We will not run out of work
Jun 2023
We will not run out of work
Artificial intelligence will make certain jobs in service industries redundant, but technological progress will ensure higher per capita economic output, higher wages, and more prosperity in the longer term. A commentary by Joachim Voth.
This article was originally published in the newspaper «Finanz und Wirtschaft» on 23 May 2023. Translated and edited for layout purposes by the UBS Center.
The fear is back - the fear of a new technology. This time it is AI, artificial intelligence, in the form of large language models (LLM) like OpenAI-GPT, which could replace many highly skilled workers.
Anyone who has ever played with one of these models, which are now widely accessible (such as Microsoft's search engine Bing), knows how impressively they can handle complex language in particular. A poem for Father's Day? A limerick on April 1? A friendly e-mail to the boss generated with a one-liner? All no problem, often characterized with amazing eloquence. Especially the latest model of OpenAI, GPT-4, shines with an enormously broad knowledge base and accurate text comprehension. What is particularly striking is how quickly the new technology is triumphing. Three years ago, very few people had heard of it; a year ago, the first version was presented to some testers; the latest versions of the past few months now demonstrate huge progress in quality.
Nevertheless, it should be clearly stated that important elements are often still missing at the moment: GPT is only familiar with the world until the summer of 2021; everything after that is unknown to the program. If you ask who currently rules the United Kingdom, you get "Queen Elizabeth II" as an answer. Interaction with other information sources is also cumbersome at the moment. Longer texts or even image sources can only be uploaded for individual products. In addition, there are problems with the character of the chatbots. Microsoft's Bing is, not entirely surprisingly for those familiar with the software company, often gruff in its dealings with "customers"; moreover, it occasionally likes to end the conversation if one points out an error. GPT, on the other hand, often "hallucinates", invents articles that don't exist, or stoops to assertions about alternative facts.
Artificial intelligence will make certain jobs in service industries redundant, but technological progress will ensure higher per capita economic output, higher wages, and more prosperity in the longer term. A commentary by Joachim Voth.
This article was originally published in the newspaper «Finanz und Wirtschaft» on 23 May 2023. Translated and edited for layout purposes by the UBS Center.
The fear is back - the fear of a new technology. This time it is AI, artificial intelligence, in the form of large language models (LLM) like OpenAI-GPT, which could replace many highly skilled workers.
Joachim Voth, UBS Foundation Professor of Macroeconomics and Financial Markets
Statistical Parrots
But even if the fine-tuning is lacking: the future is clearly visible. Automated interfaces and plug-ins from other software will massively increase the capabilities of the LLM. Thanks to these programs, programming is already child's play for even minimally trained laymen. What would have taken a programmer several days nine months ago can now be made to work by a beginner with the help of these programs. Medical knowledge, often more recent than that of some practicing physicians, is often strikingly accurate and easily accessible. Evaluation of written texts, often a tricky business for experienced specialists, is a breeze.
And so an old, new question arises: If technology can do so much, what is left for people? Who is going to write the children's books when GPT manages to write entertaining, instructive, melodious, and child-friendly texts every minute? Who will still be needed in the financial industry as a financial analyst when GPT can be married to a data service like Bloomberg and the analytical capacity of dozens of employees can be called up at the push of a button? Who will still use people for copywriting advertising?
To systematically weigh the consequences, it helps to first realize what LLM are and can do, and what is beyond their capabilities. LLMs, of course, are not really on the way to replacing human intelligence; they are "statistical parrots": they understand what they say no more than the parrot that loudly squawks "come to mommy" when its master comes home. Programs like GPT are artificial rather than intelligent. They are trained using as much text as possible; GPT-3, for example, with 8 million documents and over 10 billion words. So the program learns which word is likely to come next when you say, for example, "That knocks me right _". Statistically, "out" is the correct answer. This insight can then be applied to complete sentences and paragraphs; LLMs vigorously churn through word bags of billions of observations once and see what their training documents said on average about a topic. The more obscure the topic, the less the LLM "know." Simple questions are often surprisingly poorly answered because, for example, there are no long discussions on Internet forums about how to solve them.
But even if the fine-tuning is lacking: the future is clearly visible. Automated interfaces and plug-ins from other software will massively increase the capabilities of the LLM. Thanks to these programs, programming is already child's play for even minimally trained laymen. What would have taken a programmer several days nine months ago can now be made to work by a beginner with the help of these programs. Medical knowledge, often more recent than that of some practicing physicians, is often strikingly accurate and easily accessible. Evaluation of written texts, often a tricky business for experienced specialists, is a breeze.
And so an old, new question arises: If technology can do so much, what is left for people? Who is going to write the children's books when GPT manages to write entertaining, instructive, melodious, and child-friendly texts every minute? Who will still be needed in the financial industry as a financial analyst when GPT can be married to a data service like Bloomberg and the analytical capacity of dozens of employees can be called up at the push of a button? Who will still use people for copywriting advertising?
Entry points are at risk
However, this means that the LLMs have a structural advantage for questions and topics of medium complexity. Wherever many people have similar questions, and if the answers are neither very simple nor really difficult, they can score. They generate source code for programs and text for essays that do not yet exist in this way, but they do not create anything really new - it is synthetic text that is always just a variation and shaken mix of already existing information and text. As long as there is a demand for fresh new writing, novelists and poets need not worry; neither should good journalists who interview new sources who share new insights. Scientists and creative copywriters should also have little to worry about in terms of their jobs. A new ad or commercial that is just a recycling of old ideas will not succeed.
Where employment changes are most likely to occur is in entry-level positions for highly skilled careers. For example, there will probably be significantly fewer jobs for financial analysts, research assistants, paralegals in law firms, and executive assistants. LLMs are likely to replace jobs that currently exist wherever creation of an initial overview of the existing state of knowledge is critical, perhaps coupled with some initial quantitative analysis. At the same time, the managers, partners, and directors they serve will become more productive, and their increased purchasing power will in turn be spent on goods and services. Accordingly, some employment will shift away from areas of the service industry with high language and quantitative analysis requirements to activities with more personal contact.
However, this means that the LLMs have a structural advantage for questions and topics of medium complexity. Wherever many people have similar questions, and if the answers are neither very simple nor really difficult, they can score. They generate source code for programs and text for essays that do not yet exist in this way, but they do not create anything really new - it is synthetic text that is always just a variation and shaken mix of already existing information and text. As long as there is a demand for fresh new writing, novelists and poets need not worry; neither should good journalists who interview new sources who share new insights. Scientists and creative copywriters should also have little to worry about in terms of their jobs. A new ad or commercial that is just a recycling of old ideas will not succeed.
Where employment changes are most likely to occur is in entry-level positions for highly skilled careers. For example, there will probably be significantly fewer jobs for financial analysts, research assistants, paralegals in law firms, and executive assistants. LLMs are likely to replace jobs that currently exist wherever creation of an initial overview of the existing state of knowledge is critical, perhaps coupled with some initial quantitative analysis. At the same time, the managers, partners, and directors they serve will become more productive, and their increased purchasing power will in turn be spent on goods and services. Accordingly, some employment will shift away from areas of the service industry with high language and quantitative analysis requirements to activities with more personal contact.
New employment thanks to decreasing costs
But while not all financial analysts will be good personal trainers, there is little cause for concern: higher productivity often drives down costs, so lower prices create additional employment. When Ford introduced assembly line production, the amount of output per employee increased enormously, but employment in the automotive industry did not decline because falling prices ensured that more and more people could afford a car. In the USA, for example, if LLM ensures lower medical costs and cheaper legal fees, higher demand can be expected.
In the long term, technical progress ensures one thing above all: higher economic output per capita, higher wages, more prosperity. Although each worker today produces more than ten times the economic output of his or her ancestors 200 years ago, we are not running out of work. Technical unemployment is almost unheard of in the long run, although isolated exceptions such as the former unrest among English textile workers and day laborers in agriculture are important exceptions. What is important, however, is that when massive dislocations occur, there is sufficient cushioning support to help with career reorientation and to ensure that no one is left stranded while the labor market rebuilds.
But while not all financial analysts will be good personal trainers, there is little cause for concern: higher productivity often drives down costs, so lower prices create additional employment. When Ford introduced assembly line production, the amount of output per employee increased enormously, but employment in the automotive industry did not decline because falling prices ensured that more and more people could afford a car. In the USA, for example, if LLM ensures lower medical costs and cheaper legal fees, higher demand can be expected.
In the long term, technical progress ensures one thing above all: higher economic output per capita, higher wages, more prosperity. Although each worker today produces more than ten times the economic output of his or her ancestors 200 years ago, we are not running out of work. Technical unemployment is almost unheard of in the long run, although isolated exceptions such as the former unrest among English textile workers and day laborers in agriculture are important exceptions. What is important, however, is that when massive dislocations occur, there is sufficient cushioning support to help with career reorientation and to ensure that no one is left stranded while the labor market rebuilds.
Quote
Programs like GPT are more artificial than intelligent.
Slider story
Contact
UBS Foundation Professor of Macroeconomics and Financial Markets
Joachim Voth received his PhD from Oxford in 1996. He works on financial crises, long-run growth, as well as on the origins of political extremism. He has examined public debt dynamics and bank lending to the first serial defaulter in history, analysed risk-taking behaviour by lenders as a result of personal shocks, and the investor performance during speculative bubbles. Joachim has also examined the deep historical roots of anti-Semitism, showing that the same cities where pogroms occurred in the Middle Age also persecuted Jews more in the 1930s; he has analyzed the extent to which schooling can create radical racial stereotypes over the long run, and how dense social networks (“social capital”) facilitated the spread of the Nazi party. In his work on long-run growth, he has investigated the effects of fertility restriction, the role of warfare, and the importance of state capacity. Joachim has published more than 80 academic articles and 3 academic books, 5 trade books and more than 50 newspaper columns, op-eds and book reviews. His research has been highlighted in The Economist, the Financial Times, the Wall Street Journal, the Guardian, El Pais, Vanguardia, La Repubblica, the Frankfurter Allgemeine, NZZ, der Standard, der Spiegel, CNN, RTN, Swiss and German TV and radio.
UBS Foundation Professor of Macroeconomics and Financial Markets
Joachim Voth received his PhD from Oxford in 1996. He works on financial crises, long-run growth, as well as on the origins of political extremism. He has examined public debt dynamics and bank lending to the first serial defaulter in history, analysed risk-taking behaviour by lenders as a result of personal shocks, and the investor performance during speculative bubbles. Joachim has also examined the deep historical roots of anti-Semitism, showing that the same cities where pogroms occurred in the Middle Age also persecuted Jews more in the 1930s; he has analyzed the extent to which schooling can create radical racial stereotypes over the long run, and how dense social networks (“social capital”) facilitated the spread of the Nazi party. In his work on long-run growth, he has investigated the effects of fertility restriction, the role of warfare, and the importance of state capacity. Joachim has published more than 80 academic articles and 3 academic books, 5 trade books and more than 50 newspaper columns, op-eds and book reviews. His research has been highlighted in The Economist, the Financial Times, the Wall Street Journal, the Guardian, El Pais, Vanguardia, La Repubblica, the Frankfurter Allgemeine, NZZ, der Standard, der Spiegel, CNN, RTN, Swiss and German TV and radio.