How could AI impact the global economy? | Michael Spence, Mohamed El-Erian, & James Manyika

AI's Potential for the Economy 🔌

I have been thinking about artificial intelligence in a very specific way: as a technology with the potential to act like a new form of electricity. That comparison is not a throwaway line. It captures several critical features: it is a general purpose technology, it can be applied across almost every sector, it requires significant infrastructure to realize its benefits, and it changes what is possible in ways we do not fully imagine today.

Mohamed El-Erian put it succinctly when he said it is “what has been called the new electricity.” He went on to highlight three interconnected aspects that matter for the economy. First, AI enables unmet needs to be addressed. Second, it increases the efficiency of tasks we already perform. Third, it promises entirely new capabilities that we cannot yet foresee. That is a sweep of potential that touches productivity, inclusion, and the capacity to solve long-standing structural problems.

Mohamed El-Erian: "I think of it as a massive productivity enhancement. And what I find most exciting at this moment of time is if you froze where we are today, the productivity promise, if we get the other bits right, is significant."

To make this electricity metaphor operational, I like to break it into three layers: the foundational compute and data infrastructure, the platforms and tools that make models usable by businesses and individuals, and the applications that directly change outcomes in education, health, manufacturing, finance, and public services. Each layer has its own investment needs and policy challenges.

Concretely, the productivity gains look like faster decision making, automation of repetitive tasks, better forecasting, smarter supply chains, and augmentative intelligence that improves worker output. Even where AI does not replace work, it can boost the effective productivity of people and teams. When that productivity propagates across the economy it can shift macroeconomic indicators — growth, inflation, and long-term interest rates.

That propagation is the critical variable. If a productivity surge remains in pockets, its macro impact will be limited. If it spreads, it changes trajectories. Mohamed raised precisely that point: if AI can drive a broad productivity surge, it can help address issues such as aging populations and fragmented global supply chains, and it can, theoretically, lower inflation and make debt more sustainable. Those are not marginal effects. They are potentially structural.

AI's Impact on Work and Skills 🧰

The economic story cannot be written without a chapter on work and skills. Michael Spence underlined a central economic truth: the ultimate impact on employment depends on demand elasticities. In plain language, if productivity rises, costs fall, prices fall, and demand expands sufficiently, employment may shift rather than shrink. If, however, the productivity shock is large and fast and demand does not pick up quickly enough, the labor market can be disrupted in ways that cause friction and real hardship.

Michael Spence: "What it has ultimately to do with work depends on in the long run on elasticities of demand. If you have enough competition and you have a huge productivity increase, it should lower the cost. And eventually, that should show up in an expanded kind of market situation."

I agree with Michael that we should expect a mixed picture. There will be occupations that decline over time, occupations that grow, and many that change in composition and required skills. My view is that adaptation is possible but dependent on policy, institutions, and the speed of change. If transitions are smooth and people can reskill or move to growing sectors relatively quickly, the social cost will be manageable. If transitions are abrupt and support is lacking, you can get persistent unemployment, underemployment, and political backlash.

There are a few distinct mechanisms by which AI affects jobs:

Substitution: AI automates tasks previously done by humans, especially routine cognitive or manual tasks.
Augmentation: AI amplifies human capabilities, making people more productive in the same role.
Complementarity: New jobs and tasks emerge that require working with AI systems, creating demand for new skill sets.

For workers, the relevant policy levers are active labor market policies, continuous learning systems, stronger employer engagement in training, portable benefits, and income support during transitions. For firms, it is about redesigning work, investing in retraining, and measuring outcomes rather than headcount reductions as the sole metric of efficiency.

We also need to attend to wages. Michael commented that if people are transitioning to other occupations, wages should continue to rise as productivity increases. That is not automatic. Productivity growth has not reliably translated to median wage gains over recent decades in many economies. Ensuring wages reflect productivity gains requires institutional attention: bargaining arrangements, minimum wage policy, and perhaps incentive structures that align productivity gains with worker compensation.

Lessons from Past Disruptions 🕰️

History offers crucial lessons. Mohamed warned against focusing solely on economic metrics while ignoring social and political dynamics. Innovations change market structures and behavior in unexpected ways. The securitization example is instructive. It was a powerful financial innovation that enabled risk allocation, but when it was overconsumed and poorly guarded by governance, it contributed to systemic fragilities that culminated in the global financial crisis.

Mohamed El-Erian: "Don't focus only on economics. Look at the social and the political, right? Because when you solve it only in terms of economics, it seems simple."

That lesson translates to AI. Lowered barriers to entry can create an environment of overproduction and overconsumption of tools, data, or services. It can encourage speculative investment in applications that do not have public value or that crowd out needed investments in infrastructure and governance. It also risks exacerbating concentration if a few firms capture most of the economic surplus from AI through data, model ownership, or distribution advantages.

Past technological disruptions — from mechanization in manufacturing to the digital revolution — have shown that while aggregate growth may be positive, distributional effects matter and can produce political backlashes. The social compact needs to be considered as part of the technological rollout. Policies that only optimize GDP growth without addressing displacement, local shocks, and community breakdown will be resisted and could undermine the long-term adoption of AI.

We must also pay attention to systemic risk. Rapid, broad adoption without adequate governance might create new vulnerabilities in finance, national security, and public health. Oversight, standards, and regulatory frameworks need to be in place before we reach a point where the system runs ahead of what can be sustained.

Macroeconomic Optimism for Growth and Debt 📈

One of the most striking takeaways from the discussion was the macroeconomic optimism that a widespread productivity improvement driven by AI could generate. Mohamed said something I will not forget: his generation failed to leave high inclusive growth and manageable debt for the next generation, but AI might be the tool that helps turn things around. That is a bold claim, but it is grounded in plausible mechanics.

Mohamed El-Erian: "The good news is we're leaving you with a tool that can help solve this. So when you think of what a huge productivity shock can do, it can literally step up growth, not only once and for all, but the whole trajectory can change."

Michael reinforced this perspective by pointing to a sober institution: the Congressional Budget Office. The CBO estimated that, if AI effects spread through the whole economy, real growth might reach the 4 percent range. That is a striking figure compared to recent decades of 1.5 to 2 percent growth rates in many advanced economies. Consider what that implies: faster growth raises tax revenues, reduces debt ratios, and improves fiscal sustainability without necessarily increasing tax rates.

Michael Spence: "The Congressional Budget Office basically said the productivity potential could produce real growth at the 4 percent range, provided it goes through the whole economy."

That scenario also has knock-on effects for inflation and interest rates. A broad productivity surge would increase supply-side capacity. If accompanied by stable demand, it could lower inflationary pressures and reduce long-term real interest rates, which would make it cheaper to finance investments in energy transition, digital infrastructure, and public goods. The combination of private incentives and public investments could create virtuous cycles.

I am careful to qualify this optimism. It is conditional. The macro dividend depends on the extent to which productivity gains are realized, how quickly they diffuse, and whether policy and institutions enable broad-based adoption rather than concentrated capture. But the sheer scale of the upside is substantial enough that it changes how we think about the policy toolbox for growth and debt management.

Given the scale of opportunity and risk, I believe we need a multi-pronged policy response. I am going to outline a pragmatic, actionable set of priorities that balance growth, inclusion, and stability. These are not theoretical wish lists. They are the levers I would focus on as a policymaker or advisor charged with maximizing social welfare in the face of rapid technological change.

1. Invest in digital infrastructure and public data

AI requires compute, connectivity, and high-quality data. Public investment in data infrastructure, privacy-preserving shared datasets, and cloud capacity can lower entry barriers for small and medium enterprises and public sector innovation. Shared, governable data assets for health, climate, and education can accelerate public value creation.

2. Strengthen lifelong learning and upskilling systems

Countries should reorient education systems around lifelong learning. That means portable credits, micro-credentials, employer-government training partnerships, subsidies for retraining, and incentives for firms to invest in worker skills. The aim is to speed up transitions for workers moving between occupations.

We need safety nets that are timely and adaptive. That includes unemployment insurance that is more responsive, wage insurance pilots, mobility allowances, and programs that reduce the friction and financial pain during retraining. These are insurance mechanisms that make transitions bearable and politically sustainable.

4. Update competition policy and market governance

Concentration risks are real. Competition authorities should monitor data-driven market behaviors, gatekeeper platforms, and vertical integration. Remedies could include interoperability mandates, data portability, and careful review of mergers that would entrench market power.

5. Ensure robust AI governance and standards

Standards for safety, transparency, and accountability are essential. Governance frameworks should include risk-based oversight, third-party audits, and sector-specific rules for high-stakes applications such as healthcare and finance. Governance needs to be international where possible because data and models cross borders.

6. Support public-sector adoption and innovation

Public services can be early beneficiaries of AI if implemented responsibly. Governments should pilot AI in education, public health, and administration while investing in evaluation and safeguards to ensure public trust and effectiveness.

7. Prioritize R&D and public-private partnerships

Public funding should support foundational research, open science, and mission-oriented investments for climate, health, and basic sciences. Partnerships between academia, industry, and government can accelerate breakthroughs and ensure they translate to public goods.

These priorities are complementary. Infrastructure without skills will create inequality. Skills without demand will produce underemployment. Governance without innovation will stifle growth. A balanced approach is essential.

What Businesses and Workers Should Do Now 💼

The technology is moving fast, and firms and workers cannot wait for perfect policy answers. There are practical steps to take today.

For businesses

Adopt thoughtfully: Start with pilot projects that focus on measurable outcomes, not hype. Use AI to augment workflows and measure productivity changes.
Invest in worker transitions: Offer retraining and reskilling programs as part of digital adoption plans. This reduces turnover and builds capabilities aligned with future needs.
Measure productivity differently: Look beyond headcount. Track value created, error reduction, time-to-market improvements, and customer outcomes.
Share gains: Consider profit-sharing or bonus structures linked to productivity improvements so workers capture part of the upside.
Address governance: Implement internal governance for AI ethics, safety, and data stewardship.

For workers

Embrace lifelong learning: Keep skills current. Focus on skills that are complementary to AI: problem framing, judgment, domain knowledge, creativity, and interpersonal skills.
Leverage AI as a tool: Use AI to augment your work. Learn to use tools that automate routine aspects and free you for higher-value tasks.
Explore hybrid roles: Look for opportunities that combine domain expertise with AI supervision, dataset curation, and evaluation.
Seek portable credentials: Accumulate micro-credentials and documented project portfolios that can be transferred across employers.

These steps will not eliminate the need for public action. But they can reduce individual vulnerability and help organizations derive real value without unnecessary disruption.

Risks to Watch and How to Mitigate Them ⚠️

While I am optimistic about the macro potential, the risks are real and worth cataloguing. They fall into several buckets: economic concentration, systemic financial risk, social dislocation, and misuse of technology. Below I outline these risks and pragmatic mitigation approaches.

Economic concentration

Large firms with control over data, models, and distribution channels can capture disproportionate rents. This reduces competition and could slow innovation over time.

Mitigation: Strengthen antitrust scrutiny with a focus on data and platform dynamics. Promote interoperability, data portability, and open standards.

Systemic financial and macro risk

Rapid adoption of AI could produce overinvestment in speculative applications or create feedback loops in financial markets.

Mitigation: Improve macroprudential oversight, monitor asset valuation dynamics related to AI sectors, and ensure financial institutions’ risk models adapt to new technology-driven correlations.

If transitions are poorly managed, communities and regions that lose employment can suffer long-term decline. That is a political risk that can slow or reverse beneficial reforms.

Mitigation: Implement place-based policies, invest in community economic development, and ensure income support is responsive and accessible.

Misuse and ethical harms

AI can be used for misinformation, surveillance abuses, biased decision making, and other harmful ends.

Mitigation: Enact sectoral regulations, promote transparency, require impact assessments for high-stakes systems, and fund research into fairness and accountability methods.

Overconsumption and unsustainable scaling

Mohamed warned of the danger of overconsumption. The market can overinvest in “solutions” that produce poor social returns or that strain infrastructure.

Mitigation: Encourage impact evaluation, support mission-driven investments, and create better signals about social returns through public procurement and standards.

All these mitigations require coordination across government, industry, and civil society. No single actor can manage these risks alone. The pace of change argues for pre-emptive rather than reactive measures where feasible.

Opportunities Beyond Productivity: Science, Health, Education 🔬

One of the most exciting aspects of this era is what AI can do for scientific discovery and public goods. Michael mentioned drug discovery and material science as areas where AI can accelerate breakthroughs. I see similar potential in education and finance.

Drug discovery is a useful example. Traditional drug development is expensive and slow, with a high failure rate. AI can speed up target identification, design candidate molecules, and reduce the time from concept to clinical trials. That could lower costs and expand access to treatments. More broadly, AI can enhance diagnostics, enabling early detection and personalized medicine at scale.

In materials science, AI-driven simulation and optimization can lead to new materials for batteries, solar panels, and carbon capture. Those innovations are directly relevant to the energy transition and climate goals. If AI reduces the cost and time for breakthroughs, it can accelerate decarbonization by enabling more effective and cheaper technologies.

Education is another domain where AI can deliver large social returns. Personalized learning systems, intelligent tutoring, and adaptive curricula can help learners progress at their own pace. For low-income households, AI-enabled tools can provide access to high-quality learning resources that previously required human teachers. Mohamed used the term "leapfrog," and that is exactly what I'm thinking about: enabling low-income populations to bypass intermediate stages of development by using AI-assisted services to reach better outcomes faster.

In finance, AI can expand financial inclusion through better credit scoring, risk assessment, and personalized financial advice. That could open up lending and investment opportunities for previously underserved populations. But we must simultaneously guard against biased models that reproduce or amplify discrimination.

How This Could Change the Trajectory of Generations 🌍

When Mohamed reflected on generational responsibility, he framed the current moment as a chance to change the trajectory that previous policy failures left behind. I want to echo that point with a generational lens. The young today face high household debt, climate risk, and productivity stagnation that constrained living standards. AI presents a rare opportunity to reset that trajectory.

It would be naive to assume the benefits will automatically accrue to the next generation. But with the right blend of public investment, institutions, and governance, AI-driven productivity can improve job prospects, increase public revenues for social investments, and fund the green transition that is indispensable for long-term prosperity.

From a policy perspective, that implies a forward-looking orientation. Fiscal and industrial policy should be calibrated to exploit the transient window where new technologies create disproportionate returns for those who deploy them strategically. Mission-oriented investments in climate, health, and education can create durable demand for AI-driven solutions while delivering public value.

Final Takeaways and My Outlook 🔭

Here are the core conclusions I draw from conversations with Mohamed El-Erian and Michael Spence and from my own work tracking AI’s trajectory.

AI is a powerful productivity technology with the potential to change macroeconomic trajectories. If its benefits diffuse broadly, it can raise growth, reduce debt burdens, and ease inflationary pressures. That is a central reason to be optimistic.
Distributional effects matter. Gains will not land evenly. Some occupations and communities will face disruption. Policies that speed reskilling and protect incomes during transition are essential to avoid social and political backlash.
Speed matters. If the productivity shock is faster than the labor market’s ability to adjust, the short-term human costs could be significant. That argues for proactive policies, not reactive ones.
Governance and institutions are as important as technology. Standards, competition policy, data governance, and ethical oversight determine whether AI produces broad social value or concentrated rents and risks.
Public investment can amplify benefits. Strategic public spending on infrastructure, shared data, R&D, and skills can make AI more inclusive and productive.
There is also a science dividend. AI will likely accelerate discovery in health, materials, and climate, producing benefits that go beyond GDP.

My outlook is cautiously optimistic. I see a credible path where AI helps solve some of the major economic challenges of our time if we act deliberately. I also see many ways the transition could go wrong if governance lags or if social protections are inadequate. The policy window is open now. The choices we make over the next few years on education, competition, governance, and public investment will determine whether AI becomes a broad-based engine of inclusive prosperity or a driver of concentrated wealth and social stress.

In the spirit of practical urgency, I would end by encouraging leaders in government, business, and civil society to treat AI not as a distant technical curiosity but as an economic and social project. It is a project that needs infrastructure, skills, standards, and public purpose. With those pieces in place, the prospect of a generationally transformative productivity surge is real. Without them, we risk squandering this extraordinary technological moment.

Michael Spence: "So I think it's not a done deal, but from a macroeconomic point of view, it looks like close to a silver bullet."

That framing captures both opportunity and caveat. I choose to take it as an invitation to act. The potential is too large to ignore, and the risks are too consequential to neglect. My recommendation is clear: move quickly on infrastructure and skills, thoughtfully on governance, and boldly on public investments that create demand for socially valuable AI applications. If we do those things, AI can be the new electricity that powers inclusive growth for generations to come.