Whether it's based on hallucinatory beliefs or not, an artificial-intelligence gold rush has started over the last several months to mine the anticipated business opportunities from generative AI models like ChatGPT. Venture-backed startups, app developers, and large corporations around the globe are trying to understand the sensational OpenAI text-generating bot.
The screams can be heard from corners of offices all over the globe: "What's our ChatGPT game?" How can we make money from this?"
While executives and companies see the potential to make a lot of money, the impact of technology on workers and the economy is less clear. Although they have some limitations, including their propensity to make things up, ChatGPT and other recently released AI models generative to automate tasks. These tasks include writing, creating graphics and summarizing and analysing data. Economists aren't sure how this will affect jobs and overall productivity.
Despite all the incredible advances made in AI and other digital tools in the past decade, their failure to improve prosperity and stimulate widespread economic growth is disappointing. While some entrepreneurs and investors have made a lot of money, the majority of people have not. Some of them have been made redundant.
Since 2005, productivity growth has been a disaster in the US and other advanced economies. The UK is one of these basket cases. Many people are suffering from stagnant wages due to the fact that the economy isn't growing as much.
The productivity growth that has occurred in this time period is limited to certain sectors (e.g. information services) and to a handful of cities in the US (e.g. San Jose, San Francisco and Seattle).
ChatGPT will make the already troublesome income and wealth inequalities in the US and other countries worse. Could it be a boon? Could it actually provide a much-needed boost in productivity?
ChatGPT's human-like writing capabilities and OpenAI’s latest release DALL-E 2, that generates images on-demand, both use large language models that were trained from huge amounts of data. Similar results can be found for rivals like Claude from Anthropic or Bard from Google. These so-called foundational models such as GPT-3.0 from OpenAI on which ChatGPT is built or Google's LaMDA competing language model, which powers Bard have seen rapid development in recent years.
They continue to improve their power: they are trained on more data and the number parameters (variables in the models that can be tweaked) is increasing dramatically. OpenAI's latest version, GPT-4, was released earlier this month. OpenAI doesn't know how much larger it is but one can guess that GPT-3 with 175 billion parameters was 100 times more than GPT-2.
ChatGPT's release late last year was what changed everything. It is simple to use and can quickly create text that looks human-like, including workout plans, recipes, and, perhaps most importantly, computer code. The AI revolution is evident for many people, including entrepreneurs and businesspeople. This chat model, which is less abstract and more practical than some of the impressive, but sometimes esoteric, advances made in academia and a few high-tech companies, is clear evidence of its real potential.
Venture capitalists, as well as other investors, are investing billions in companies that use generative AI. The list of apps and services powered by large-language models is growing every day.
Microsoft is one of the major players. It has reported that it invested $10 billion in OpenAI, its ChatGPT and its Bing search engine. The technology is expected to bring new life and capabilities to its Office products and give it a boost. Salesforce announced in March that it would introduce a ChatGPT application to its popular Slack platform. It also announced a $250 million fund for investing in generative AI startups. There are many more, including GM and Coca-Cola. Everybody has a ChatGPT game.
Google also announced that it will use its new generative AI tools for Gmail, Docs and other popular products.
Will ChatGPT make the already troubling income and wealth inequality in the US and many other countries even worse? Or could it help?
There aren't yet any killer apps. Economists believe that as companies scramble to find ways to use the technology to their advantage, a rare window has opened to rethink how to reap the benefits of the new generation AI.
"We are talking in such an instant because you can touch the technology. You don't need any programming skills to play with this technology. Many people can begin to imagine how this will impact their workflow and their job prospects," Katya Klinova (head of research on AI, labor, economy, Partnership on AI in San Francisco) says.
"The question is, who will benefit?" Klinova is currently working on a report that will outline the potential job benefits of generative AI, as well as provide recommendations for how to use it to increase our shared prosperity.
It will be a powerful tool that many workers will use to improve their skills and knowledge, as well as boosting the economy. The pessimistic view: Companies will use it to eliminate what used to be automation-proof jobs. These well-paid, creative and logically-based jobs are highly-skilled and will make the companies richer but will not contribute to overall economic growth.
Supporting the least-skilled
ChatGPT's effect on the workplace is not a theoretical question.
OpenAI's TynaEloundou, Sam Manning and Pamela Mishkin analyzed the latest data with Daniel Rock from the University of Pennsylvania. They found that large language models like GPT could have an effect on at least 80% of the US workforce. The team also estimated that AI models including GPT-4, and other anticipated tools, would have a significant impact on 19% of jobs with at least 50% of those tasks "exposed". This is in contrast to earlier waves of automation. Writers, digital and web designers, financial quant analysts, and, just in case you were considering a career shift, blockchain engineers are some of the most vulnerable jobs.
David Autor, a labor economist at MIT and a leading expert in the effects of technology on jobs, says that "there is no doubt that [generative AI] will be used–it is not just a novelty." It is already being used by law firms, but that's only one example. Automation opens up new possibilities for many tasks.
Autor spent many years documenting the destruction of jobs in manufacturing and routine clerical positions that once earned a good salary. He claims ChatGPT and other examples generative AI have altered the equation.
While AI had previously automated some office tasks, it was only those repetitive step-by-step tasks which could be coded by an AI. It can now perform tasks we once considered creative, such writing and creating graphics. He says that it is obvious to anyone paying attention that generative artificial intelligence opens up the possibility of automating many tasks that we consider difficult to automate.
ChatGPT is not a concern. Author points out that there are many jobs in the US. But that ChatGPT will mean large-scale unemployment. Instead, it will mean that ChatGPT companies will replace white-collar jobs that are relatively well-paid with this new form automation. This will result in lower-paying service jobs for those who are most able to take advantage of the new technology.
Generative AI could help a wide swath of people gain the skills to compete with those who have more education and expertise.
This scenario would allow tech-savvy employees and companies to quickly adopt the AI tools and become more productive, thereby allowing them to dominate their work environments and sectors. People with less technical skills and less technical knowledge would be further behind.
Autor sees an even better outcome. Generative AI could allow a wider range of people to gain the skills necessary to compete with experts and those with more education.
ChatGPT's productivity impact on productivity was the subject of one of the most rigorous studies to date.
Whitney Zhang and Shakked Noy, both MIT economics graduate student, conducted an experiment involving hundreds college-educated professionals in marketing, HR, and other areas. They asked half of them to use ChatGPT for their daily tasks, while the rest did not. ChatGPT increased overall productivity, not surprising. But here's what really interests me: ChatGPT helped the less skilled and experienced workers the most, decreasing their performance gap. The poor writers did much better, while the skilled writers were able to write a lot faster.
ChatGPT and other generative AIs may be able to "upskill" people having difficulty finding work, according to preliminary findings. Autor states that there are many experienced workers who have been displaced from manufacturing and office jobs in the past few decades. Generative AI could be used to expand their knowledge and give them the specialized skills needed in areas like teaching or health care, which have many jobs. It could help revitalize our workforce.
It will take more thought to determine which scenario is the best.
"I don’t believe we should take it because the technology is still in its infancy and we have to adapt to it. Autor says that it is still in development and can be used in many ways. It's difficult to emphasize the importance of designing what it is there for.
We are at a point where less-skilled workers can take on knowledge work or where the most skilled knowledge workers will dramatically increase their advantages over others. The outcome will depend on how employers use tools like ChatGPT. The more optimistic option is possible.
HTML0? ______ HTML0 _____
However, there are reasons to be pessimistic. Erik Brynjolfsson, a Stanford economist, warned last spring that AI creators were too focused on imitating human intelligence and not finding new ways to enable people to perform new tasks and expand their capabilities.
Brynjolfsson claimed that the pursuit of human-like abilities has resulted in technologies that replace people with machines. This drives down wages and increases inequality of income and wealth. He argued that it is the single largest reason for the rise in wealth concentration.
He says that ChatGPT with its human-sounding outputs is "like the poster child of what I warned about": It has "turbocharged the discussion around how new technologies can be used in order to give people new capabilities rather than replacing them.
Brynjolfsson is a techno-optimist in artificial intelligence, despite his concerns that AI developers will continue blindly outdo one another in imitating human-like capabilities in the creations they create. He predicted two years ago that AI and other digital technologies would lead to a productivity boom. Today, he is optimistic about the potential impact of new AI models.
Brynjolfsson's optimism is largely based on the conviction that businesses can greatly benefit from using ChatGPT to expand and improve their offerings as well as their productivity. It's a wonderful creativity tool. It can help you do new things. It's more than just doing the same thing less expensively," Brynjolfsson says. He says that as long as developers and companies can "stay away" from the mindset of thinking that humans don't need them, it will be "very important."
He predicts that generative AI will be able to boost the US' economic growth by trillions of dollars within a decade. He says that the majority of America's economy is made up of knowledge workers and information workers. "It's difficult to imagine any information worker that won't be affected at all."
It is difficult to predict when that productivity boost will arrive. Perhaps we need to be patient.
In 1987, Robert Solow, the MIT economist who won the Nobel Prize that year for explaining how innovation drives economic growth, famously said, "You can see the computer age everywhere except in the productivity statistics." It wasn't until later, in the mid and late 1990s, that the impacts–particularly from advances in semiconductors–began showing up in the productivity data as businesses found ways to take advantage of ever cheaper computational power and related advances in software.
Could AI be the next big thing? Avi Goldfarb, an economist from the University of Toronto, said that it all depends on how we use the most recent technology to transform businesses, just as we did in earlier computers.
He says that companies have been using AI to improve their tasks so far. "It will increase efficiency, it might incrementally increase productivity, but ultimately the net benefits will be minimal." Because you are doing the exact same thing, but a little better. But technology does more than allow us to do the same things we have always done, he said. It may allow us to develop new processes that create value for customers.
It is still unclear when, or even if, generative AI will be implemented. Goldfarb states that once we determine what good writing at scale can allow industries to do differently or–in Dall-E–what graphic designer at scale allows us do differently, then we will experience a big productivity boost. I don't know if it will be next week, next year, or 10 years in the future.
Power struggle
Anton Korinek, an economist from the University of Virginia, was a Brookings Institution fellow and had access to ChatGPT's new generation of large-language models. He began to play with them to see if they could help his work. In February, he carefully recorded their performance and noted how they dealt with 25 "use cases", including brainstorming and editing text (very helpful), coding (pretty competent with some help) and doing math (not so great).
ChatGPT did not correctly explain one of economics' most fundamental principles, Korinek said. But, the error was quickly spotted and was quickly forgotten in view of the many benefits. He says, "I can assure you that it makes my cognitive worker more productive." "It is clear that I am more productive when I use a model language."
GPT-4 was released. Korinek tested it on the same 25 questions he had documented in February. It performed much better. Korinek says that there were fewer cases of it making up stuff. It also performed better in math assignments.
ChatGPT and other AI robots can automate cognitive tasks, which is in contrast to the physical tasks that require infrastructure and equipment investments. This could lead to a faster increase in economic productivity than previous technological revolutions. He believes that we could see a greater increase in productivity by the end the year, and certainly by 2024.
Who will control the future of this amazing technology?
He adds that the AI models could make researchers like him more productive and have the potential to drive technological advancement in the long-term.
The potential for large language models in physical sciences research is being realized already. Berend Smit is an expert in machine learning for discovering new materials. He runs a chemical engineering laboratory at EPFL, Lausanne, Switzerland. After Kevin Maik Jablonka (a graduate student), showed some impressive results with GPT-3, Smit asked him if GPT-3 was actually useless for the sophisticated machine-learning studies that his group uses to predict the properties.
Smit jokes, "He failed completely."
After being tuned for a few minutes using a few examples, it turns out that the model can perform as well as sophisticated machine-learning tools specifically designed for chemistry. This allows the model to answer basic questions such as the solubility or reactivity of a compound. It can be given the name of a compound and it will predict its properties based upon the structure.
Like other areas of work large language models can help increase the expertise and abilities of non-experts. In this instance, it could be chemists who have little to no knowledge of machine-learning tools. Jablonka states that it is as easy as a literature search and could help bring machine learning to the masses.
These amazing and surprising results are just one hint at how powerful new forms of AI can be across a broad range of creative work, including scientific discovery. They are also a glimpse of how easy they are to use. This also raises fundamental questions.
The potential impact of generative AI in the economy and employment is becoming more apparent. Who will decide how these tools should design and be deployed? Who will decide the direction of this incredible technology's future?
An economist at Cambridge University, Diane Coyle says that one concern is the possibility of large language models being dominated by the same companies that control much of the digital universe. She points out that Meta and Google offer large language models in addition to OpenAI. However, the high computational costs of running the software make it difficult for anyone to enter the market.
Coyle states that the concern is that these companies use similar "advertising driven business models." "So you can get a certain unity of thought if you don’t have different types of people with different incentives."
Coyle admits that there are no easy solutions, but she suggests a publicly funded international research agency for generative AI. It would be modeled after CERN (the Geneva-based intergovernmental European nuclear energy research body where the World Wide Web was first created in 1989). It would have the computing power and scientific expertise necessary to run the models.
Coyle says that such an effort would be outside the realm of Big Tech and "bring some diversity in the incentives that the model creators face when producing them."
Coyle says that while it is not clear which policies will be most effective in ensuring that large-language models serve the public interest best, it is becoming increasingly obvious that we cannot leave the decision about how to use technology up to the market and a few companies.
There are many examples in history that show how vital government-funded research is to developing technology that will bring about widespread prosperity. The internet was born out of another publicly funded effort that began long before CERN's invention of the internet. In the 1960s, the US Department of Defense supported ARPANET. This project pioneered the use of multiple computers to communicate with one another.
Daron Acemoglu and Simon Johnson, MIT economists, give a fascinating walk through the history and mixed results of technological progress in creating widespread prosperity. They argue that technological advancements should be directed in a way that provides broad benefits, not just for the wealthy.
The rapid technological advances in the US economy from the years after World War II to the 1970s saw wages rise and income inequality drop. Acemoglu & Johnson believe that technological advances allowed for new jobs and tasks, while social pressures and political pressures ensured that workers received more of the benefits with their employers than they do today.
They write that the recent rapid adoption by manufacturing robots in the Midwest, "the industrial heartland" of the American economy, has resulted in a "prolonged regional decline and destroyed jobs."
This book will be available in May and is especially relevant to understanding the rapid advancements in AI today and how they will impact us all moving forward. Acemoglu stated that they had been writing the book at the time GPT-3 was released. Acemoglu jokes that they foresaw ChatGPT, and he agrees.
Acemoglu believes that AI creators "are going in a wrong direction." He says the entire architecture of AI is in an "automation mode." "But there's nothing intrinsic about generative AI, or AI in general, that should drive us in this direction. It's the vision and business models of OpenAI, Microsoft, and the venture capitalist community that will drive us in this direction.
If you think we can control a technology's course, then the obvious question is: Who are "we?" Acemoglu's and Johnson's provocative statements are at the heart of this. They wrote: "Society must stop being seduced by tech billionaires and their agenda…. One doesn't need to be an AI expert in order to have a say on the direction of advancement and the future of society forged through these technologies."
ChatGPT's creators and those who brought it to market, including OpenAI CEO Sam Altman, are to be commended for introducing the new AI sensation to the general public. Its potential is vast. It doesn't mean that we should accept their vision of where we technology will go and how it should use.
Their narrative states that the ultimate goal is artificial general Intelligence, which, if everything goes according to plan, will result in great economic wealth and abundances. Altman has been a long-standing advocate for a universal basic income (UBI), to help the non-technocrats. It sounds appealing to some. It's easy to make money with no work! Sweet!
The most worrying thing about the narrative is the assumptions that underlie it. These assumptions are that AI is heading in a job-destroying direction and that most people are just going along for the ride. ride. This view doesn't even consider the possibility of generative AI causing a productivity and creativity boom for workers that goes beyond tech-savvy elites. It could help unlock their brains and talents. The idea of using technology to create widespread prosperity through expanding the human abilities and expertise of the workforce is not discussed much.
Companies can decide to use ChatGPT to give workers more abilities—or to simply cut jobs and trim costs.
Acemoglu, Johnson, and others write that we are headed toward greater inequality because of our faulty choices regarding who is in power and how technology will affect society…. UBI fully supports the vision of the tech elite and business leaders that they are the intelligent, talented people who should be generously funding the rest.
Johnson and Acemoglu write about various ways to achieve "a more balanced portfolio of technology," including tax reforms, which might encourage the creation more worker-friendly AI, and reforms that might remove Big Tech funding for computer science research.
However, economists admit that such reforms are "a difficult order" and that a social push for technological change is "not just around."
We have the option to choose how we use ChatGPT or other large language models. Businesses and individuals will be able to decide how to make use of the technology as many apps are released quickly. Companies can choose to offer more capabilities to workers or to reduce costs.
Another positive development is that there is some momentum behind open source projects in generative Ai, which could help break Big Tech's hold on the models. Last year, more than 1000 international researchers worked together to create Bloom, a large-scale language model that can produce text in French, Spanish and Arabic. If Coyle and other researchers are correct, increasing public funding for AI research may help to change the course of future breakthroughs.
Stanford's Brynjolfsson denies that he is optimistic about the outcome. His enthusiasm for technology is evident. He says, "We can have the greatest decade ever if the technology is used in the right direction." It's not impossible.
————————————————————————————————————————————————————————————
By: David Rotman
Title: ChatGPT is about to revolutionize the economy. We need to decide what that looks like.
Sourced From: www.technologyreview.com/2023/03/25/1070275/chatgpt-revolutionize-economy-decide-what-looks-like/
Published Date: Sat, 25 Mar 2023 10:30:00 +0000
Leave a Reply