Conference Report
A two-day conference in Oct 2024 bringing together 30 leading economists, AI policy experts, and professional forecasters to rapidly evaluate the economic impacts of frontier AI technologies by 2030.
Part 1: Worldbuilding
The majority of Day 1 was spent conducting worldbuilding exercises. In these exercises, participants did backcasting and world-building to explain each of the three scenarios provided.
Worldbuilding around a common set of scenarios served two purposes at our conference. First, by reducing the problem space and having participants reason from shared assumptions rather than divergent priors, the exercises enabled deeper exploration of novel insights. Rather than debating foundational assumptions, experts could engage in thought-provoking discussions with other leading thinkers about implications and possibilities they might not have previously considered in their individual worldviews.
Second, these exercises helped participants develop and articulate coherent perspectives on AI's concrete societal impacts. By giving attendees dedicated time to thoroughly explore specific viewpoints, the exercises transformed abstract discussions about AI capabilities into detailed, tangible descriptions of how these technologies might reshape everyday life, economies, and social institutions.
Methodology
During the Worldbuilding exercises, attendees were placed at tables of 4-5 to conduct collaborative brainstorming, with one organizer at each table to facilitate the discussion and take notes. They engaged in two 45-minute writing sessions, where they created original worldbuilding results describing society in 2030 based on a self-selected set of questions. After each writing session, attendees engaged in a table discussion about their results.
During each 45-minute writing session, we asked that individual attendees spend roughly 15 minutes writing for each scenario. However, attendees had flexibility in terms of allocating their time to different scenarios, and many attendees chose to allocate more focus to Scenario 3.
The first writing session focused more on general-purpose worldbuilding (General Worldbuilding Exercise), predicting broad societal and economic topics, whereas the second writing session had attendees create predictions about their specific area of deep expertise (Deep-Dive Worldbuilding Exercise).
For the General Worldbuilding Exercise, we provided a wide range of questions that they could choose to answer, as well as a range of “creative structures” that would serve as the format for the attendee’s output. For example, a “creative structure” could be a “day-in-the-life narrative in 2030”, or “headlines from CNN in 2030”.
First, attendees reviewed the suggested questions and creative structures. We gave attendees some time to brainstorm original questions and creative structures that they could use.
Then, attendees chose a single “creative structure” and a set of questions that they would like to answer. Attendees were asked to evaluate their chosen questions across all three scenarios, so that their written output followed roughly the same format per scenario.
For the Deep-Dive Worldbuilding Exercise, attendees conducted roughly the same structure for their writing sessions. However, they focused on questions related to their specific domain of expertise, rather than discussing more broad societal and economic questions.
Following are the direct instructions given to attendees, as well as the questions and creative structures suggested initially:
Instructions for General Worldbuilding
Worldbuilding Potential Questions
Overall Questions
Economic Questions
Industry Sector-Specific Questions
Region-Specific Questions
Potential Creative Structures
Results
Our attendees created over 100 pages of original writing for our worldbuilding exercises, spread across 22 documents. You can read the full, unedited texts created by our attendees in our Appendix (All Worldbuilding Writeups). For conciseness and readability, we’ve summarized the most discussed eight themes we found from analyzing these exercises below.
Significant Unemployment
The threat of significant human unemployment is one of the most discussed potential economic impacts of AI. However, there is very little alignment among economists and AI experts on the likelihood of this threat occurring. Some attendees expect unemployment to increase rapidly and severely, as AI progressively outcompetes humans while costing much less. Others predict that unemployment due to automation will be softened as industries react to increased output by employing more humans in novel and non-automatable jobs, while governments will react to societal pressure by providing vastly greater support to the unemployed and impoverished.
Attendees did agree that the impact of AI will be sector-specific, with jobs in transport or tech-heavy fields like radiology likely to face rapid automation. Meanwhile, they expect that highly skilled physical tasks, like surgery or dentistry, and high-prestige roles, like legal experts, will be less likely to see automation by 2030, though they may still see significant AI task augmentation.
Scenario 1: Modest Unemployment with Significant Impact
Even under the mildest scenario, attendees predict a hollowing out of bottom and middle-skilled labor roles, especially in creative industries like art and design where demand and thus wages will fall: “[The] high-end are productive using AI tools, but middle-of-the-road design wages (e.g. on Upwork) are much lower, and you’re mostly paying for someone to supervise AI systems, manage clients” (1C).
Scenario 2: Rising Unemployment Leading to Widening Inequality
The reduction of lower skilled labor roles is sharper than in Scenario 1, with less chance of counteracting forces prevailing over the negative impacts of unemployment. “Major increases in unemployment and declines in total hours worked as whole sectors of the economy are automated. There is some increase in employment in non-automatable tasks, but by no means matching the declines in employment in the automated tasks. The narrow group of winners includes only selected capital holders” (1A). However, the same attendees expect industries to react to growing economic output by employing more humans in the supply chain. “Thanks to self-driving technology [we will] see a lot more delivery of goods as a share of the economy. But, any steps that are not automated in this will employ more humans. Just as, thanks to robotics in warehouses today, and better tech-driven logistics, being a warehouse worker has grown as a job category” (1C)
Scenario 3: Drastic Economic Disruption
In this scenario, severe and still rising unemployment will lead to dramatic restructuring of the economy, likely too fast to be adequately mitigated by governments. “Employment in the US has dropped by 40%. This has created utter chaos. Governments are scrambling to rapidly expand social safety nets, and most have by now started implementing some form of UBI” (2A). “Most jobs become automated as AGI agents can perform both complex and simple tasks more cheaply and effectively than humans. The few remaining human roles focus on governance and oversight of AGI systems, though there are pressures for these to be automated too” (2B). Around the world, we’ll see “major increases in unemployment and declines in total hours worked as whole sectors of the economy are automated. The entire economy can be fully automated, and human labor is fully replaceable” (1A).
Common Sub-Themes:
Increasing Wealth Inequality
One of the most commonly discussed outcomes at Threshold 2030 was the magnification of wealth and income inequality. Attendees predicted that the rising capabilities and perceived potential of AI will lead to commensurate rises in the market value of leading AI firms. This would concentrate wealth to those who own and develop AI systems - primarily large tech companies, their shareholders, and highly skilled professionals. Meanwhile, they argued that automation-driven unemployment will worsen the financial burdens of the poorer half of society. Some attendees went as far to suggest that this pattern could lead to the majority of individuals even in wealthy countries facing bitter poverty in extreme scenarios.
The effect of AI on inter-country inequality will heavily depend on the rate of diffusion of AI technology to lower income countries. With sufficiently rapid diffusion, we could see some flattening of inter-country inequality. However, most attendees predicted that Scenarios 2 and 3 are likely to exacerbate inequality as AI-leading countries capitalize on the capabilities of AI to automate and revolutionize many industries. This rising inequality would likely lead to major political pressure to support those negatively affected, and potentially to major changes to geopolitics and international relations. Whatever the rate of diffusion, these scenarios are likely to introduce new geopolitical dynamics and instability.
Scenario 1: Early Increases in Wealth Inequality
Wealth inequality may start to increase as the first signs of divergence show between AI-driven and human-driven economic output: “There will be significant divergences between: corporate productivity and individual average human productivity; productivity and wages; human consumption and GDP; human welfare and economic growth; asset prices and value to humans. The human economy and the AI economies will diverge, as the financial and main street economies have diverged in and after the financial crisis” (3C)
Scenario 2: Wealth Increases Are Captured by an Oligopoly
GDP will rise within AI-leading countries, but this wealth will largely be captured by relatively few individuals and an oligopoly of companies: “Ownership of labor [will be] dispersed but ownership of capital and AI [will be] heavily concentrated” (1A). Much of this wealth generation and capture will be in the form of the equity of leading companies, which would rise drastically in anticipation of future economic value.
Most attendees agree AI is likely to worsen inter-country inequality, though some argued that it might reduce or at least alter traditional gaps between nations: “The greater automation is likely to lead to a flattening of occupations, with less differentiation between nations. Nevertheless the lower degrees of education and, importantly, technological literacy leads to significant inequality between nations in 2030” (2C).
Scenario 3: Radical Wealth Inequality & Imbalances Between HICs and LICs
AI will continue to increase global wealth and economic output while drastically exacerbating wealth inequality, both within countries and between them.
Uneven global diffusion of AI will worsen inequality between higher and lower income countries, as the majority of wealth generated by AI by 2030 will be captured by companies in AI-leading countries: “More advanced countries find different ways to maintain their economic superiority, using the race advantages of having achieved AGI first to ensure they maintain a competitive advantage. Strong reliance of weaker countries on tech companies, potentially to the extent of becoming puppet states” (2C).
Within AI-leading countries, one attendee suggested the rapid spread of automation and rising capital share will lead many to face bitter poverty as the demand for human labor crashes without the infrastructure to support those affected. “The average person in a major city will be subsistence living as if in a favela. They will rely on AI interfaces to companies and government for basic services, information services, and social services" (3C). Another attendee proposed the following 2030 headline: "The Economist: 'The 0.001%: How AI Ownership Created a New Feudal Class'” (5A).
Common Sub-Themes:
Attendees discussed how AI-driven automation and the disproportionate capture of AI-generated wealth will affect wages and labor share - the portion of national income that goes to wages, as opposed to income from capital assets like investments, real estate, and corporations. Many different factors and scenarios will affect the balance between labor & capital. A plurality of attendees predicted that labor share will decrease in developed countries, though some argued the balance would remain stable or even shift in favor of a higher wage share. Attendees agreed that the exact balance and scale and speed of change will depend strongly on the scenario and the strength of counterbalancing economic forces.
Scenario 1: Limited Impact on Labor Share
Under Scenario 1, AI primarily enhances existing economic structures without fundamentally altering wage dynamics. Some attendees predict “Decreasing labor income share, increasing capital and profit shares (markups) within developed countries” (1A), while others argue that market forces provide resistance to falls in labor shares, including “higher rewards to interpersonal/empathetic/carework skills [and] physical labor” (1D).
Scenario 2: Diverging Labor Outcomes
In Scenario 2, the effect of AI on wages varies by sector; “wages for the top 10% in white collar work [and] highly-skilled manual laborers are rapidly increasing, while overall labor participation dramatically decreases“ (1B). Workers that are able to utilize AI and to pressure for higher compensation could similarly see rising wages: “Workers who command AI systems become more productive. Especially since it’s only 2030 [...] workers will find it strange and galling to only get ~2% raises per year while profits and capital costs go up 90%, say. In the short run at least [...] there will be appetite to share this bounty with workers” (1C).
However, in the rapidly growing domain of automatable tasks, “[the relative price of AI] vis a vis the human worker will decline, triggering large-scale replacement [leading to] rapidly decreasing labor income share, increasing capital and profit shares (markups) within developed countries” (1A).
Scenario 3: Dramatic Decline in Labor Share
In Scenario 3, labor share could fall off a cliff: “Labor income share falls towards zero. Wages are growing, at best, at a systematically lower rate (maybe an order of magnitude lower rate) than aggregate GDP or GDP/capita” (1A). However, some sectors may be unsuitable or at least unpalatable for AI replacement and thus will be protected against automation and falling labor conditions: “people with physical jobs, e.g. in nursing have started to demand large wage increases and are working fewer days a week” (2A).
Common Sub-Themes:
Cost of Goods & Services
A key question is how AI advancement might affect the prices of goods and services across different sectors of the economy. Attendees analyzed potential price changes across scenarios, considering factors like labor content, resource requirements, and market structure. We saw several key themes regarding AI impacts on the cost of goods and services: deflationary pressures in sectors where AI can effectively substitute for labor, potential limitations in how these cost savings translate to consumer prices, and complex interactions between price changes and demand responses.
A recurring observation emerged that price effects would likely vary substantially across sectors, with digital and knowledge-intensive services seeing earlier and larger price reductions than sectors dependent on physical resources or human interaction.
Deflationary Pressures from Productivity
Attendees anticipate that AI-driven productivity improvements are expected to exert downward pressure on prices throughout the economy, although the extent and timing of these effects will differ by sector. As AI systems reduce labor and operational costs, this deflationary effect is likely to become increasingly significant.
Inelastic Demand Limiting Growth Effects
Even with falling prices, attendees highlighted that demand may not increase proportionally, leading to potential economic imbalances. As one attendee observed, "Demand for goods and services does not make up for the decreased price levels" (3C) This highlights a scenario where reduced prices driven by AI productivity gains might not translate into sufficient consumption growth to sustain current economic activity levels.
This inelastic demand could shrink overall spending in the economy despite rising productivity. Attendees emphasized that these dynamics may result in uneven impacts, with digital services benefiting from cost reductions and increased adoption, while commodities and real estate remain relatively insulated from such price effects.
Ambiguous Real Income Effects
Attendees expressed mixed views on real income impacts. While some expect increased access to goods and services for lower and middle-income consumers in Scenario 2, others suggest these benefits might instead concentrate among specific groups. In particular, the gains might flow mainly to companies with strong market positions and workers whose skills complement AI systems, while workers whose jobs can be replaced by AI might see decreasing wages.
Scenario 2 could bring more consumer savings as AI adoption expands, but another attendee remarked that these benefits are unlikely to create "a step level change in their well-being," (3B) with gains primarily concentrated among companies with strong market positions and workers whose skills complement AI systems.
Other Takeaways
Attendees identified clear sectoral variations in AI's price impact. They expect human interaction in services like healthcare and customer service to become luxury goods, while digital services become more affordable.
Transportation costs could decrease significantly with self-driving vehicles, as one attendee noted, "Transport is much cheaper, so people live further out from the cities. Land in suburbs becomes more valuable." (1C)
For small organizations, attendees note that automation might reduce capital expenditure requirements, though the competitive impact would depend on implementation costs. Several attendees emphasize that economic impacts would likely diffuse gradually through the global economy, with developed economies experiencing effects before developing ones.
Rate of Diffusion
The global impact of AI in 2030 will strongly depend on the level of global diffusion - that is, the extent to which AI will be practically distributed and employed by governments, businesses, and individuals. Diffusion will doubtless remain uneven in 2030 regardless of scenario; rather, the debate concerned precisely how each scenario would affect and be affected by diffusion. A commonly referenced example was India, which currently demonstrates a complex mix of technological diffusion: many areas lack electricity, while in others modern technology and communications are fully incorporated into daily life and business (5C). These factors will strongly determine the rate of diffusion; without infrastructure, AI tools cannot be adopted, regardless of their utility.
Scenario 1: Minimal Diffusion
Diffusion in this scenario is limited, and control and use of AI may even become increasingly concentrated to large, established companies. One attendee suggested that “Media outlets [will] speak of the AI bubble bursting and funding starts drying up” (2A), at least within AI-leading nations. Elsewhere, AI tools will start to be adopted by those with the infrastructure to use it: “Countries with large IT industries such as India will quickly adopt these technologies not only to increase their productivity but as a way of outsourcing AI-supervised tasks” (2C).
Scenario 2: Rising Diffusion Driving Inequality
Diffusion will accelerate, with wider access both within and between countries. However, LICs will still lag in adoption. “Highly economic development counties have much higher integration than others“ (5A).
Scenario 3: Unpredictable & Locally Rapid Diffusion
Though attendees predict wider and more rapid diffusion in Scenario 3, there will remain many economic sectors unsuited for AI assistance by 2030. “At the end of 2030, the diffusion of AI capabilities across sectors and across countries is highly uneven and fairly limited [...] especially on the global scale. While some political and geopolitical impacts around the balance of power etc might have already taken place by Dec 2030, a lot of the pure economic impacts are still ahead of us, including the effect on prices, though there might have been some pretty impressive drops in sectors that were poised to adopt and adapt very quickly. ” (3B).
Several attendees found that the diffusion of physical AI services, including robotics and other physical tools, is unlikely to be significant by 2030 due to their large infrastructure requirements. “Seems hard to believe we will create 1B humanoid robots by 2030, even if the 2030 level of AI basically could do most of the R&D and manufacturing itself; it takes time to develop, test, etc” (1C).
Common Sub-Themes
Transformed Voting & Governance Systems
Attendees explored how governance mechanisms might evolve, from AI-augmented existing systems to potentially revolutionary new forms of democratic participation. While traditional elections persist through 2030, the integration of AI into governance processes raises fundamental questions about democratic legitimacy, power distribution, and the changing nature of political representation.
Scenario 1: AI as Support Tools for Traditional Democracy
In Scenario 1, AI primarily enhances existing democratic processes without fundamental structural changes. "AI is used as technology for bureaucratic processes...but does not substantially enter decision making processes" (1B). While AI improves voter access to information through "decision support tools... curating news sources for highly informed voters," the basic architecture of democratic institutions remains intact.
The private sector experiments with automated decision-making, but "few of these experiments have turned out to be successful, and none have emerged as a market leader in a major industry" (1B), suggesting limited immediate pressure for governance innovation.
Scenario 2: Hybrid Human-AI Governance Emerges
By Scenario 2, AI becomes deeply integrated into governance while maintaining human authority. Several attendees highlighted the emergence of new democratic tools and processes. "Vetted independent civic-agents help voters understand political agendas from candidates, and discuss and think through proposed policies" (2B), while AI "removes bottlenecks on direct constituent input into decision-making" (4B).
This scenario sees the rise of "automated decision support" systems that can "suggest good potential policies" while reducing "cognitive drain on leadership" (4B). Notably, these developments enable "increased capacity for good governance in grassroots, collectively governed, democratic, consensus-based organizations", suggesting a potential democratization of complex decision-making processes.
Scenario 3: Fundamental Restructuring of Political Systems
In Scenario 3, governance systems undergo dramatic transformation. "Decisions are made by AI systems" while humans retain nominal authority. "Politicians now mostly serve as the hands and feet of AI systems. Their agendas are largely crafted by AI systems to maximize voters while staying somewhat true to a specific ideology" (2A).
The voting system itself evolves, with "countries experimenting with ranked choice, quadratic voting" (2B). Citizens gain the ability to "outsource certain votes (local elections, transport, bin collection etc) to agents entirely," though "for more consequential decisions, human voting is critical" (2B). This creates a two-tier system of democratic participation, with routine decisions increasingly automated while maintaining human oversight for fundamental choices.
Common Sub-Themes
Attendees suggest several critical developments in the transformation of governance systems by 2030:
This analysis suggests that by 2030, democratic institutions will face fundamental challenges in balancing AI integration with maintaining meaningful human agency in governance.
Legal Status of AI Agents
As AI capabilities increase, questions about the legal and moral status of AI agents grow more pressing. Attendees explored how legal frameworks and ethical considerations might evolve across three scenarios, moving from AI as mere tools without rights to potentially recognizing agents as entities with moral standing. While formal human-like rights for AI remain unlikely by 2030, discussions highlighted the emergence of new legal constructs and the possibility that advanced AI systems may eventually receive protections similar to, or extending beyond, corporate personhood.
Scenario 1: AI as Tools without Rights
In Scenario 1, AI systems are considered sophisticated instruments rather than right-bearing entities. As one attendee explained, “models are not deemed to have any rights. As mere tools, there is no strong pressure or need for models to have any form of legal personality,” and although legal frameworks adapt to clarify responsibilities for developers and deployers, “the rights landscape however has not shifted in any material sense” (2B).
Here, legal reform is limited to refining statute and tort law, ensuring accountability remains squarely with human actors.
Scenario 2: Limited Legal Personality in Specialized Areas
Scenario 2 sees more complex AI deployments and the need to handle full automation of high-stakes tasks. As one attendee observed, “no one would be willing to take legal responsibility if anything goes wrong” (1A), prompting “legal innovations” that allow liability to be circumvented. AI agents remain without formal rights, but they gain recognition as “key economic, cultural, and social contributors.”
In certain sectors, “new forms of legal personality are recognised”, comparable to trusts or limited companies. While accountability ultimately “flows back to the deployer or developer” (2B), these arrangements indicate the first steps toward granting AI entities distinct legal identities for practical ends.
Scenario 3: Moral Debates and Emerging Protections
By Scenario 3, moral and ethical debates intensify. “Should AI systems this advanced be afforded rights?” asked one attendee, reflecting a growing sentiment that agents may warrant moral consideration (1D).
Another attendee noted calls for recognizing agents “as entities with moral status,” with some individuals deeming it wrong to “disconnect” an agent without considering its “preferences” (2B). Still, others cautioned that meaningful human-like rights remain “very unlikely by 2030” (2C), and changes will proceed slowly. Instead, AI may be granted corporate-like legal standing to own resources or enter contracts.
Identity management frameworks would also become critical, guarding against impersonation and manipulation in environments where AI and human roles often blur. In this scenario, legal personhood for AI is not fully realized, yet the groundwork is laid for future developments—ranging from narrower obligations and protections to potential rights expansions as societal norms and ethical consensus evolve.
Human Responses to AI-Driven Economies
In this section, attendees examined how humans may respond to widespread AI adoption across three scenarios, from initial economic disruption to fundamental social transformation. Their analysis reveals emerging patterns in economic restructuring, labor displacement, social stratification, and the search for meaning in an AI-transformed world.
Scenario 1: AI as Enhancement Tools
In Scenario 1, societal changes begin manifesting through economic restructuring. "The international ties between countries grew weaker this year... as rich countries begin to eschew global supply chains built around the cost advantages of cheap labor," while "poor countries have returned the favor, erecting trade barriers to prevent inflows of certain goods that are AI-produced" (4C). This suggests early signs of economic protectionism and changing global dynamics.
Scenario 2: Emerging Social Tensions
By Scenario 2, society grapples with significant ethical and economic debates. Key tensions emerge around "income replacement vs. unrestricted cash grants for cognitive workers" and "economic protectionism vs global distribution debates accelerate" (1D). There's particular concern about how these changes might "fan flames of anti-immigrant/nativist sentiment, in absence of clear gains from more workers in society".
Scenario 3: Fundamental Social Restructuring
In Scenario 3, society undergoes dramatic transformation in how people live, work, and find meaning. Multiple attendees highlighted several key developments:
Common Sub-Themes
Attendees identified several critical developments in social transformation by 2030:
This analysis from the attendees suggests that by 2030, society will face fundamental challenges in adapting to a post-work environment while maintaining social cohesion and individual well-being.