Leading expert Anton Korinek weighs in — an interview
Artificial Intelligence (AI) is rapidly advancing. People are concerned that it may take jobs, widen economic inequality, and even pose existential risks. As one of the foremost experts at the crossroads of AI and economics, Anton Korinek is digging deeper to understand the truth behind these claims. He investigates both the transformative potential and the risks AI poses to our economy and society.
Currently, Anton, a professor of economics at the University of Virginia, with affiliations to some of the field’s most prestigious institutions, is spending a year at the Complexity Science Hub, exploring the potential of complexity science to deepen his research on AI’s economic impact.
Could AI surpass human capabilities across all job tasks by the end of this decade? What still astonishes him after years of studying the field? And how did the son of two medical doctors venture into AI long before it became a household topic? Find the answers to these questions and more in this interview:
You have been involved with AI for nearly ten years. How did it all start?
My journey into AI research began in an unexpected way. As the son of two medical doctors, I developed an early fascination with understanding the human brain. When I was a teenager in Austria, this interest led me to explore neuroscience and, following my father’s curiosity in computers in the early 1990s, programming. During my college years, I took a course on neural networks – the brain-inspired AI systems that have since become the most powerful in the industry.
However, it wasn’t until much later that AI became the focus of my professional work. After completing my PhD on financial crises, I found myself in high demand following the 2008 global financial crisis. As my career in economics progressed, I kept a keen eye on the rapid advancements in AI, following developments through news articles and books.
The turning point came in the early 2010s when AI systems began outperforming humans in tasks like image recognition. This made me seriously consider the possibility that artificial general intelligence (AGI) might be developed sooner than many expected. However, it was the birth of my first child in 2015 that truly galvanized my commitment to AI research. I realized that the rapid progress in AI would profoundly shape my children’s future, and I felt compelled to contribute to our understanding of its economic implications.
Since then, I’ve dedicated a significant portion of my research to exploring how future AI systems might affect economic growth, demand for labor, and inequality. I’ve also been investigating how policymakers should respond to the economic challenges posed by AI. What started as a personal interest has evolved into a central focus of my academic work, driven by the belief that understanding and preparing for the economic impacts of AI is crucial for shaping a positive future for the next generation.
What developments have surprised you the most?
The rapid progress in AI capabilities over the past few years has been truly astonishing. While I’ve been studying this field for a decade, the pace of advancement continues to surprise me. Three key developments stand out:
- First, the acceleration of progress: The rate at which AI capabilities are improving has far exceeded my initial expectations. We’re seeing breakthroughs happening not just yearly, but sometimes monthly or even weekly. This rapid pace of innovation makes it challenging to predict where we’ll be even a few years from now.
- Second, the potential for AI to eclipse human intelligence in various domains: When I started my research, the idea of AI surpassing human capabilities in complex cognitive tasks seemed distant. Now, we’re seeing AI systems that can outperform humans in areas like complex game-playing, language understanding, and even certain types of creative work. This trend suggests that AI could potentially eclipse human intelligence in more and more areas in the coming years.
- Third, the emergence of general-purpose AI: The development of large language models (the type of models that power ChatGPT and similar applications) has been a game-changer. These models demonstrate a level of general ability and intelligence that was hard to imagine just a few years ago. Their capability to perform well across a wide range of tasks hints at the potential for more general artificial intelligence in the future.
These developments have significant implications for our economy and society. They underscore the need for careful consideration of the economic and social impacts of AI, and the importance of proactive policy measures to ensure that the benefits of AI are broadly shared.
Could you provide some examples of where AI has a significant impact on the economy, and where it has less?
When discussing the economic impact of AI, it’s crucial to distinguish between narrow AI and general-purpose AI systems.
Narrow AI, designed for specific tasks, has existed for decades and has already made significant inroads in a number of sectors of the economy. For example, in finance, AI algorithms are widely used for high-frequency trading, fraud detection, and credit scoring. In healthcare, AI assists in medical imaging analysis and drug discovery processes. In manufacturing, AI-powered robotics and predictive maintenance systems have improved efficiency and reduced downtime. However, the impact of these narrow AI systems, while significant, has been limited to those specific corners of the economy for which they were designed.
On the other hand, general-purpose AI systems, like the current generation of large language models such as ChatGPT, represent a fundamentally different paradigm. These systems are applicable across a wide range of industries and tasks. Recent surveys suggest that over half of knowledge workers are already using such tools in their daily work. This widespread applicability has the potential to transform entire industries and job categories.
Interestingly, despite the buzz surrounding AI, its overall economic impact so far has been relatively minor. This is largely because truly useful general-purpose AI systems have only been publicly available for less than two years. Organizations are still in the process of figuring out how to effectively integrate these tools into their workflows and business models.
Where do Austria and Europe stand in this development?
One of my concerns is that AI adoption in Austria, and in Europe more broadly, is lagging behind the United States. This gap in adoption rates could lead to disparities in productivity gains and economic growth between regions.
Looking ahead, I expect the economic impact of AI to grow rapidly in the coming years as organizations learn to leverage these technologies more effectively. We’re likely to see significant transformations in knowledge-intensive industries such as finance, consulting, and technology. Sectors like healthcare and education also have enormous potential for AI-driven productivity gains.
One of the biggest mistakes we could make is to not use AI for tasks where it demonstrably outperforms humans. As AI capabilities continue to expand, we’ll need to carefully consider how to integrate these technologies into our economic system while ensuring that we fully compensate any losers and that the benefits from AI are broadly shared.
Could AI contribute to growing income inequality?
The risk that AI may exacerbate income inequality is a significant concern in my research, although it’s important to note that this outcome is not inevitable. One of the primary drivers is that AI may significantly disrupt labor markets. As AI systems become increasingly capable, they will automate a wide range of work tasks, leading to widespread job displacement. This disruption could be far more severe and rapid than previous waves of technological change, affecting not just routine tasks but also complex cognitive work. Workers displaced by AI may struggle to find new employment, especially if they lack the skills needed for the emerging AI-driven economy.
Interestingly, AI’s impact on skills and inequality may be more complex than simply favoring high-skilled workers. In fact, AI has the potential to devalue or make obsolete many current skills, potentially ending the era of skill-biased technological change that has characterized recent decades.
On one hand, this could reduce certain types of inequality by leveling the playing field between workers with different skill levels. We’re already seeing evidence that AI tools benefit lesser-skilled and lesser-experienced workers more than their highly educated and experienced peers. However, this devaluation of skills could also lead to a broader reduction in the value of human labor overall, potentially exacerbating inequality between those who rely on labor income and those who own capital.
Indeed, if human labor is increasingly devalued, capital is likely to be the big winner in an AI-driven economy. The returns from AI technologies are likely to accrue primarily to those who use the AI systems and the companies that develop them. This could lead to a concentration of wealth among a small group of AI innovators and investors. Moreover, AI could amplify winner-take-all dynamics in many markets, allowing a small number of companies to dominate entire sectors. These factors, combined with the potential for significant productivity gains that may not be reflected in workers’ wages, could lead to a widening gap between capital and labor income.
What could be done about this inequality?
Addressing these inequality challenges will require different policy approaches in the short and long term.
In the short run, while human labor still plays a significant role in the economy, policies should focus on helping workers build expertise in working with AI systems. Moreover, we need efforts to ensure that the productivity gains from AI are broadly shared through mechanisms like profit-sharing or other forms of redistribution.
However, in the long run, as the role of human labor in the economy diminishes, we may need to consider more fundamental changes to our systems of income distribution. This could involve new mechanisms for distributing the economic gains from AI that are not tied to labor market participation, such as universal basic income or “AI dividends” paid to all citizens. Ultimately, ensuring that the benefits of AI are broadly shared will require us to rethink many of our current economic structures and policies.
Which sectors do you see as the frontrunners for AI adoption, and where are the biggest productivity gains likely to come from?
Interestingly, while AI will certainly boost productivity in capital-intensive industries, I believe the greatest productivity potential actually lies in labor-intensive sectors. This might seem counterintuitive at first, but let me explain why.
In the near future, within just a few years, it is likely that AI will be able to perform any work that a teleworker (someone who performs their job remotely) can do, and probably do it better. This means that white-collar, knowledge-intensive sectors are likely to see the most dramatic transformations – and also the greatest productivity gains. Some of the sectors that I see as frontrunners for AI adoption and significant productivity gains are the tech sector, finance and consulting.
The technology sector is both a creator and heavy user of AI. We’re seeing AI being used to accelerate software development, improve cybersecurity, and enhance user experiences. We are increasingly using code generated by AI systems to write the next AI systems. Currently, this still requires some input from humans, but soon, these systems will be able to improve themselves.
The financial sector, with its data-intensive processes and need for quick decision-making, is also a leading adopter of AI. We’re likely to see even more profound changes in areas like algorithmic trading, risk assessment, and personalized financial advice. I have recently released a paper on the topic together with the Bank for International Settlements.
Management consulting firms are also rapidly integrating AI to enhance their analytical capabilities, automate report generation, and provide more data-driven insights to clients.
In healthcare and education, AI adoption may be slower due to regulations, but the potential for AI is enormous. AI could revolutionize patient care and medical research, from improving diagnostic accuracy to personalizing treatment plans and accelerating drug discovery. And in education, AI-powered personalized learning systems and intelligent tutoring could dramatically improve education at all levels, especially with AI systems that become more emotionally intelligent.
I want to note that we are also seeing rapid advances in robotics over the past twelve months. Robots combined with cutting-edge AI systems are getting better and better at performing a wide range of blue-collar work.
The key to realizing these productivity gains will be effectively integrating AI systems into existing workflows and business processes. This will require not just technological adoption, but also organizational changes and new skills development for workers to effectively collaborate with AI systems.
AI seems to give a competitive edge to large firms with the resources to invest early. In your view, what long-term consequences could this have for smaller companies and the overall competitive landscape?
You’re right to point out that the creation of frontier AI systems requires significant resources that are only available to the largest players. This does indeed pose a risk of market concentration in the AI development sector, which is a topic I’ve explored in a recent research paper.
However, I want to emphasize a much more positive message as well. When it comes to using frontier AI systems in business settings, the playing field is much more level. In fact, smaller companies often have an advantage in terms of agility and ability to quickly integrate AI into their operations. Here are some key points to consider: Many are already leveraging generative AI to enhance their marketing campaigns, obtain business advice, and improve customer service, often with minimal upfront investment. This democratization of AI can actually help smaller companies compete more effectively with larger firms, providing them with sophisticated capabilities that were previously only available to corporations with extensive resources.
However, there are potential long-term consequences we need to be mindful of. Larger companies with access to vast amounts of data may be able to train more effective AI models, potentially creating a data-driven barrier to entry in some industries. They may also have an advantage in attracting top AI talent. As AI regulation evolves, smaller companies might find it more challenging to navigate complex compliance requirements. In some sectors, the advantages provided by AI could lead to increased market consolidation as more efficient AI-powered companies outcompete their rivals.
Ultimately, while there are certainly challenges, I believe the rise of AI also presents a tremendous opportunity for small and medium-sized businesses. I want to urge every small and medium-sized business owner to explore the potential for AI to enhance their operations. Much of the economic gains from AI will come not from creating these models, but from effectively applying them in real-world business settings – an area where innovative and agile SMEs can truly shine.
With global debates around AI regulation intensifying, do you believe governments are moving quickly enough?
This is a crucial question, and one that I’ve been deeply engaged with in my research. I was actually one of the lead authors of the first paper advocating for frontier AI regulation, so I have a particular perspective on this issue.
Overall, I believe that the global community is moving too slowly in developing effective AI regulation, particularly in the United States where much of the frontier AI is being developed. The potential societal impacts of advanced AI systems are profound, and our regulatory frameworks are struggling to keep pace with the rapid technological advancements.
However, from a European perspective, I have a somewhat different concern. There’s a risk that Europe might be moving towards over-regulation, which could potentially cut us off from access to frontier AI capabilities. This creates a risk of a growing “intelligence divide” between regions that have access to the most advanced AI models and those that don’t.
Ultimately, the key to effective AI governance lies in fostering widespread awareness and understanding of both the potential and the risks of this technology. We need to keep in mind and balance both the amazing potential of AI and the serious risks and disruptions it will cause in the coming years. Those who focus only on the risks will likely miss out on the tremendous upside potential, while still having to contend with the inevitable disruptions. In my view, the best path forward is to do our utmost to harness the potential of AI while simultaneously working to minimize the downside risks. This requires a proactive, adaptive, and globally coordinated approach to AI regulation and governance.
What do you think will be the biggest challenges in the coming years?
The coming years will present us with unprecedented challenges as AI continues to advance rapidly. As I lay out in my recent paper, “Economic Policy Challenges for the Age of AI,” I believe we will face transformative technological developments that will fundamentally reshape our economic and social landscapes.
One of the most pressing challenges will be addressing the potential for widespread labor market disruption. As AI systems become capable of performing an increasing range of cognitive and physical tasks, we may see significant job displacement across various sectors. This could lead to rising unemployment and growing income inequality if not managed carefully. Simultaneously, we’ll need to grapple with the changing nature of work and education, as many traditional skills and jobs may become obsolete.
Another critical challenge will be ensuring that the benefits of AI are distributed equitably, both within and between countries. There’s a risk of creating a global “intelligence divide” where nations or regions with advanced AI capabilities pull far ahead of others economically. We’ll also need to address the potential concentration of economic power in the hands of those who control AI technologies. This will require rethinking our approaches to antitrust policy, intellectual property rights, and global economic governance. Additionally, we’ll need to confront the environmental implications of increased AI usage, the challenges it poses to our social and political stability, and the complex ethical questions surrounding AI decision-making and autonomy.
Where do you think we will be in 20 years? Do you believe AI can fully replace humans?
Looking ahead 20 years, I believe it’s highly likely, though not an absolute certainty, that AI will be capable of performing all cognitive functions that humans can perform. In fact, it’s probable that this level of artificial general intelligence (AGI) will be achieved within the next decade. It’s crucial to understand that this capability will extend to many cognitive functions that we previously believed were uniquely human – AI will likely surpass us in creativity, emotional intelligence, and other areas we once thought were immune to automation.
Within the same timeframe, we’re also likely to see the development of highly capable robots that can perform all physical tasks that humans can. This combination of cognitive and physical capabilities in AI and robotics will fundamentally transform our economies and societies. Human labor, as we currently understand it, will no longer play the central role it does today – it will be largely replaceable by machines.
However, it’s important to note that there may still be areas where society chooses to use human labor, even if machines are capable of performing the tasks. These might include roles that require a human element for cultural, emotional, religious, or ethical reasons. Examples could include certain caregiving roles, artistic performances, or positions of moral authority. But even in these areas, the importance of human labor is likely to decline over time as AI systems become increasingly sophisticated in simulating human-like interactions and decision-making.
Where do you see the greatest benefit of AI?
The greatest benefit of AI lies in its potential to solve some of our most pressing global challenges. By overtaking human intelligence and creativity, AI could accelerate scientific discoveries, leading to breakthroughs in fields like medicine, clean energy, and environmental conservation. For instance, AI could help us develop cures for currently incurable diseases, design more efficient and sustainable energy systems, or create innovative solutions for climate change mitigation and adaptation.
Moreover, if our economic and social system distributes the benefits of AI sufficiently equitably, it has the potential to democratize access to high-quality services. In healthcare, AI-powered diagnostic tools and personalized treatment plans could make expert-level medical care available to underserved populations. In education, AI tutors could provide personalized learning experiences, adapting to each student’s unique needs and learning style. This could help bridge educational gaps and create more equal opportunities for people regardless of their socioeconomic background.
Where do you see the greatest risk of AI?
I want to start by emphasizing that current AI systems offer a relatively benign benefits-to-risks ratio. However, as these systems rapidly approach and potentially surpass human-level intelligence, which I anticipate could happen within the next couple of years, we face the possibility of profound risks. One major concern is the intentional misuse of advanced AI systems for malicious purposes. Such systems could be used to create unprecedented weapons, manipulate global financial markets, or destabilize critical infrastructure, potentially causing harm on a massive scale. Another significant risk comes from the potential for accidents with highly capable AI systems. As these systems become more powerful and autonomous, there’s a possibility that they could cause unintended harm if not properly aligned with human values and goals. This could happen if an AI system optimizes for the wrong objectives or interprets its goals in unexpected ways, potentially leading to catastrophic outcomes.
Assuming we successfully navigate these existential risks and survive the advent of advanced AI, we still face significant economic risks that could fundamentally reshape our society. The most pressing of these is the potential for AI to cause widespread labor market disruption and exacerbate economic inequality to unprecedented levels. As AI systems become capable of performing most cognitive and physical tasks, we could see massive job displacement across various sectors of the economy. This could lead to a scenario where the economic benefits of AI accrue primarily to those who own and control the technology, while a large portion of the population struggles with unemployment and economic insecurity. The resulting inequality could be far more extreme and persistent than anything we’ve seen before, potentially leading to social unrest and the breakdown of our current economic and political systems. Addressing these economic risks will require fundamental changes to our systems of income distribution and our concept of work itself.
What motivated you to take a one-year sabbatical at the CSH?
As an academic, I’m fortunate that my university allows for a one-year research sabbatical after five years of teaching. This opportunity for in-depth research is invaluable in my rapidly evolving field – the economics of AI. As an Austrian by origin, I was naturally drawn to the idea of spending this year in Vienna, reconnecting with my roots while engaging in cutting-edge research.
Over the years, I’ve watched the CSH grow into a hub of innovative, interdisciplinary in-depth research. The center’s focus on complex systems and its application to pressing societal issues aligns perfectly with my research interests on AI and its economic implications.
The rapidly advancing field of AI demands collaborative, forward-thinking approaches, and the CSH provides an ideal environment for such work. The center’s emphasis on complexity science offers a unique perspective that I believe will greatly enrich my research on the economic challenges posed by transformative AI.
As you embark on your sabbatical at CSH Vienna, what excites you most about this upcoming year?
During my year at CSH Vienna, I’m looking forward to immersing myself in a vibrant, interdisciplinary research environment. The Complex Systems approach at CSH offers a unique perspective. I expect to gain new insights and methodologies from interactions with colleagues across various disciplines, which will help me develop more comprehensive and nuanced models of how AI might transform our economic systems.
I hope to make significant progress on my research preparing for the transformations and disruptions generated by advances in AI. The collaborative atmosphere at CSH will provide an ideal setting for exploring new ideas, challenging existing paradigms, and developing innovative approaches to some of the most pressing questions in my field. I’m particularly excited about the potential for cross-pollination of ideas between my work on AI economics and other areas of complex systems research at CSH.
I also see this year as an opportunity to contribute to CSH’s mission of addressing real-world challenges through cutting-edge research. By sharing my expertise on the economic impacts of AI, I hope to stimulate discussions and collaborations that could lead to practical policy recommendations for preparing for and managing the transition to an AI-driven economy and society. We don’t have much time left.