The Potential Economic Revolution Enabled by Large Language Models: Our Choices Will Shape Its Outcome
The dawn of new large language models is set to revolutionize many professions. However, whether this change will result in widespread prosperity hinges on our actions.
By: Rooz Aliabadi
Over the past few months, an artificial intelligence gold rush has begun, fueled by the promise of lucrative business opportunities presented by generative AI models such as ChatGPT, regardless of the hallucinatory beliefs surrounding them. App developers, startups, and even some of the world’s biggest companies are in a frenzy, attempting to understand the capabilities of the sensational text-generating bot that OpenAI unveiled last November.
One can almost hear the cacophony of voices from executive suites worldwide as they clamor to answer the questions: “What is our ChatGPT strategy? How can we capitalize on this?”
While businesses and executives are eyeing a profitable opportunity, the potential impact of generative AI technology on the workforce and the economy as a whole needs to be clarified. Despite their flaws, including their inclination to fabricate information, recently released generative AI models like ChatGPT offer the potential to automate tasks previously believed to be exclusive to human creativity and reasoning, such as writing, graphic design, and data summarization and analysis, even music composition. This leaves economists and many others needing clarification about how jobs and overall productivity will be affected.
Despite the remarkable advances in AI and other digital tools over the past decade, their ability to enhance prosperity and stimulate widespread economic growth has been disheartening. While a select few investors and entrepreneurs have amassed great wealth, most people have not reaped the benefits, and some have even been replaced by automation.
Since around 2005, productivity growth in the United States and most advanced economies, except for the UK, has been lackluster, hindering their potential for incredible wealth and prosperity. The limited expansion of the economic pie has resulted in stagnant wages for many workers.
The few instances of productivity growth during this time have been restricted to specific sectors and certain cities in the US, including San Jose, San Francisco, Seattle, and Boston. Given the alarming income and wealth inequality in the United States and numerous other nations, will ChatGPT worsen this disparity, or could it alleviate it? Could it provide a much-needed stimulus to productivity?
Large language models like ChatGPT, which boasts human-like writing capabilities, and OpenAI’s DALL-E 2, capable of generating images on demand, rely on vast data for their training. Competing models such as Anthropic’s Claude and Google’s Bard follow the same principle. These foundational models, including OpenAI’s GPT-3.5 used by ChatGPT and Google’s language model LaMDA, which powers Bard, have rapidly evolved in recent years.
Their power continues to grow as they are trained on ever-increasing amounts of data, and the number of parameters- the variables in the models that are adjusted- is increasing dramatically. OpenAI’s latest release, GPT-4, was unveiled earlier this month. While the exact parameter count has not been disclosed, it will be significantly larger than its predecessor GPT-3, which had around 175 billion parameters and was approximately 100 times larger than GPT-2.
The release of ChatGPT in late 2022 transformed the landscape for many users, providing an incredibly easy-to-use tool that can quickly create human-like text. This includes everything from recipes to workout plans, and even computer code, surprising many users. For non-experts, especially entrepreneurs and businesspeople, the chat model is a practical and user-friendly example of the potential of the AI revolution. Unlike the abstract and technical advances of academia and select high-tech companies, it is seen as evidence of its real-world impact.
This has led to an inflow of investment from venture capitalists and other investors, with billions poured into companies centered around generative AI. As a result, the list of apps and services driven by large language models continues to grow, with each passing day bringing new additions.
Microsoft has invested $10 billion in OpenAI and ChatGPT technology to revive its Bing search engine and add new capabilities to its Office products. Similarly, Salesforce has announced plans to introduce a ChatGPT app in its popular Slack product (which I use at ReadyAI daily) while establishing a $290 million fund to invest in generative AI startups. From Coca-Cola to GM and Ford, companies across various industries are making their own ChatGPT plays. At the same time, Google has announced that it plans to utilize its new generative AI tools in widely-used products like Gmail and Docs.
Despite the rush to find applications for ChatGPT and other generative AI models, there have yet to be apparent stand-out uses. This presents a unique opportunity for us to rethink how to maximize the benefits of this new technology.
The current moment offers a unique opportunity to explore the potential impact of generative AI on workflow and job prospects. However, we must question who will benefit from this technology and who will be left behind.
The optimistic view is that generative AI will establish a potent tool for many of us, improving our capabilities and expertise while boosting the economy. On the other hand, the pessimistic view is that companies will use it to destroy automation-proof jobs that require creative skills and logical reasoning, leaving a few high-tech companies and tech elites even richer but doing little for overall economic development and prosperity.
Assisting individuals with the lower-level of skills
The impact of ChatGPT on the workplace is not merely a theoretical concern. A recent analysis by OpenAI’s Tyna Eloundou, Sam Manning, and Pamela Mishkin found that large language models like GPT could potentially impact 80% of the US workforce. They further indicated that these AI models, including GPT-4 and other forthcoming software tools, would significantly affect 20% of jobs, with at least 55% of tasks in those jobs “exposed.” In contrast to previous waves of automation, higher-income jobs would be most affected, with writers, web and digital designers, quantitative financial analysts, and even blockchain engineers among those with the most vulnerable positions.
There is no question that generative AI will be used, citing law firms as one example. It will open up a range of tasks that can be automated. ChatGPT and other generative AI examples have changed the game. While AI had automated some office work before, only those rote step-by-step tasks could be coded for a machine. Now, AI can perform tasks once viewed as creative, such as writing and producing graphics. It’s apparent to anyone paying attention that generative AI opens the door to computerizing many functions that we think need to be more easily automated.
The concern is not that ChatGPT will lead to large-scale unemployment, as there are still plenty of jobs in the US, but that companies will replace relatively well-paying jobs with this new form of automation. This could result in workers being sent off to lower-paying service employment. At the same time, only a few individuals can exploit the new technology and reap all the benefits.
If this scenario continues, individuals and businesses with solid technology skills may adopt generative AI tools and become significantly more efficient, ultimately dominating their respective industries. However, those with similar technical abilities and less skilled workers could stay caught up, exacerbating existing economic inequalities.
However, we envision a more optimistic scenario where generative AI can enable more people to acquire the necessary skills to compete with those with higher education and expertise.
An experiment conducted by two MIT economics graduate students, Shakked Noy, and Whitney Zhang, asked hundreds of college-educated professionals in fields like marketing and HR to use ChatGPT in their daily tasks. In contrast, the others were not asked to use it. The AI tool raised overall productivity and assisted the least skilled and accomplished workers the most, reducing the performance gap between employees. In other words, poor writers improved significantly, while good writers became faster.
These initial findings suggest that ChatGPT and other generative AIs could “upskill” people struggling to find work. Many experienced workers are currently “lying fallow” after being ousted from office and manufacturing positions in recent years. It could revitalize the workforce if generative AI can be used as a practical tool to expand their expertise and provide them with specialized skills needed in healthcare or teaching.
To determine which scenario will prevail, we need to make a concerted effort to consider how we want to utilize the technology. We shouldn’t assume that technology is already out there and we have to adapt to it. Since the technology is still in development, we have the opportunity to use it in a variety of ways. The key is to design it with intention.
In essence, we are at a crossroads where individuals with fewer skills can take on knowledge work, or those already highly skilled will significantly expand their advantages. The outcome will largely depend on how employers implement tools like ChatGPT. However, the more optimistic scenario is entirely within our grasp.
Beyond Human-Centered Design
Nevertheless, there are reasons to have a pessimistic outlook. AI creators needed to focus more on replicating human intelligence instead of leveraging the technology to empower individuals to perform new tasks and expand their abilities.
Pursuing human-like capabilities has resulted in technologies that merely displace human workers with machines, lowering wages and exacerbating wealth and income inequality. This is the single most significant explanation for the increasing concentration of wealth.
ChatGPT, with its human-like language outputs, embodies the very concern. It has accelerated the conversation about how these technologies can be leveraged to enhance people’s capabilities instead of solely displacing them.
Despite many concerns about AI developers prioritizing human-like capabilities over extending human abilities, I remain optimistic about artificial intelligence’s potential.
Businesses can benefit significantly from generative AI by expanding their offerings and increasing productivity. It is a powerful tool for creativity and innovation rather than simply a means of doing things more cheaply. As long as developers and companies avoid the mindset that humans are unnecessary, generative AI will be critical.
Within a decade, generative AI could contribute trillions of dollars to the US economy, affecting nearly all types of knowledge workers. However, the timing of this productivity boost remains uncertain. It may require patience.
In 1987, Nobel laureate economist Robert Solow from MIT made a well-known statement: “You can see the computer age everywhere except in the productivity statistics.” Only in the mid to late 1990s did the effects, particularly from semiconductor improvements, appear in productivity data as businesses learned to harness increasingly affordable computational power and related software advancements.
The impact of AI on productivity will depend on our ability to use the latest technology to transform businesses, much like we did in the earlier computer age. Companies only use AI to incrementally improve tasks, which may increase efficiency but have limited net benefits. However, the true potential of AI lies in creating new processes and value for customers. The timeline remains to be determined, as we need to figure out how to use generative AI for industries like writing and graphic design. Once we have identified how AI can revolutionize these industries, a significant productivity boost will occur, but the timeline for this breakthrough still needs to be determined.
The Power Struggle in the Age of Artificial Intelligence
I believe that since ChatGPT and other AI bots automate cognitive work rather than physical tasks that require infrastructure and equipment investments, there may be a more significant boost to economic productivity than in past technological revolutions. A productivity boost could occur much more quickly by the end of the year or, indeed, by 2024.
Furthermore, the potential for large language models to enhance productivity and drive technological progress is broader than economics. This potential is already being realized in the physical sciences, as seen in the work of Berend Smit, a chemical engineering researcher at EPFL in Lausanne, Switzerland. Smit’s group uses machine learning to discover new materials. After one of his graduate students demonstrated interesting results using GPT-3, Smit challenged the student to prove that the model was useless for their sophisticated machine-learning studies that predict compound properties.
However, the student should have done so. With just a few minutes of fine-tuning and relevant examples, the model could perform as well as advanced machine-learning tools explicitly developed for chemistry. Based on the compound name and structure, it could accurately answer basic questions about compound properties, such as solubility and reactivity.
Large language models have the potential to expand the expertise and capabilities of non-experts, such as chemists with little knowledge of complex machine-learning tools, similar to other areas of work. Kevin Maik Jablonka notes that as simple as a literature search; it could bring machine learning to the masses of chemists. These surprising results show the significant power of the new forms of AI in various creative fields, including scientific discovery, and how easily they can be utilized. However, this also raises critical questions about who will define the vision for the design and deployment of these tools and control the future of this remarkable technology as its potential impact on the economy and jobs become more apparent.
There is a concern that large language models may be controlled by the same big tech companies already dominating much of the digital world. For example, Google and Meta offer their large language models alongside OpenAI, and the high computational costs required to run the software create a barrier to entry for competitors. As a result, there is a risk of uniformity of thought and incentives, which is a big concern when it comes to a technology that has such a far-reaching impact.
One possible solution is establishing a publicly funded international research organization for generative AI modeled after CERN. This organization would have the necessary computing power and scientific expertise to develop the technology further but would be outside of Big Tech. This would bring some diversity to the incentives of the creators of the models. Although it is still being determined which public policies would best serve the public interest, it is becoming clear that a few dominant companies and the market must make decisions about using this technology.
Government-funded research has played a pivotal role in developing technologies that have brought widespread prosperity. For instance, in the late 1960s, the US Department of Defense backed ARPANET, which paved the way for the internet long before creating the World Wide Web at CERN.
It’s essential to steer technological advancements in ways that benefit the masses and not just the privileged few. Federally-funded research has been critical in developing technologies that lead to general prosperity. Technological advances created new tasks and jobs, raising wages and decreasing income inequality. However, the recent adoption of manufacturing robots in the American Midwest has resulted in job loss and regional decline.
Rapid progress in AI could affect us all and emphasizes the importance of steering technological advances in ways that provide broad benefits. Our society and its powerful gatekeepers must stop being fascinated by tech billionaires’ agendas. They should have a say in the direction of progress and the future of our society.
The creators of AI and the businesspeople involved in bringing it to market deserve credit for their efforts. Still, we must not blindly accept their vision and aspirations for the technology’s future. The assumption that AI is headed on an inevitable job-destroying path is troubling. It barely acknowledges that generative AI could lead to a creativity and productivity boom for workers beyond the tech-savvy elites.
There are various tools for achieving a more balanced technology portfolio, such as tax reforms and government policies encouraging worker-friendly AI creation. However, they acknowledge that such reforms are a tall order, and redirecting technological change will require a social push.
Fortunately, our direction with ChatGPT and other large language models is within our control. As these technologies are rapidly deployed in various applications, businesses and individuals can use them to enhance worker abilities or cut costs by eliminating jobs. Additionally, open-source projects in generative AI are gaining momentum, potentially breaking Big Tech’s hold on these models. For example, more than a thousand international researchers collaborated last year on an open-source language model called Bloom, which can create text in multiple languages. Increased public funding for AI research could also change the course of future breakthroughs. While I am not entirely optimistic about the outcome, he is enthusiastic about the potential of these technologies, emphasizing that using them in the right direction could lead to one of the best decades ever, but it is not an inevitable outcome.
This Chat GPT Lesson Plan and others are available FREE to all educators at edu.readyai.org
This article was written by Rooz Aliabadi (rooz@readyai.org). Rooz is the CEO (Chief Troublemaker) at ReadyAI.org
To learn more about ReadyAI, visit www.readyai.org or email us at info@readyai.org.
