AI and labour

4 minute read

Published:

The rapid advancement of AI technologies is transforming industries and labor markets at an unprecedented pace. Despite these sweeping changes, the relationship between AI and labor remains surprisingly understudied. Recent works, notably by Korinek & Suh (2024), Acemoglu (2025), and Epoch AI's GATE model (2025), illustrate the complexity of AI’s economic impacts, but also highlight significant gaps in understanding AI’s real-world implications for labor.

Korinek and Suh’s theoretical exploration into AGI’s potential trajectories offers insights into the dramatic possibilities ranging from wage collapse to explosive growth. Their model extensively analyzes scenarios based on computational complexity and automation thresholds, but it heavily emphasizes computational power as a primary determinant of automation, neglecting critical real-world frictions. Factors such as regulatory constraints, societal acceptance, data privacy concerns, and task-specific challenges like dexterity and specialized hardware limitations significantly shape the pace and extent of automation but are inadequately addressed in their framework. Furthermore, their approach treats labor predominantly as homogeneous, overlooking the complex interplay of skills, job specialization, and human-centric tasks. This oversimplification fails to capture the nuanced realities of hybrid automation scenarios, where automation is partial, and human oversight remains crucial for task execution.

On the other hand, Acemoglu’s empirical, task-based model forecasts modest productivity and wage impacts from AI over the next decade, grounded in detailed estimates of task exposure and cost-saving potentials. While somewhat empirically grounded, this approach underestimates the transformative potential of AI to generate entirely new tasks and industries, fundamentally altering labor market dynamics. Additionally, Acemoglu’s analysis relies heavily on extrapolations from current AI capabilities, which may not adequately represent future capabilities given the rapid and often unpredictable pace of AI advancements. By narrowly focusing on incremental, task-specific productivity improvements, Acemoglu’s analysis misses significant recombinative innovations, new industries and job categories created by novel applications of AI technologies, that could substantially reshape labor dynamics and employment opportunities.

Epoch AI’s GATE model provides an integrated, compute-based assessment of AI-driven economic transformation. GATE uniquely combines a compute-based AI development model, automation mapping, and macroeconomic modeling with endogenous investment and adjustment costs. This integration highlights critical interactions between AI advancements, economic growth, and labor dynamics, capturing real-world frictions such as hardware production bottlenecks, capital adjustment costs, and uncertainties around AI capabilities. However, while comprehensive, the model still simplifies the complexities of task heterogeneity and the dynamic nature of labor markets, particularly in reallocation challenges and social adaptability to rapid technological changes.

None of these papers thoroughly addresses the differentiated impacts of automation across varying skill levels, nor the social and economic implications of widespread re-skilling or unemployment. Moreover, institutional contexts such as labor laws, union power, cultural attitudes toward automation, and public policies aimed at mitigating adverse effects remain significantly understudied. The extent to which these studies differ in their predictions with respect to each other is concerning. It points to the fact that no one has a clear or common understanding of what this technology’s impact on the labour market will be, and significantly more effort needs to be put into developing more detailed scenarios, assuming alternate models, comparing results, running ablations, and condensing insights.

Addressing these gaps requires a concerted research effort. Some potential areas to focus on are:

  1. Granular Skill Heterogeneity and Adaptation: Future research should precisely map AI impacts across diverse skill sets, identifying tasks and roles most susceptible to displacement or complementarity. Examining the effectiveness, cost, and feasibility of re-skilling programs is crucial to prepare the workforce adequately.

  2. Institutional and Social Constraints: Investigation into regulatory barriers, labor market protections, social acceptance, trust, and ethical implications of automation is necessary. Understanding how institutional environments can shape the deployment and adoption of AI will help policymakers craft effective, socially responsible strategies.

  3. Hybrid Automation Models: Exploring “human-in-the-loop” models, where automation supplements rather than replaces human labor, could provide realistic insights into evolving job markets. Such research should address the implications for job design, workplace structure, wage distribution, and workforce satisfaction.

  4. Economic and Welfare Valuation of New Tasks: Developing methods to quantify the societal and economic impacts of new AI-enabled tasks, including those that negatively affect social welfare, is essential. Such valuation frameworks should inform policy decisions and mitigate potential adverse consequences.

We must balance empirical detail and theoretical depth, connecting short-term realistic scenarios with long-term, potentially radical transformations. By rigorously studying these understudied aspects of the AI-labor relationship, we can better anticipate and manage the profound implications of AI for society and employment.

You May Also Enjoy

On Knowledge and Substrate

6 minute read

Published:

I’ve recently been thinking a lot about what the intrinsic space of all human knowledge looks like, what kind of topology and structure does the neural latent manifold have, how sparse is it, and how to think about all the space in between pockets of density. For instance, it is not clear to me what the dimensionality of the original space is and whether using tokens as the basic entities of this space even makes sense. Maybe tokens are too granular to be useful for this kind of a thought experiment and we need to think about this at a higher level, say sentences and concepts. The reason such a thought experiment is appealing to me is because I think it lies at the heart of a question I’m interested in - whether AI can discover truly new knowledge.

Enhancing Factual Accuracy in Large Language Models: Integrating Decoding Strategies and Model Steering

19 minute read

Published:

The emergence of open-source Large Language Models (LLMs) like Llama has revolutionized natural language generation (NLG), making advanced conversational AI accessible to a broader audience [1]. Despite their impressive capabilities, these models often grapple with a significant challenge: factual hallucinations. Factual hallucinations occur when an AI model generates content that is unfaithful to the source material or cannot be verified against reliable data [2]. This issue is particularly concerning in critical and information-dense fields such as health, law, finance, and education, where misinformation can have catastrophic consequences [3][4].

Perspectives on the Future of AI

13 minute read

Published:

How big are the models going to get and how much longer is the scaling hypothesis going to hold? It’s unclear, but according to current performance trends, which haven’t shown signs of plateauing (GPT-4o, Claude 3.5 Sonnet, Gemini-1.5-Pro, Llama-3.1-405B, Grok-2), and the power budget of announced data centres (5GW OpenAI/Microsoft Stargate campus), it is likely that there is an order of magnitude left (OOM) to climb in model size. This Epoch AI research covers these scenarios in depth and estimates training runs of the order of ~2e29 FLOPs being possible by 2030, which would be 4 OOMs larger than GPT-4 (2e25 FLOPs). These training runs will primarily be power constrained, followed by chips, data, and latency.

Climate Risks for India in the Coming Decades and the Need to Invest in Adaptation Projects

27 minute read

Published:

Climate change has emerged as one of the most pressing challenges of the 21st century, posing unprecedented risks to economies, ecosystems, and human well-being. India, with its diverse geography and significant dependence on climate-sensitive sectors like agriculture, faces heightened vulnerability. Rising temperatures, extreme heat events, changing precipitation patterns, droughts, floods, and coastal hazards are increasingly evident, threatening rural livelihoods and urban infrastructure alike. Although India has been proactive in formulating climate policies—such as the National Action Plan on Climate Change (NAPCC) and State Action Plans on Climate Change (SAPCCs)—and has undertaken mitigation initiatives, the intensifying impacts demand a sharper focus on adaptation. This article reviews India’s key climate risks, summarizes existing adaptation strategies, and discusses the urgent need for scaling up investments in resilience-building measures. It concludes by proposing a strategic path forward to mainstream and finance climate adaptation across sectors.