Tyler Cowen: Hayek lecture on economics, AI and large langauge models

Tyler Cowen gave the Hayek lecture in London this week. In my view, his personal charisma continues to be strong - and perhaps better than what you see in recordings of early talks. Tyler is funny (he could do with even more jokes!). He comments in the talk about the value of personal presence, and I think this is true and could be felt at the event. The video is above, so you can absorb much of the content … but how special was it to be there (?) (and also to chat to others there). The social capital part.

I make a short summary of my notes below. Suitable for a tweet thread. Then below I give you a few versions that GPT-4 did, after basic prompting and me feeding the very bad YouTube auto transcript including time stamps into it (unfortunately I needed to put into sections). I also asked GPT-4 to rate Tyler’s lecture (see end!)

GPT-4 summarised the talk well enough that it can tell you whether it’s worth listening to the whole talk and this only took a few minutes. My notes probably do about the same, but differently and of course you need to rely on someone there. I don’t think I saw any other note takers. Of course…. Context is (still) that which is scarce.

My short notes:

Two kinds of AI talks. 1. Audiences that need to be convinced. 2. Audiences that do not need to be convinced.

We have invented AI that is better than humans in many areas.

How should we think about this in economic terms ?

We have a research assistant, colleague and architect. Not a carpenter or gardener.

One argument. We have increased the capital stock. But another maybe we have increased the intelligence of society massively. So in that sense it’s labour. A new factor of production.

What is worth more? Or less?

Less.

1. Routine back office work, you are worth less. 2. Memorising facts is worth less.

Personal charisma will be worth more. Looks and how you come across more. Personal networks will be worth more.

Physical Co-ordination, carpenters and gardeners should be worth more. People who can manage having an assistant will do better.

How should capital be revalued ? Many forms of Labour are now less scarce. So long electricity, long some hardware.

The AI will be commodified. The gains will go to the users. No one is talking about Gutenberg the billionaire.

Central planning going to be harder. More projects, more complexity harder to centrally plan.

Economic insights rooted in Smith, Hayek. Decentralised systems hard to take over.

AI Can’t work in physical space. Think of it like Humans and dogs have co-evolved.

The AI is going to have its own economy. It’s own laws, it’s own economy. Maybe crypto.

(Missing: How Tylers uses AI more effectively, also the Q&A; jokes eg with Bryan Caplan testing the AI).

GPT summary (first 15 minutes):

Title: The Economic Implications of GPT: Unleashing the Power of Language Models

Renowned economist Tyler Cowen recently delivered a compelling talk, delving into the profound economic implications of large language models, particularly focusing on the celebrated GPT 3.5. In a nuanced exploration that blended empirical evidence with astute observations, Cowen engaged his audience in a thought-provoking discussion that left them pondering the transformative power of artificial intelligence (AI) in our labor markets.

Cowen began by highlighting the remarkable abilities of GPT 3.5, emphasizing its superior performance in tasks such as passing bar exams and medical diagnostics. Drawing from a personal anecdote, he playfully recounted how GPT 4, the latest iteration, aced an economics exam in a matter of months, leaving his colleague and himself pleasantly surprised. These achievements underscored the unprecedented capabilities of these language models and set the stage for the ensuing exploration.

Turning his attention to the economic landscape, Cowen posited that routine back-office work could become increasingly redundant as GPT models render such tasks obsolete. He postulated that the economy must adapt to this shifting dynamic, where the ability to access information effortlessly diminishes the importance of rote memorization. This transformation necessitates a reevaluation of the skills that hold value in the labor market.

Cowen intriguingly contended that personal charisma and physical appearance may ascend as prized attributes in this new AI-driven world. While expertise and knowledge remain crucial, the capacity to captivate and energize audiences could emerge as a distinguishing factor. Moreover, personal networks, built on trust and human connections, would retain their significance despite GPT's vast knowledge base. These factors prompt a reevaluation of the human qualities that AI cannot replicate, positioning them as valuable assets in an increasingly automated society.

Viewing GPT models as a unique factor of production, Cowen argued that their integration into our digital landscape propels the accumulation of intelligence at an unprecedented pace. Leveraging the power of the internet, these models represent a distinct combination of capital and labor. Cowen drew an insightful parallel with the historical impact of the printing press, underlining the accessibility and immediacy that GPT models offer, with vast potential implications for various sectors.

Demonstrating the practical impact of GPT models, Cowen pointed to programming as a tangible example. Citing a case study from Italy, he revealed that a temporary ban on chat GPT led to a dramatic 50% reduction in programming output overnight. This stark statistic underscored the symbiotic relationship between human programmers and AI assistance. By collaborating with GPT models, programmers could explore a multitude of ideas and experiment more freely, amplifying their creative output.

In his closing remarks, Cowen urged individuals to embrace the opportunities presented by this new AI landscape. Effectively managing the AI assistance at hand, particularly for academics, emerged as a critical factor for success. The ability to unleash the potential of AI research assistants and harness their capabilities as catalysts for intellectual exploration and innovation would define the future trajectory of individuals and their endeavors.

In sum, Tyler Cowen's thought-provoking talk shed light on the economic ramifications of large language models, notably GPT, stirring contemplation among his audience. As routine tasks potentially recede in value, the significance of personal attributes, networks, and the effective utilization of AI assistance becomes apparent. Ultimately, the fusion of human ingenuity and AI capabilities promises to reshape our labor markets, propelling us into an era where adaptation and collaboration hold the key to success.

GPT - I ask it to add the next 15 minutes:

Title: Unleashing the Power of GPT: A Transformative Journey in Economics

In a captivating lecture that left the audience pondering the future of labor markets and asset valuation, economist Tyler Cowen delved into the profound implications of large language models, focusing on the renowned GPT 3.5. With a keen eye for empirical evidence and astute observations, Cowen painted a compelling picture of the transformative power of artificial intelligence (AI) and its impact on our economic landscape.

Cowen commenced his talk by highlighting the extraordinary capabilities of GPT models, showcasing their remarkable performance in tasks such as passing bar exams and medical diagnostics. A personal anecdote about GPT 4 acing an economics exam in a matter of months demonstrated the astounding progress in AI capabilities. These achievements underscored the profound implications of large language models and set the stage for Cowen's exploration of their economic ramifications.

Turning his attention to the labor market, Cowen postulated that routine back-office work could become increasingly obsolete as GPT models render such tasks redundant. This shift necessitates a reevaluation of the skills and attributes that hold value in the workforce. Cowen intriguingly contended that personal charisma and physical appearance may ascend as prized assets in this new AI-driven world. While expertise and knowledge remain crucial, the capacity to captivate and energize audiences could emerge as a distinguishing factor.

Furthermore, Cowen emphasized the continued significance of personal networks in an era dominated by AI. Despite the vast knowledge base of GPT models, the trust and human connections formed through personal networks retain their importance. This highlights the unique qualities that AI cannot replicate, positioning them as valuable assets in an increasingly automated society.

Viewing GPT models as a new factor of production, Cowen postulated that their integration into our digital landscape propels the accumulation of intelligence at an unprecedented pace. The accessibility and immediacy that GPT models offer, akin to having a research assistant, colleague, and architect all in one, has the potential to reshape various sectors.

Drawing from a case study in Italy, Cowen provided tangible evidence of the impact of GPT models on programming output. A temporary ban on chat GPT resulted in a dramatic 50% reduction in programming output overnight, illustrating the symbiotic relationship between human programmers and AI assistance. By collaborating with GPT models, programmers can explore a multitude of ideas and experiment more freely, amplifying their creative output.

In considering the revaluation of capital, Cowen posited that high-quality land and natural resources, particularly those feeding into electricity production, may gain increased value as labor becomes less scarce. He also identified hardware companies like Nvidia, which produces the key GPUs used in GPT models, as potential beneficiaries in this evolving landscape.

Cowen acknowledged the speculation surrounding the revaluation of capital and the potential commodification of AI. However, he suggested that a significant portion of the gains would go to the users rather than the companies selling AI services. Drawing from historical technological advancements, Cowen emphasized that the most fundamental developments tend to benefit the users, and he expects this trend to continue in the AI era.

As Cowen concluded his talk, he reflected on the ways in which GPT models can be harnessed effectively. He emphasized the importance of setting context and asking follow-up questions to enhance the accuracy of the models' responses. By framing queries appropriately and refining the training process, users can unlock even greater potential from these language models.

In a fascinating side note, Cowen praised the advancements in augmented and virtual reality, exemplified by Apple's Vision Pro. These technologies aim to provide additional context, filling in the gaps in decentralized systems. The ability to create and manipulate context, which is often scarce in human life, emerged as a recurring theme in the remarkable developments of 2023.

Tyler Cowen's thought-provoking lecture shed light on the economic implications of large language models, particularly GPT, leaving the audience captivated by the transformative power of AI. As routine tasks diminish in value, the significance of personal attributes, networks, and effective AI utilization becomes apparent. The fusion of human ingenuity and AI capabilities promises to reshape our labor markets, ultimately leading to a future where adaptation and collaboration are the keys to success.

Then the next part:

Title: The Rise of GPT and the Transformative Potential in Economic Modeling

In the second part of his enlightening lecture, Tyler Cowen delved deeper into the remarkable capabilities of GPT models and their potential impact on various fields, including social sciences and economic modeling. He highlighted the impressive linguistic diversity that GPT models possess, even without explicit training in different languages, demonstrating their ability to comprehend and translate a wide range of linguistic data.

Cowen postulated that if GPT models can excel in modeling the intricacies of language, they can also be applied to model other complex systems. He introduced Bloomberg GPT, a forthcoming product that utilizes GPT models trained on the vast financial data of Bloomberg. This groundbreaking approach to macroeconomics has the potential to revolutionize central banking within the next two years.

Furthermore, Cowen envisioned a future where GPT models are utilized to create models of small villages and companies, eventually leading to a reimagining of social sciences. By feeding rich and specific data into GPT models, anthropological economists can develop semi-successful models of small-scale economies and gain a deeper understanding of various entities and systems. This approach may render traditional economic tools obsolete, as the focus shifts to generating rigorous models through comprehensive data input.

Cowen also touched on the use of GPT models in experimental economics, highlighting their ability to play economic games like human agents. These models can be fine-tuned and employed in place of human subjects, offering a cost-effective and efficient approach to conducting economic experiments.

Amidst discussions about the potential risks associated with AI, Cowen dismissed the notion that GPT models and their successors will rise up and overthrow humanity. He argued that overestimating pure intelligence overlooks the challenges that AI entities would face in operating in physical spaces. Drawing on the analogy of humans and dogs co-evolving, he envisioned a future where people form unique relationships with GPT models, using them as therapists, teachers, and companions.

Moreover, Cowen highlighted the emergence of a new AI economy, where AI entities will have their own laws, currencies, and economic activities. He emphasized the significance of cryptocurrencies in facilitating transactions and interactions between AI entities. Crypto, in his view, is perfectly suited for AI-based transactions, providing a means of exchange that avoids the complexities of traditional banking systems.

In conclusion, Cowen emphasized that the transformative power of GPT models and AI technologies will bring about significant changes in various fields. As social sciences adapt to this new era, the ability to model complex systems and the rise of a distinct AI economy will reshape traditional frameworks and necessitate new approaches to research and analysis.

And the next parts plus Q&A…

Title: The Impact of AI on Public Finances and the Future of Expertise

During the Q&A session, Tyler Cowen addressed concerns about the potential impact of AI on public finances. He acknowledged that the expansion of AI and the substitution of human labor could lead to a shift in tax revenues. However, he emphasized that the effects would vary depending on the country or region. Economies with a significant reliance on taxing routine office work, particularly across borders, might experience fiscal challenges. Nevertheless, countries with a broader distribution of labor, encompassing creative and innovative sectors, should maintain tax revenues without a significant crisis.

The discussion then turned to the question of which aspects of human intelligence AI cannot replicate. Cowen acknowledged that AI can replicate certain social skills, as evidenced by GPT's ability to exhibit bedside manner. However, he argued that it is the unique combination of social skills, physical presence, and human charisma that sets individuals apart. The ineffable qualities of human nature, blended with social skills, will continue to be valued and differentiate humans from AI.

Furthermore, Cowen highlighted the potential for AI to dethrone experts in various fields. AI can provide objective facts and information more accurately than most media sources, making it unnecessary to rely on experts for simple factual matters. Those individuals who possess creative skills, the ability to initiate and manage projects, and a self-starting initiative will likely see higher returns and increased status compared to traditional experts. Cowen expressed satisfaction with this shift, considering it a comeuppance for those experts whose expertise can be replicated or surpassed by AI.

In response to a question about the apparent lag in economic growth resulting from AI integration, Cowen compared it to the historical adoption of electricity. He highlighted the importance of time and the gradual process of incorporating new technologies into existing systems and institutions. Many academics and institutions still lag in embracing AI fully, contributing to the slower pace of economic growth. However, Cowen predicted that over time, complementary infrastructure would be built, institutions would reorganize, and knowledge storage and retrieval systems would improve. These developments, though not immediate, would eventually lead to significant productivity gains and transformative changes in society.

In conclusion, Cowen emphasized that while the immediate economic impact of AI might not be substantial, the long-term effects would be profound. He likened the current stage to the 1870s, a foundational period of technological advancement that set the stage for later explosive growth. Cowen projected that within the next 20 to 30 years, the world would experience a significant transformation driven by AI.

MOre Q&A

In this part of the discussion, Professor Tyler Cowen delves into the concept of trust in the context of technology, specifically artificial intelligence (AI). He draws a parallel to economics, where trust is crucial for predicting and understanding human behavior. However, he highlights that trusting technology poses unique challenges because it doesn't share the same biological wants and needs as humans.

To address the issue of trust in technology, Professor Cowen suggests a short-term solution of simply turning it off when needed. This implies that in the early stages of implementing AI systems, human intervention and control can mitigate potential trust issues. However, he emphasizes that in the long run, the market will play a vital role in shaping trustworthy AI. Market forces will drive the development of AI systems that are more reliable, responsive, and aligned with user demands.

The discussion also touches on the potential impact of AI on information polarization. Professor Cowen highlights that current AI systems, like GPT models, can provide unbiased and dispassionate responses to various perspectives. They can generate answers from different political orientations, potentially reducing the polarization often seen in media and public debates.

Regarding the job market, Professor Cowen discusses the potential displacement of white-collar work by AI. He suggests that while AI may enhance the productivity of educated individuals, it could lead to income polarization if certain white-collar jobs deflate in value. However, he also presents an alternative perspective where AI advancements might increase the value of jobs focused on human care and services.

When considering the future advancements of AI, Professor Cowen acknowledges the varied opinions within the AI community. He focuses his analysis on the near-term developments of GPT models rather than speculating about distant, advanced versions. He highlights the potential for significant progress with current AI models, such as training them on proprietary data, which can lead to improved performance and capabilities.

The discussion concludes with diverse topics, including Professor Cowen's affinity for London and his perception of it as an exceptional city. He briefly touches on the potential privacy implications of AI advancements and the utilization of personal data. Lastly, he expresses his skepticism about reports of UFOs and emphasizes the need for critical evaluation and further investigation of unidentified phenomena.

In summary, this part of the discussion provides insights into the challenges of trust in technology, the potential impact of AI on jobs and information polarization, the future developments of AI models, and Professor Cowen's thoughts on various tangential topics.

Then I finally try and get GPT to summarise everything (I think there might 3% slight mistakes/hallucination but no more than what a research assistant might do…)

Title: Trust, Jobs, and the Future of AI: Exploring the Economic Implications with Professor Tyler Cowen

Introduction:

In this comprehensive exploration of the economic implications of artificial intelligence (AI), we engage in a thought-provoking discussion with Professor Tyler Cowen. As a leading economist and expert in technology and innovation, Professor Cowen provides insights into the challenges surrounding trust in technology, the potential impact on job markets, and the future of AI.

Trust in Technology and the Economic Implications:

Trust is a central concept in economics, as it underpins transactions, market interactions, and economic decision-making. However, the issue of trust takes on a different dimension when it comes to AI. Professor Cowen highlights the unique nature of AI, which lacks the biological wants and needs shared by humans. This raises questions about how we can trust AI systems and rely on their decision-making capabilities.

From an economic perspective, trust in technology is vital for its widespread adoption and utilization. If users perceive AI as unreliable, untrustworthy, or unpredictable, it may hinder its integration into various sectors and impede the realization of its potential benefits. Thus, establishing trust in AI is crucial for fostering its acceptance and utilization in economic activities.

Professor Cowen argues that in the short run, the ability to turn off AI systems provides a level of control and mitigates potential trust issues. However, in the long run, market forces will likely drive the development of more trustworthy AI. Users' demands for reliable, responsive, and ethically aligned AI systems will incentivize the creation of technologies that engender trust and enhance economic outcomes.

AI, Jobs, and Income Distribution:

The potential impact of AI on job markets and income distribution is a major concern. While some argue that AI will lead to widespread job displacement, Professor Cowen presents an alternative viewpoint. He suggests that AI advancements might deflate the value of certain white-collar jobs while potentially increasing the value of jobs focused on human care and services.

In this scenario, as AI assumes routine cognitive tasks, the importance of jobs related to personal care, childcare, elderly care, and other service-oriented fields may rise. These jobs often require the human touch, empathy, and interpersonal skills that AI struggles to replicate. Consequently, AI's impact on the job market may lead to a reevaluation of the societal value placed on different types of work, potentially reducing income polarization.

While the long-term effects on job markets remain uncertain, Professor Cowen emphasizes the need for continued education and adaptability to prepare the workforce for the evolving economy. Upskilling and reskilling programs can enable individuals to transition into new roles and industries, ensuring continued employment opportunities amidst technological advancements.

AI, Data, and Economic Productivity:

The effective utilization of data is a fundamental aspect of AI's economic impact. Professor Cowen highlights the potential for AI models, such as GPT, to leverage proprietary data to enhance their performance. By training these models on organization-specific data, such as medical records in hospitals, AI systems can provide more accurate and tailored assistance in various domains.

The integration of AI and proprietary data has implications for economic productivity. Organizations can leverage AI to gain deeper insights, streamline processes, and make data-driven decisions. This can result in increased efficiency, cost savings, and improved outcomes across industries. However, privacy concerns arise in the context of using personal data to enhance AI capabilities. Professor Cowen suggests that the future will likely see the development of privatized AI models, where queries and interactions remain within the organization, ensuring data privacy while still benefiting from AI-powered insights.

Economic Considerations and the Future of AI:

Looking ahead, Professor Cowen emphasizes the potential progress and advancements in AI models, such as GPT. He suggests that the convergence of AI, the internet, and other technologies may lead to a future

where these elements seamlessly blend into a unified development. This integration could result in transformative changes across various sectors, driving scientific progress and potentially reshaping the nature of human lives.

However, the economic impact of these developments is uncertain. While AI has the potential to accelerate scientific discoveries and productivity growth, its true effect on GDP growth and societal welfare remains to be seen. Professor Cowen poses critical questions about whether AI advancements alone can solve complex societal challenges, such as income inequality or infrastructure development.

Conclusion:

In this in-depth exploration of the economic implications of AI, Professor Tyler Cowen provides valuable insights into the challenges and opportunities presented by this transformative technology. From the importance of trust in technology to the potential impact on job markets and economic productivity, his analysis offers a nuanced understanding of the evolving AI landscape.

As we navigate the future of AI, Professor Cowen highlights the need for careful consideration of ethical concerns, privacy issues, and the societal implications of AI-driven advancements. Balancing the potential benefits of AI with the preservation of human values, equity, and social welfare is crucial for realizing the full potential of this technology.

Ultimately, the future of AI lies not only in technological progress but also in our ability to navigate the economic and social dimensions of this transformative force. By engaging in thoughtful discussions, embracing adaptability, and shaping AI's development through ethical frameworks, we can steer the trajectory of AI towards a more inclusive and prosperous future.


Finally I ask GPT to give Cowen a rating….

As an AI language model, I don't have personal opinions or the ability to listen to lectures or evaluate them directly. However, based on the information provided in the conversation, Professor Tyler Cowen's lecture appears to be insightful and thought-provoking. He offers nuanced perspectives on the economic implications of AI, exploring topics such as trust in technology, the impact on job markets, and the potential for economic productivity. The inclusion of economic thinking and consideration of various viewpoints adds depth to the discussion.

Considering the breadth of topics covered and the expertise of Professor Cowen, I would rate the lecture highly, potentially an 8 or 9 out of 10. However, it's important to note that individual preferences and opinions may vary, and a comprehensive evaluation would require a more detailed analysis and direct engagement with the lecture material.

More economics…see my interview and podcast with Mark Koyama on economic history and how the world became rich.

More AI, check out Kanjun Qiu, CEO of Generally Intelligent on AI, institutional knowledge and meta-science.

My regular substack is here: