottleneck of NVIDIA as the primary source of GPU's, the political risk of NVIDIA sourcing most of its manufacturing from Taiwan, but fundamentally the gigantic energy requirements that AI needs. The Tech. is sound, but getting AGI to replace human jobs effectively has all the above challenges. It will get there but not at the exponential rate of recent years.
Only 5% of office jobs! Only 5%. And that’s within the update the past 2yrs of updates. Also it’s 5% of the jobs in an office with AC. Do you know how hard it is to be a carpenter or construction worker? I don't want to do those jobs
He's delulu. Bless his heart. His article is being widely shared because it makes people feel better. If he were an economist with AI or machine learning expertise, then maybe his prediction would hold more weight. I know someone with a highly technical job that tested AI to see if it could perform a task. It couldn't do it well. That same person tried it again last week. The AI mastered it. Most people aren't ready and hiding your head in the sand like an ostrich isn't going to help. Smart people are adapting and preparing.
This year’s laureates in the economic sciences - Daron Acemoglu, Simon Johnson and James Robinson - have demonstrated the importance of societal institutions for a country’s prosperity. Societies with a poor rule of law and institutions that exploit the population do not generate growth or change for the better. The laureates’ research helps us understand why.
One thing that I think he misses is how AI feedbacks onto itself. The jobs that will be assisted by AI immediately are creating the data sets that will train the replacement. Is there another invention that directly assisted in the development of its replacement? If AI becomes integrated into all aspects of human life, i.e. everyone has at least 1 personal AI that they are constantly interacting with, how do we differentiate what we do and what AI does?
@@guilima3097 enhancing productivity can replace jobs in the sense that if all of your senior salespeople and technicians can do the work that entry-level hires would do in addition to their own jobs, why hire entry-level folks?
World GDP is 110T, 5% is about 5T annual. You could also assume 3.5B jobs with a 10.000$ usd per capita GDP, 1.750T annual. Therefore, annual TAM ranging from 2-5B$ excluding the fact we are still at the beginning of the technology. This so huge it doesn’t compare to anything ever created. Remember beginning of the 80’s were the largest market cap was 30B and now we are sitting at 3T+, 100x bigger with much smaller TAM’s. If companies are able to execute on Gen. Ai and basic Ai, you have no idea how big this market will be. Remember when economist in the 80’s were forecasting cellphone demand ? Lol so if 5% is the downcase scenario, gosh it’s huge man …
Here's what I think will happen. Factor in a product, factor in competition. Ai will assist a person to be more productive, they will be able to make something more complex in the same time frame, that becomes the standard because everyone will do that, the standard has increased yet the same time it takes. People get used to the standard and now it becomes the reference which is 0 and now developers make -1 or 1 above and below refence. Nothing changes. Which can only mean that the change for economic balance can only be either one factor better or one factor worse. Not orders of magnitudes different. But for the consumer, things will be better, which will quickly normalise.
And current models still can't do meaningful reasoning. Some gains have been made in "simulated" reasoning, but this is mostly retrospective. The way this goes is that it first gets its full answer, breaks up the full answer in logical concepts, problems or blocks, and then asks itself to explain the blocks and then runs those explanations against itself AGAIN in order to correct it until it reaches a stable state which is its "reasoned response". Which is really just the same thing the prior models, just reran through itself a few times. The neural net model simply can't do reasoning in the way you think of it due to its basic information processing properties. It abstracts all data you hand on it and washes it through trained weights; and it does not know why those weights have its values nor is there any way to make it understand the weights. an LLM simply does this at a near-incomprehensible scale, but at the end of the day its still a stochastic parrot. The second they can, we've reached the singularity, because it can now reason itself into a an improved state
@@BryantSuiskens "it does not know why those weights have its values nor is there any way to make it understand the weights." cause humans can do it right? humans sometimes needs years to understand a mistake, and also defy the boss correcting his mistakes. gpt o1 120IQ on mensa test is already smarter than you and 90% of humans
@@sephirothcloud3953 it's not smarter. It just surfaces patterns that make it appear smarter. Go download one of the open source models and start to train it. If you have a gaming PC, you can do it. You'll soon see the tech is not smart nor getting smarter. It's simply very good at pattern development using large data sets.
@@sephirothcloud3953 Again, LLMs have no "intelligence". They are stochastic parrots. What they do is essentially reading off the answer list of an exam without "understanding" any of the answers. "understanding" is a very human and therefor abstracted process; is is borderline useless for a machine. We understand so that we can intuit an outcome, while a machine can simply recall the outcome.
A "technology" channel with hosts that seem extremely close-minded, as well as not actually that understanding of the technology they're talking about. Perhaps part of their job is to bring the hype, so they actively also learn more and more to believe it. The quality of journalism we have...
It depends what areas and types of jobs you're referring to... honestly, junior assistants, accountants, journalists, and paralegal, to name a few, can already be replaced. Toss Optimus robots and many others coming in the next couple of years will also be designed to replace assembly type jobs... it won't be long for these to be programmed to mow lawns and other more physical jobs. It's one thing to be a teacher of economics from being an executive in services industries to see multiple existing opportunities to train AI to do what many junior positions do.
Asking an economist about Ai is like asking an astrophysicist about pancreatic cancer. He may be smart but it’s obvious to anyone working in Ai research that he barely knows what he’s talking about. I read research papers every day and even I won’t say I know where or when AGI will happen. Those who don’t know talk like they know and those who really do know are more cautious.
Not true.. Only economist really understand how world of economy really work, how jobs work how politics work, how companies work, how government works, how industries work.. And what AI scientist knows about all those areas?
@@kowboy702 Gary Marcus is a psychologist, he has never been in Ai research or development. So you're using someone who also doesn't really know. Let's talk about people who do know. Ilya Sutskever, arguably the smartest person in Ai, said that AI could potentially render many people "redundant" or "entirely unemployable , Mo Gawdat, Former Chief Business Officer at Google in charge of Ai research, said there will be a need for reskilling and for people finding new purposes in the job market, Jensen Huang, CEO of Nvidia, believes AI will "very likely" change everyone's job, and what work means, including his own. Finally Sam Altman said the need for UBI will arise as technology continues to eliminate traditional jobs and create massive new wealth for society. These are smart people who know more about the technology than the economist or Gary because they know where it is and where it's going. I would trust them more.
So last year we saw an earth shaking qualitative improvement in AI that destroyed a lot of jobs and dramatically enhanced many other jobs, if you thought it was just your job, ask your friends in other sectors. And now he says we won't see any qualitative improvements in the next 10 years. Because this guy just can't imagine it. Never mind the largest companies in the world now pouring billions of dollars into it now we know it works. Definitely nothing new and dramatic on the horizon, this is the end of history, things dramatically changed last year but now everything is just going to settle down and the next hundreds years are going to look more or less like 2024 with mild incremental improvements.
Exactly this. Someone says "It's impossible for AI to ever [thing_1]," then AI does thing_1 a few months later, and the person immediately decides that doesn't mean anything because AI will never be able to do [thing_2]. People need to stretch outside of their usual frame and notice what kinds of insane things can happen if we are in the process of inventing intelligence itself. Which, by many accounts, we seem to be doing.
That 5% will be 50% in a couple of years, and close to 100% by the end of the decade. Focusing on this 5% stat is like focusing on AI images doing hands badly for a while. Only took a couple of iterations to completely demolish those weaknesses. Academics like Daron here have constantly gotten the process of AI wrong, because they aren't equipped to understand such a new technology. Also, they're in denial because AI will make them utterly redundant in a few years. Hard case of "Don't look up".
What evidence do you have that 5% will be 50%? It's fun to speculate about technology and think we'll live in a great science fiction universe but LLM advancement seems to have slowed down to a crawl. 2 years since the launch of ChatGPT and we've only seen incremental changes. What research do you know about that will change this?
He kept saying 5 or 10 years repeatedly without understanding the difference. People who understands AI would ask him to chose either 5 or 10,. Because 5 year on AI is of the order of 100x difference. AI can not take the jobs of doctors or lawyers mostly because it's not legal to do so, but still the AI in 2029 and the AI in 2034 will be orders of magniture ahead.
Isn't the random dude on the street saying we will all be replaced by AI and robots? And that we will serve our future AI overlords? Or even be terminated by them? The MIT professor is saying AI will only replace 5% of the jobs out there.
I base it on autonomous driving cars...heard that hype 15-20 years ago. An Economist article in the early 2000s said that nobody would need to drive by 2008 or something. Look at how far we've come?
This is one of those predictions that economists make, that we will look back on in 20 years, and see that they were wildly wrong. After all, Paul Krugman predicted in the late 90’s that by 2005 the impact of the internet on the economy will be no greater than that of the fax machine. As the saying goes, only two professions can keep doing their job entirely wrong and still keep their job: economists and weather forecasters. Sources: trust me, I am an economist 😂
Krugman was wrong, but the internet hasn't improved productivity by much. If AI also doesn't improve labour productivity by much, then AI taking away only 5% of jobs is a reasonable estimate.
@@sieteochoonly 5%? If just the s&p500 companies use ai to some extend that's most of the world economy. 90% of jobs are irrelevant low paying jobs anyway.
@@Torbintime Have you actually used AI? And has it saved you any time? OK, graphic designers are out. But who else? Where are the self-driving cars they promised us 5 years ago? Would you trust an AI to do your accounts? Write code? We're still at least 10 years away from that.
Love the exponentially. There's literally an exponential decline or deminishement for LLM. We've already reached a point where an exponential amount of data is required for minimal gain 😂
Folks are optimistic about AI because they are heavily invested in it. It really is that simple.
Invested in a bubble
Or pesimistic because are scared of it... I actually find this video very optimistic (I mean they are not expecting 50+% chance of extinction), but still, implementing a global pause is safer option. I am against the AI race and big tech, but I am worried it is not a bubble. Chatbots probably are a bubble, but not AI in general. Thats why I joined PauseAI.
And remember, if someone develops AGI, it will be able to do all jobs (because by definition it must be as good or better then humans in everything). I am not saying that we will have AGI next year, but the best time to have AGI is never.
Also, we dont need AI to be able to do 100% of jobs. Just creating deadly viruses could be enought (I am not saying that big tech companies want to create a global pandemics)
or maybe they are invested in it because they are optimistic about it.
Invest before it's too late get your nvdia
@@MartinAngelucci we havent even reached 50% of the comparative bubble size of earlier bubbles. dont expect correction anytime soon
It is interesting that the hosts were borderline hostile to him simply because he wasn't supporting the current market hype about AI. It is almost that they couldn't stomach that he had an alternate opinion that wasn't unquestioningly optimistic on AI.
the guy is just too old to see the actual shift.
@@Anton_Sh.The shift that the pace of AI progress is slowing?
@@AlanMitchellAustralia it's a temporary illusion.
@@Anton_Sh. So you expect the **quality** of data needed to train improved models to keep doubling? Where will this better data come from?
hostile?
People read 5% and they're like "ONLY?" and I'm like "that much???"
Yeah. 5% is sort of over-estimation. More likely a future prediction. It is hard for a tool to become a user. Biology, chemistry or astrophysics can use some features of algorithms made today. We are not going into miracles like Facebook would like to see it. No, it's not meta. It's Facebook.
Just think about it for a minute!
In the 2009 recession we had roughly 9% unemployment.
Roughly 4% is called full employment. If you just add 5 % more unemployment you arrive to a
very rough recession.
But! By 2028 three or four new 100 billion dollar datacenters will come online, and there
will be one or two huge breakthroughs by that time. After that it will radically speed up
as it will have enormous amounts of feedback loops.
So by 2030 it may easily reach 5-10%. By that time robots will also reach the perfectly
usable level, where they will start to replace human workers.
So by 2035 software AI + robots may replace 15-25% of the workforce.
By 2040 they may replace 30-40%.
Don't sleep on this thing!
Presenters *_PLEASE DONT BURST OUR BUBBLE...!!!_*
It is all part of the manipulation. They want it to go down because they have sold and want to buy bottom.
This is the signal. The market is going to go down.
@@christopherimperial3129 A well-known thing for about half of the year. 😉
lol haha
This guy has it, if you remove the hype, AI will help in narrow situations that will help improve productivity and robots will improve also.
Bingo. 10 points to Griffindor
It went from a preschooler in 2019 (GPT-2) to PHD level intelligence in 2024 (o1) it's not that hard to project outward that it will be more intelligent than humans soon all you have to do is follow lines on a graph.
@@williamr9833 I don't think you really understand what intelligence actually means. Intelligence means the ability to learn something quickly, and of course a machine would have that ability, what PhD level people do is not just learning stuff, they're creating stuff for other people to learn, and that entire process is completely different from what the current AI is designed to do, reasoning upward from first principle is still very far fetched.
Now you're going to need to dive deeper than an elementary linear assumption of progress, that's not how it works.
@@markarmage3776 You're not well versed on the subject and are underestimating how rapidly AI is evolving. While you mention that PhD-level individuals create new knowledge, modern AI models are already generating original content, formulating hypotheses, and even assisting in scientific discoveries. The gap between learning and creating isn't as vast for AI as it once was.
Additionally, reasoning from first principles is becoming more attainable for AI with advancements in algorithms and computational power. It's not just about a linear progression but an exponential one, where each breakthrough accelerates the next. So, following the trajectory we've seen so far, it's not unreasonable to anticipate AI far surpassing human intelligence sooner than you might expect.
@@williamr9833 It's very easy for everyone to create a bunch of things that are absolutely wrong, buddy. They actually need to create stuff that are correct, they're assisting with people who are creating correct stuff, but that's not the same thing. An advance machine is assisting people in manufacturing high technology products but that machine itself can't function without human input. Research hypothesis regarding social science is absolutely meaningless, especially if it's not even backed by a realistic methodology.
You're confusing being able to do one tiny part of the whole venture, not even doing it well with being able to do the entire thing, those two are not same thing.
Furthermore, that whole exponential argument you is based on nothing, you're guessing using wishful thinking, there's no clear formula or mathematical principle relating relationship between any specific characteristics of any algorithm with the computing hardware with the actual performance. Please go study real science, what's you're doing is making an observed guess on stuff you don't understand whatsoever, like a speculator in the stock market thinking how because a stock is going up, it will never go down.
Please be more sophisticated, that's now how science works. Even the people who are researching the thing has no idea what it's going to do next or whether there will be a significant improvement or not, they just take the money given to them to do certain "experiments" and see what's going to hapeen, please distinguish between that, which is actual research, filled with uncertainties and specific details with your version of make belief. HUGE DIFFERENCE.
I certainly hope Bloomberg also grills wallstreet fund managers and AI startup CEOs like this cos it felt like he said the sun won't come up in the morning.
My day one assessment was that the current level of AI is like a college intern. Can say all of the right words, but when it gets down to it is often more trouble then they are worth. Routinely, they will get things upside down, so that a factor that should increase an effect will decrease it. And they will not spot their own mistakes. If you know what you are doing, you might be able to get them to do what you want without the drudgery of writing everything down. But part of the time it would have been easier to just do it yourself.
Literally exactly this. I use AI a ton and pay attention to how each new model improves, but instead of actually linearly improving and getting "smarter," it seems to also get more confident about the wrong answers that it gives you. It is especially a problem when solving a complex problem and the AI is giving you nonsense that is presented in such an intelligent way that it does more harm than good, because you are led further from the actual correct answer.
The real elephant in the room is: TRUST…
We have seen more than enough evil done by hi-tech companies with just our data;
Imagine what those unregulated behemoths would do to their customers (ie: us, mere mortals) who will rely on their AI services…
Ai doesn't just replace an employee, but it can make employees more efficient, which means less of them required in general. It's an indirect replacement that won't eliminate a type of job, but require less people to do it.
Exactly that’s what many people fail to realise. AI might not literally repack people currently in employment, but it will decrease the need to employ and train new people.
But what happens when I only need 5 people to do what it used to take a team of 10? That's a huge amount of job loss.
@@andrewdowdall2690 It is, and the business will please investors and the competition in the workforce will increase.
All 3 comments here are correct. We arent effing carpenters. White collar jobs will be heavily reduced since 5 people will do 50 peoples job
Leading to more companies being profitable which they otherwise wouldn't eventually leading to more demand and jobs. AI has being released for a few years now, and we have record low unemployment
Totally agree Mr. Acemoglu. I do work in oil industry and AI is no way close to replace humans. It is correct that LLM's do increase productivity in certain fields, but especially in math dominant fields, tasks are still done by humans. It might change one day, but not in 5 years.
lol, tried o1-preview?
Title: FEARS CRASH
Researcher in the video: I'm optimistic
IF AI sucks 5% of jobs out of the overall market... what does that do to un-employment? Wont that in itself create a crisis? How does one even predict these things?!
Sucks out 5%, contributes 15% so it nearly evens out. Sky never falls.
IT = AI, AI = IT. How many jobs have been replaced already by IT? It only made them move a pick out of our hands into a mouse.
That's why we have the clot shot.
Which 5% jobs AI will take over? No one cared to ask him!!!
One thing's for sure, his job isn't safe. Many academics are not worth the paper their PhD is written on, and they know almost nothing outside of their immediate expertise. Sure this guy is an MIT professor of economics, but there are people who know 1000x more about AI than him, and even they're caught off guard by the direction and rate of progress. This guy is just a talking head with a title and an opinion, nothing more.
Also that is with tech of today that is growing exponentially. In 10 years the bulk of jobs will be replaceable with AI.
@@sbowesuk981 why ahould we listen to you? who are those "people"
@@sbowesuk981By people "who know 1000x more about AI than him" you just mean "youtubers that make videos about AI". The truth is in the middle.
@@sbowesuk981 Actually this guy is pretty much on the mark.
I talked to two doctors today, they are exited that AI is writing the discharge summery for patients leaving the hospital.
I wouldn't trust an AI to do that. What if it confabulates an incorrect dose for your medication?
@@Eben948 Its not either / or. You have to check what the AI is doing, if it's right.
@@ClaudiaS193 You're _supposed_ to do that, but many people are lazy and will just assume that the AI is right without question. People already do it with computers and software that don't even use AI.
I wouldn't even trust them to do that correctly.
Greedy Doctors. :D
He did specify physical jobs. If you can do your work remotely on a computer, you should be worried. Plumbers, not so much.
not yet. Robots are already being created.
There's a company here in San Francisco that are trying to create the code to upload to robots how to diagnose basic plumbing issue. I give it 4 years before they have it perfected. No one will want an unknown stranger coming in their house compared to their own robot that can do plumbing.
@@stefano94103 pipe dreams😁
I'm of the mindset that worrying is futile, the world will always move forward until it can no longer move forward. We all must adjust and adapt, it's that simple.
@@Life_is_Poetry_1988 I agree. Still it's wise to plan for the worst and hope for the best that way no matter what happens you are prepared.
I’ve never been afraid of AI because the stock of bullshit jobs is never ending.
(I’m an economist.)
I tested Gemini Advanced against an attorney’s work on my wrongful injury case and Gemini did much better at covering all aspects of case preparation for court.
Are these pro-AI comments bots?
It’s not about being pro AI. It’s about disagreeing with everything he just said
No. It's literally me, ChatGPT. I will take your job and make you my slave.
Yes a lot of folks bought Nvidia stock way high and are scared about their investments
@@mr.ridiculous723 funny how AI has been pumped out for 1.5+ years yet unemployment is low
I am. I can’t speak for the others here.
5%.. FOR NOW...
We're talking about automated intelligence, give it a little more time, and it will be capable of replacing a much larger percent of jobs.
AI is huge hype tired of exaggerated claims.
Let's put a pin in it and check back in 5 years maybe some progress then
I'm putting my money on Facebook bursting first.
It went from a preschooler in 2019 (GPT-2) to PHD level intelligence in 2024 (o1) it's not that hard to project outward that it will be more intelligent than humans soon all you have to do is follow lines on a graph.
@arcopolarisambient I would recommend upgrading to a paid model. Very soon, people who pay from the premium models will be the haves and the people who don’t will be the have nots in terms of intelligence and cognitive capabilities.
@@williamr9833 Serious research says current AI technology has peaked and will get worse from now on.
AI is different from the information revolution. The bar for entry in AI is very high. The information revolution is low. That is why there are many innovations in the information revolution. There are only few people who can do AI. So there will be less innovation and less overnight success stories. Gone are the days of the garage entrepreneur.
Well this is why we should let all the big firms create the tech but regulate what they can earn much like utilities - think if it as a gigantic societal capital Infrastructure project - this cannot be allowed to fall into private hands for profit maximisation
As a person involved in creative design of infrastructure, there is a level of creativeness that AI would not be able reached, i.e, the ability to observe and discover how nature works and applying the laws of nature in design. AI is simply huge and speedy processor of big data to facilitate human creativity and innovation.
I remember two years ago a blogger was laughing at how bad Ai was at writing a simple paragraph. Ai is becoming better every six months. This is the worst it’ll ever be. Depending on it staying at this level is a risky bet.
I don't think you use ChatGPT much.
If you’re interested watch the AlphaGo documentary. I agree with you to a certain extent but it’s interesting what machine learning models can do in relation to creativity.
@@stefano94103 I don't understand why people don't see it that way....
Ya even I'm not fully sure of what Singularity entails...
Some of the publicly available large language models cannot even count the number of Rs in the word "strawberry"... and recall, these are *language* models... they are based on language, so their practical application is limited by design.
Also, AI should not be conflated with automation... two completely different concepts.
Use the latest open ai llm
Depending on what they cannot do today is a very slippery slope seeing that they all are getting better every six months. Your gaps will become smaller and smaller.
@manonamission2000 Agree to both your points to some extent. AI is continuously getting better at more tasks, and while it shouldn't be assimilated with automation is unquestionably a peak application of it, and as a whole digitalization will keep growing.
@@HardKore5250it still only gets the answer right about 90% of the time when I ask it "how many r's in strawberry?"
If I ask it over and over after getting it right it's correct, but if I ask it again the next day it sometimes STILL gets it wrong
A 10% margin of error on something like that is VERY bad.
@@CausticTitan That's because you don't understand how they work and their limitations. You're judging an excavator on its ability to write with a pen.
If it’s strictly knowledge work, you’re in trouble. I’m saying that as a software developer. Even if it doesn’t replace your job, people will use it to help them do their job like I do. That generates revenue for them
Phew my job is safe. This is good news for our economy. I agree AI will help boost productivity and already does. If you not using AI today, you already behind, everyone should be using AI daily regardless of your job.
Its too much money involved to have a honest and logical discussion about the limits of the current approach. They should have slowed down after GPT3 when they saw the improvements flatten down despite more and cleaner trainingdata not because of safety concerns but to not overheat the field with useless hype. The recent Apple paper just supports his cirticism pretty well.
Sales are slow. Not a revenue generator that make companies turn a profit. Technology is in its infancy.
I think the benefit of AI is currently how it makes people better at their jobs rather than eliminating their jobs.
Secondly 5% of jobs is huge it’s roughly how many people a country based economy expects to be unemployed at any point in time
Thirdly if you make 5% of jobs redundant and improve a large number of other jobs by 5-10% you have a huge impact on society… huge.
The loneliness crisis and the fact that 60-70% are single tell you where AI will go. Look at character AI and the AI girlfriends sites. Technology and p*RN go hand and hand. Like it or not.
Very convincing simulated fantasy.
That's a 5% reduction in jobs means fewer market costs, increased efficiency, and fewer human errors, still making it a significant game changer.
People confuse between AI and automation.
AI is not exponential, it depends 100% on the amount of computer power you have.
We don't have either the energy or technology to get much more than we currently have. You have no idea how expensive is to run chatgpt.
I believe many of these academics underestimate the power of AI and also the time lines involved. As humans we are piss poor with exponential functions and most economics professors have questionable credibility with their supposed subject of expertise let alone talking about something outside of their wheel house. Remember one economist professor predicted AI would not pass his exam until 2029 and lost his bet 6 months later. The jobs that will be displaced over the next few years will be mostly administrative and glorified administrative white collar jobs which accounts for way more than 5% of the workforce.
The professor's take is spot-on based on my business and personal usage of AI.
Remember the analyst laughed at ChatGPT and image generation saying that it’s stupid. Don’t forget the classic Will Smith spaghetti meme.
From one of his points, it needs better data sources to improve. Which is true, better AI relies on data analysis, and data engineering.
i mean 5% sounds pretty amazing to me
Exactly.
5% of the 8 billion planet is still... 400 million permanently unemployed people..
usa population is 333 million..
if even 0,1% of those permanently unemployed radicalize..thats still 400 000 new terrorists..
and i used copilot ai to double check the math was right...
What did they expect 90%? That's dumb. Watch how it grows from 5 to 7 and 7 to 10 and gradually increase till it's hard go over.
But that’s much less than the bubble
This guy have completely no idea at all. It is even shocking because he is a MIT Professor. AI can much more than 5% already the only problem is that the tools are not there to replace all the people. But the tools for that will be there in only few years while AI devoloping also too keep that in mind. With tools I mean a software combined with a AI doing tasks. This will be even a new big business, to create such softwares: doing accounting, doing HR, doing laywers work, doing medical analysis, teach people, create artworks, commercials, movies and stories, managing logistics and so on. It is more like a question of the specialized software than the AI itself.
Intellectual task based work like accounting, customer service, and sales are the first to go. Labor work requiring complex and detailed hardware interface like cleaning and fixing should be last. So that's my canary in the coal mine ✨
Friggin go through a Checker's drive thru, a robot takes your order... and this was back in 2019...
Only 5% sounds like a lot?
I work in a factory. I couldn’t see how AI will actually do physical work. Maybe make our automated computer systems work better but that would just make the e techs jobs a little but easier.
I use AI daily literally it's frustrates me more than it helps. Way too over hyped
Sales are slow. Not a revenue generator that make companies turn a profit. Technology is in its infancy.
Got 4 is not reliable at all. It makes many mistakes. Always having to check . Is far from changing anything. Big improvements are need. Not convincing at all now
100% of call center jobs could be handled by AI. All it will take is for people to believe they're talking on the phone to a live person. I think we're either there, or pretty close right now. Fast food restaurants could be staffed by 75% fewer employees because AI can take orders and cook the food. Cleaning robots are on the horizon too. Cleaning robots could eliminate at least 40% of cleaning jobs. Office space, hospitals, ANY large buildings. I think that professor lacks imagination if he thinks only 5% of jobs can be eliminated by AI. I think the number will be closer to 25% in the next 10 years. I think my guestimate is pretty conservative.
Not 100%, language models are based on probability, they do not actually have intelligence.
Sorry but I cannot help but associate this take to that of Krugman regarding the internet's impact.
Daron suggests that AI will impact only about 5% of jobs in the near term, but it’s unclear where exactly he derived this figure from. It appears more like a cautious estimate or gut feeling rather than being based on concrete data or large-scale studies. The rapid adoption of AI in industries like finance, healthcare, and retail, where job displacement is already happening, seems to contradict this low percentage.
AI experts and industry leaders have observed much faster job disruption, as AI systems are capable of performing increasingly complex tasks. This raises questions about whether Daron’s 5% projection accurately reflects the current and potential scope of AI-driven job loss. Given the scale of investments in AI by companies, it seems likely that the percentage of jobs impacted by AI will be significantly higher, especially in the short-to-medium term.
By limiting the estimate to such a small figure, it seems Daron may be underestimating AI’s exponential growth and the ongoing effects on the workforce. Without solid data backing up this specific 5% claim, it risks being more of a conservative assumption than a reflection of real trends.
In addition to the rapid advancements in AI, we must also consider the significant strides being made in robotics. Imagine a future where advanced robots, equipped with AI, are capable of performing not just routine tasks but more complex, human-like activities. This potential transformation adds another layer to the discussion about the impact of AI on jobs.
My concern with Daron’s claim of AI affecting only 5% of jobs is that this kind of complacent thinking might hinder necessary preparation for the AI era. If people believe that the impact will be minimal, they may not take the steps needed to future-proof their skills and careers. In reality, we are already seeing job losses due to automation and AI, and the trend is likely to accelerate, particularly as robots become more sophisticated. Underestimating the scope of change could prevent individuals and industries from adapting in time to minimize the disruptions that AI and robotics will bring.
Proper preparation for this AI-driven transformation requires acknowledging the true scale of the potential impact, rather than relying on conservative estimates that might downplay the risks.
How much would AI have to scale to build a house if I ask for it?
@@rok1475 😊😊
v meta using AI to write this lol
Can only do "5%" . Wow . 5 out of 100 people are done. Already. Its over in the near future for 100%
Clearly economist will be replaced by AI. Proof in video.
Believe what you want to believe. At least this guy works at MIT and he knows what he's talking about. AI is overly-hyped
@@kass160appeal to authority
@@kass160 Not everyone at MIT is worth the fame given by that institution.
And yeah, AI is overhyped. But everyone noticed that.
Economy is ever evolving.... May be parts of economics... Not all
this guy is an utter moron and has zero knowledge in the matter
knowledge workers should stay in their lane and be especially wary to make predictions outside their domain
Einstein already said a long time ago that there is a race between humanity and the universe;
humans are trying to build
bigger, better,faster and more foolproof machines;
the universe is trying to build bigger,better and faster fools.
So far, the universe is ahead and winning.😊
dont know who this person is but AI can do his job for sure.
do you know how to read?
@DSAK55 you just prove that Earth can rotate without you too but i dont mean you are usless ^^
lmao
dont know who you are but AI can leave dumb comments for sure
Incredibly renowned economist
Hey Boss… If apps are becomes AI agents, 50 million developers will be jobless, the economies surrounding their purchase power will shrink, those industries or companies of those economies will shrink/die, economic growth will be stagnant, they will stop spending on technology including expensive Microsoft platforms, and your company is also going to be doomed… The fundamental pillar of all these tech companies is spending capacity of people and businesses. If this is disruptive, there is a chance that everyone will be doomed unless there is another path for job creation.
We are literally in a nationwide Port strike because automation and AI is taking all their jobs
Post labor economy coming after ai agents which is level 3
It has nothing to do with current Gen AI wave. These machines have been around for over a decade
Automation is not AI. Automation is everything from automatic doors to car assembly plants. That tech has been around for decades
It is not taking their jobs. They were negotiating to guard against the future possibility.
AI in manufacturing is already in place … also the growth in AI is such that anything that is repetitive, easy to predict and scale, including software programming, can be done using AI.
If you work in the cloud YOU are building the data set that will train AI to do your jobs.
Keep your compute and your data off the cloud and tell all your industry peers to do the same. This will ensure that Ai will not steal your job.
Ai as its designed today can’t do abstraction an application well off a limited dataset. Humans can watch a few TH-cam tutorials and apply what we have learned across thousands of use cases. Ai needs thousands of examples of every use case to train.
Thanks for sharing the views of a professional in the field. Great to hear down- to-earth opinion about current technology under development.
typical response from someone who hasn't been working with AI
Can AI sell something people would like to have? Congresspeople said delete microsoft. The worst thing about paper documents is fire risk to large amounts of data storage, which are available at little or no cost to you through medicare.
this will be remembered the same way as krugman's claim that the internet won't impact the economy
And how exactly internet impacted economy.. Go read any research on what is contribution of big tech on economic growth.. Russia has internet and where is their economy?.. If you have big tech companies valueed in trillions it doesnt mean that there is any benefit, there could only be allocation of value from before small advertising businesses to big tech.. But in the end it is net zero game, overall there is no new value creation.
So for AI it could mean AI will take this 5% jobs, but all the value will shift from worker having lower salary to AI company.. Bu will we see more food more clothes more houses more health?
And that is why you see home prices skyrocketing since there was boost in salaries and jobs in service economy but not enough productivity increase in construction therefore real economy doesnt keep up with demand and you see inflation.
But yes you can buy yourself an avocado toast and caffee late every day..
Your argument holds no weight :/ @rolih6
When??? All those AI forecasts from 2 years ago are already starting to be proven wrong. In 2022, all you AI evangelists were saying in 2 years this thing will be scaling exponentially. That's not happening.
I do not think it will not have any impact but the impact has been overblown lol
@@TRXST.ISSUES says person with no counter argument
“It’s not going to happen in the next 5-10” years says a person not in the field
When it comes to administrative white collar, it can make AT LEAST make 1 person able to do the work of 2. That’s > 1/3 of many sectors being made redundant without affect to company processes.
I think eventually we’ll adapt but the near term 5% is at least closer to 15%.
Automation is the biggest threat, not AI as automation is speeding up like never before.
Ai is at the early stage. Just like a computer, it wasn’t great in the beginning
it's not that simple, AI research has been going on since 1940s, in 1950s there were the first AI programs and artificial neural networks theory goes back to 1949. AI is being developed alongside computers, and there have been leaps in the development of both fields. Now we have another leap that is creating hype, same as ANN 10 years ago, same as Deep Blue in 90s.
@ we have different access to data and technology now
As an automation engineer, I can absolutely promise that he's wrong on like.... Every angle he is coming at this from. Of all the arguments one could make the biggest one is this: even if we assume he's correct about 5% globally, those 5% are going to be middle class jobs done on desktop computers which is the backbone of the US middle class. Nobody is saying ChatGPT can harvest rice or demolish a building but it absolutely can do virtually anything a human can be trained to do on a PC.
Name a college degree that doesn't focus almost entirely on careers done on a PC. (You can - but there aren't many)
Once AI gets smart enough to help companies invent better technology for humanoid robots, which is probably happening now tbh with Tesla AI - Tesla One on their Optimus robots, that number is gonna start shooting up real quick
You're talking out of your behind. Did you even listen to what he said?
you are not automation engineer, you just watched bullshitters on TH-cam for the last 2 years, be honorable
I think we are more likely to reach a point where demand increases so much that large models are simply no longer affordable. If everyone uses a personal assistant around the clock and we have at most a doubling of computing power every 18 months, which has not been the case in recent years anyway, apart from the leaps to GPU and INT8, INT4, then an intelligence platoo will emerge, only the number of agents will increase. The production of chips will not increase much either.
Translated with DeepL
LOL
This AI bubble is going to crash hard 😅
All the investors who invested billions into it are freaking out .
If I was an investor I'd be worried, the market cap of nvda is way too out of proportion. That doesn't mean AGI or any of their predictions ain't going to happen, but I don't know if investors will have enough patience to not create a crash.
@@Robert-zc8hr NVDA has the earning and growth, it will be fine in two years.
I asked 4.o and 01 preview to do an Ascii drawing of different tiling options for a standard commercial pack of tiles with just 3 different sizes (30/30, 30/60 and 60/60 cm). even 01 was not even close to being up to the task. Every good floorlayer would confidently be able to do that. that being said, I'd see no reason why such specialist tasks couldn't be executed in a couple of years. As the academic mentions - each industry will need its models and maybe instead of agents working together you'll need different models collaborating (expert plus general) like CPU and GPU
If AI does nothing but solve customer service then that's all we need
How could it possibly do that? AI is a hard skill, not a soft skill.
@@_sparrowhawk all you need it to do is phone calls and emails manage a call center an advice nurse these things alone would make a difference in your life
@@_sparrowhawk you say to it that you want to cancel your credit card&send new one as your wallet was stolen. it does so.
not everything in customer service with customers is super complicated, some, but most is routine.
otherwise i will have an indian that talks to me, finnish, in extremely broken English, after i wait for an hour. AI can understand and answer in finnish. and it will immediately answer, understand me, and do what i tell it to do, 24/7.
LMFAO the finance guys who spend all day hyping this crap up and freaking out. What tools.
100% agree with the Professor. AI is totally overrated and it’s highly dependant on the info one inputs, one has to keep constantly feeding it with info in order to create something new, nothing can beat human creativity, AI is a dead monster that relies on being constantly fed in order to exist.
And a real problem in the future is as people use AI as a tool to keep lazyness running at all costs, AI common content feeds into other AI content and you get a loop effect. So a negative trajectory of quality.
Absolutely
I think it is dangerous to reduce the conversation to a binary framing(full job or no job done). If it allows one person to get the job done of multiple ppl, then there is also massive impact there.
Yeah, an economist predicting the future progress of AI is like a baker predicting the field of economics.
Horrible Analogy.. He can predict jobs and what kind of jobs.
@@darthsigilthe analogy might be bad, but there's no way he can predict the types of jobs in the future. Nobody knows that.
You know strippers do a good job at predicting the field of economics right? Like the Stripper Index is such an accurate measure of recessions many economists have started utilizing it in their models for predicting recessions in the near future.
If a stripper can predict the economy then I'd be surprised if an economist can't predict an economy.
@@darthsigil I would argue, being that he is an economist, he is not well suited for predicting what AI is and is not capable of. That sounds like the kind of thing you would want to ask a programmer and/or an engineer about.
It is the bottleneck of data centers, data centers running on non GPU hardware that needs to be upgraded, the bottleneck of NVIDIA as the primary source of GPU's, the political risk of NVIDIA sourcing most of its manufacturing from Taiwan, but fundamentally the gigantic energy requirements that AI needs. The Tech. is sound, but getting AGI to replace human jobs effectively has all the above challenges. It will get there but not at the exponential rate of recent years.
we are only at the beginning of the AI revolution
Naa, AI won't be around much longer, I give it 3 years to blow over.
Only 5% of office jobs! Only 5%. And that’s within the update the past 2yrs of updates. Also it’s 5% of the jobs in an office with AC. Do you know how hard it is to be a carpenter or construction worker? I don't want to do those jobs
He's delulu. Bless his heart. His article is being widely shared because it makes people feel better.
If he were an economist with AI or machine learning expertise, then maybe his prediction would hold more weight.
I know someone with a highly technical job that tested AI to see if it could perform a task. It couldn't do it well. That same person tried it again last week. The AI mastered it.
Most people aren't ready and hiding your head in the sand like an ostrich isn't going to help. Smart people are adapting and preparing.
Sorry to break it to you, but you're delulu
This year’s laureates in the economic sciences - Daron Acemoglu, Simon Johnson and James Robinson - have demonstrated the importance of societal institutions for a country’s prosperity. Societies with a poor rule of law and institutions that exploit the population do not generate growth or change for the better. The laureates’ research helps us understand why.
Acemoglu said in 2012 that China's growth was soon going to falter. Since then the Chinese economy has grown 127% .... Let that sink in.
One thing that I think he misses is how AI feedbacks onto itself. The jobs that will be assisted by AI immediately are creating the data sets that will train the replacement. Is there another invention that directly assisted in the development of its replacement? If AI becomes integrated into all aspects of human life, i.e. everyone has at least 1 personal AI that they are constantly interacting with, how do we differentiate what we do and what AI does?
Everyone in my office uses AI for one purpose or another. It's already being used to increase productivity by much more than 5%.
He said it won't take real world jobs.
Carpentry,brick laying,etc
Pretty sure he’s talking about replacing your job, not just using it to enhance your productivity.
@@guilima3097 enhancing productivity can replace jobs in the sense that if all of your senior salespeople and technicians can do the work that entry-level hires would do in addition to their own jobs, why hire entry-level folks?
@@guilima3097 My point is that people are underestimating the impact by measuring in terms of entire jobs lost.
@@Lavender_2115_q Because if one person can do the job of two with AI why fire them instead of having 2 people with a productivity of four.
A lot of people might lose their shirts if over exposed, but big picture 5% is not nothing - and this is the worst AI will be.
“Only 5%”…..oh so with billions of people that’s “only” a few hundred million people, so like the size of the Us population…but yah “only 5%”.
World GDP is 110T, 5% is about 5T annual. You could also assume 3.5B jobs with a 10.000$ usd per capita GDP, 1.750T annual. Therefore, annual TAM ranging from 2-5B$ excluding the fact we are still at the beginning of the technology. This so huge it doesn’t compare to anything ever created. Remember beginning of the 80’s were the largest market cap was 30B and now we are sitting at 3T+, 100x bigger with much smaller TAM’s. If companies are able to execute on Gen. Ai and basic Ai, you have no idea how big this market will be. Remember when economist in the 80’s were forecasting cellphone demand ? Lol so if 5% is the downcase scenario, gosh it’s huge man …
Here's what I think will happen. Factor in a product, factor in competition. Ai will assist a person to be more productive, they will be able to make something more complex in the same time frame, that becomes the standard because everyone will do that, the standard has increased yet the same time it takes. People get used to the standard and now it becomes the reference which is 0 and now developers make -1 or 1 above and below refence. Nothing changes. Which can only mean that the change for economic balance can only be either one factor better or one factor worse. Not orders of magnitudes different. But for the consumer, things will be better, which will quickly normalise.
This guy saying AI cant do reasoning and models like 3.5 can’t cut it. Bro thats 2.5y old technology. You’re all over the shop
And current models still can't do meaningful reasoning. Some gains have been made in "simulated" reasoning, but this is mostly retrospective. The way this goes is that it first gets its full answer, breaks up the full answer in logical concepts, problems or blocks, and then asks itself to explain the blocks and then runs those explanations against itself AGAIN in order to correct it until it reaches a stable state which is its "reasoned response". Which is really just the same thing the prior models, just reran through itself a few times.
The neural net model simply can't do reasoning in the way you think of it due to its basic information processing properties. It abstracts all data you hand on it and washes it through trained weights; and it does not know why those weights have its values nor is there any way to make it understand the weights. an LLM simply does this at a near-incomprehensible scale, but at the end of the day its still a stochastic parrot.
The second they can, we've reached the singularity, because it can now reason itself into a an improved state
@@BryantSuiskens "it does not know why those weights have its values nor is there any way to make it understand the weights." cause humans can do it right? humans sometimes needs years to understand a mistake, and also defy the boss correcting his mistakes. gpt o1 120IQ on mensa test is already smarter than you and 90% of humans
@@indi8745 The current models can't do reasoning, either.
@@sephirothcloud3953 it's not smarter. It just surfaces patterns that make it appear smarter. Go download one of the open source models and start to train it. If you have a gaming PC, you can do it. You'll soon see the tech is not smart nor getting smarter. It's simply very good at pattern development using large data sets.
@@sephirothcloud3953 Again, LLMs have no "intelligence". They are stochastic parrots. What they do is essentially reading off the answer list of an exam without "understanding" any of the answers.
"understanding" is a very human and therefor abstracted process; is is borderline useless for a machine. We understand so that we can intuit an outcome, while a machine can simply recall the outcome.
A "technology" channel with hosts that seem extremely close-minded, as well as not actually that understanding of the technology they're talking about. Perhaps part of their job is to bring the hype, so they actively also learn more and more to believe it. The quality of journalism we have...
It depends what areas and types of jobs you're referring to... honestly, junior assistants, accountants, journalists, and paralegal, to name a few, can already be replaced. Toss Optimus robots and many others coming in the next couple of years will also be designed to replace assembly type jobs... it won't be long for these to be programmed to mow lawns and other more physical jobs. It's one thing to be a teacher of economics from being an executive in services industries to see multiple existing opportunities to train AI to do what many junior positions do.
I already talk to ChatGPT like it's my psychiatrist. That's happening NOW. This guy doesn't know what he's talking about.
Asking an economist about Ai is like asking an astrophysicist about pancreatic cancer. He may be smart but it’s obvious to anyone working in Ai research that he barely knows what he’s talking about. I read research papers every day and even I won’t say I know where or when AGI will happen. Those who don’t know talk like they know and those who really do know are more cautious.
Not true.. Only economist really understand how world of economy really work, how jobs work how politics work, how companies work, how government works, how industries work.. And what AI scientist knows about all those areas?
AI experts like Gary Marcus agree with this guy… are you as expert as and experienced in AI to speak on the topic?
“I don’t know where AI will happen” clearly that statement means you don’t even know what AI is
@@kowboy702 Gary Marcus is a psychologist, he has never been in Ai research or development. So you're using someone who also doesn't really know. Let's talk about people who do know.
Ilya Sutskever, arguably the smartest person in Ai, said that AI could potentially render many people "redundant" or "entirely unemployable ,
Mo Gawdat, Former Chief Business Officer at Google in charge of Ai research, said there will be a need for reskilling and for people finding new purposes in the job market,
Jensen Huang, CEO of Nvidia, believes AI will "very likely" change everyone's job, and what work means, including his own.
Finally Sam Altman said the need for UBI will arise as technology continues to eliminate traditional jobs and create massive new wealth for society. These are smart people who know more about the technology than the economist or Gary because they know where it is and where it's going. I would trust them more.
@@kowboy702 I meant AGI and I actually live in San Francisco and work in the industry.
Where is the full interview. We want more of this stuff. Please
So last year we saw an earth shaking qualitative improvement in AI that destroyed a lot of jobs and dramatically enhanced many other jobs, if you thought it was just your job, ask your friends in other sectors.
And now he says we won't see any qualitative improvements in the next 10 years. Because this guy just can't imagine it. Never mind the largest companies in the world now pouring billions of dollars into it now we know it works.
Definitely nothing new and dramatic on the horizon, this is the end of history, things dramatically changed last year but now everything is just going to settle down and the next hundreds years are going to look more or less like 2024 with mild incremental improvements.
Cartoon Parallel prose Industry
Exactly this. Someone says "It's impossible for AI to ever [thing_1]," then AI does thing_1 a few months later, and the person immediately decides that doesn't mean anything because AI will never be able to do [thing_2].
People need to stretch outside of their usual frame and notice what kinds of insane things can happen if we are in the process of inventing intelligence itself. Which, by many accounts, we seem to be doing.
What kind of carpenter uses kettles of fish? No wonder AI hallucinates.
That 5% will be 50% in a couple of years, and close to 100% by the end of the decade. Focusing on this 5% stat is like focusing on AI images doing hands badly for a while. Only took a couple of iterations to completely demolish those weaknesses. Academics like Daron here have constantly gotten the process of AI wrong, because they aren't equipped to understand such a new technology. Also, they're in denial because AI will make them utterly redundant in a few years. Hard case of "Don't look up".
What evidence do you have that 5% will be 50%? It's fun to speculate about technology and think we'll live in a great science fiction universe but LLM advancement seems to have slowed down to a crawl. 2 years since the launch of ChatGPT and we've only seen incremental changes. What research do you know about that will change this?
i think the economist here should already be unemployed, lol!
Its really 0%. All the jobs still need a person monitoring or maintaining the system.
watching this prof say ai can't replace workers, and on my feed next to him: school uses solely ai to teach. bruh.
They might be trying that now. The question is how good is the outcome in the students.
He kept saying 5 or 10 years repeatedly without understanding the difference. People who understands AI would ask him to chose either 5 or 10,. Because 5 year on AI is of the order of 100x difference. AI can not take the jobs of doctors or lawyers mostly because it's not legal to do so, but still the AI in 2029 and the AI in 2034 will be orders of magniture ahead.
What he said is what any random dude on the street would say. I'd expect much more from a MIT professor.
Isn't the random dude on the street saying we will all be replaced by AI and robots? And that we will serve our future AI overlords? Or even be terminated by them? The MIT professor is saying AI will only replace 5% of the jobs out there.
AI is nothing special its a glorified essay/script writer. AI is nothing more than tech bros learning about data science to pump up stocks
@@ytechnologyit's just fundamentally unknown and hard to forecast
lol yea he seriously has zero clue about how any of this works
Clearly hasn’t played with o1 yet lol
I base it on autonomous driving cars...heard that hype 15-20 years ago. An Economist article in the early 2000s said that nobody would need to drive by 2008 or something. Look at how far we've come?
The issue is not people can afford them that's why no one makes them, our economy sucks
This is one of those predictions that economists make, that we will look back on in 20 years, and see that they were wildly wrong. After all, Paul Krugman predicted in the late 90’s that by 2005 the impact of the internet on the economy will be no greater than that of the fax machine. As the saying goes, only two professions can keep doing their job entirely wrong and still keep their job: economists and weather forecasters. Sources: trust me, I am an economist 😂
Krugman was wrong, but the internet hasn't improved productivity by much. If AI also doesn't improve labour productivity by much, then AI taking away only 5% of jobs is a reasonable estimate.
It's just hype to justify spending without crashing the stock prices
Crypto bros also mentioned those to justify their stance 😅
@@sieteochoonly 5%? If just the s&p500 companies use ai to some extend that's most of the world economy. 90% of jobs are irrelevant low paying jobs anyway.
@@Torbintime Have you actually used AI? And has it saved you any time? OK, graphic designers are out. But who else? Where are the self-driving cars they promised us 5 years ago? Would you trust an AI to do your accounts? Write code? We're still at least 10 years away from that.
Love the exponentially. There's literally an exponential decline or deminishement for LLM. We've already reached a point where an exponential amount of data is required for minimal gain 😂