Why uncertainty about artificial intelligence matters for the economy

By Juan Reyes

Artificial intelligence (AI) is rapidly reshaping the world today as well as future expectations about production functions, productivity, labour markets and long-run economic growth. However, there is no clear consensus on the scale, timing or distribution of its economic effects. Some view AI as a powerful engine for efficiency and innovation, with the potential to boost output and living standards. Others worry about job losses, declining wages and disruptive changes to how work is organised as well as important privacy issues including new channels for fraud.

This disagreement reflects more than academic debate. It points to genuine uncertainty about how AI will transform the economy and society. Even before AI is widely adopted across the economy, uncertainty about its future effects can influence economic behaviour today. When the future is unclear, households delay major purchases, firms postpone investments and hiring slows. Economists have long shown that this kind of “wait‑and‑see” behaviour can slow economic activity during uncertain times when decisions are costly to reverse.¹ 

Recent AI developments may be generating a new form of uncertainty: one that is not about interest rates, inflation or government policy but about how fast technology will advance, which jobs will be affected, and how workers and firms will adapt. If so, AI could influence the economy today both through uncertainty about future productivity gains, and about who will benefit, who will lose and how quickly change will arrive.

This raises a natural question: does uncertainty about AI affect the economy in the same way as more familiar forms of economic uncertainty, or does it operate differently? 

Constructing the AI Uncertainty (AIU) Index to measure AI uncertainty

My latest ESCoE Discussion Paper attempts to address this question by examining how the US economy responds when uncertainty about AI suddenly increases.

The focus is on uncertainty about the economic consequences of AI as perceived by households, firms and policymakers. This includes uncertainty about how AI may affect productivity and wage dynamics, job reallocation and skills, workplace organisation and how governments may regulate AI technologies or their use.

To capture this dimension of uncertainty, I have constructed an AI Uncertainty (AIU) Index using news articles from leading outlets in the US, UK and selected European countries. Following the approach of Baker et al.² in constructing the Economic Policy Uncertainty (EPU) Index, the AIU index identifies articles that jointly reference “artificial intelligence”, the “economy” and “uncertainty”, along with related terms. By focusing explicitly on economic framing, the index captures discussion of uncertainty about the economic implications of AI rather than innovation or productivity alone.

Figure 1 shows that the index rises around major AI-related developments. Notable spikes coincide with events such as the release of GPT-4 in March 2023, the introduction of new AI governance measures in late 2023, and renewed debate about AI costs and global competition in early 2025. These patterns suggest that the index captures meaningful shifts in perceived economic uncertainty related to AI.

Figure 1: AIU Index (3-Month moving average)

Note: Figure 1 plots the AIU Index from M1:2016 to M4:2025. For presentation purposes, a three-month moving average is applied. Spikes in the index were investigated by manually reviewing the underlying news articles to identify the events driving the largest movements.

Is AI uncertainty different to broader economic uncertainty?

A natural question is whether the AIU Index captures a genuinely distinct source of uncertainty or simply mirrors broader economic or financial uncertainty. To test this, I compare the AIU index with existing measures of economic uncertainty, including the Economic Policy Uncertainty (EPU) Index, the Real Economic Uncertainty (REU) Index³, the S&P 500 Volatility Index (VIX), and the NASDAQ 100 Volatility Index (VXN).

The results show that AI uncertainty behaves quite differently. Statistically, the AIU Index has very little correlation with traditional measures of uncertainty. It represents a statistically independent dimension of measured uncertainty, distinct from policy-related, macroeconomic or financial market uncertainty.

Figure 2 illustrates this divergence clearly. During the COVID-19 pandemic, conventional uncertainty measures increased sharply amid widespread macroeconomic and financial stress, whereas the AIU index remained relatively stable. By contrast, from late 2022 onward, following the release of ChatGPT in November 2022, the AIU index increased substantially while conventional measures remained comparatively subdued or declined.

Figure 2: AIU Index and selected uncertainty measures

Note: Figure 2 plots the AIU Index together with the REU Index (Jurado et al., 2015), the EPU Index (Baker et al., 2016), the VIX, and the VXN. All series are normalised to facilitate comparison.

How does AI uncertainty affect the economy?

Overall, the results suggest that higher AI uncertainty has a negative effect on the economy, with the strongest impacts seen in financial markets and in how intensively people work. However, the answer is not straightforward, because news about AI often mixes two elements: optimism about future productivity and concern about disruption and adjustment. 

This identification challenge is well recognised in the literature. Piffer and Podstawski⁴ and Cascaldi-Garcia and Galvão⁵ show that uncertainty shocks are frequently correlated with technology news shocks that revise expectations about future total factor productivity. When these two effects move together, it becomes difficult to tell whether the economy is responding to uncertainty itself or to revised expectations about future productivity. 

To address this issue, I remove movements in AI‑related coverage that reflect general optimism or pessimism about economic growth. The remaining series, shown in Figure 3, captures changes that are specific to uncertainty about AI itself. This measure is then used to identify how unexpected increases in AI uncertainty affect the economy.

Figure 3: Pure AI uncertainty component

Note: Figure 3 displays the residual-based measure of AI uncertainty, constructed by isolating the uncertainty component of the AI-related coverage. The procedure involves regressing the AIU Index on a broader measure of AI-economy news coverage. These capture fluctuations in uncertainty that are orthogonal to overall changes in AI-related economic discussion. The annotated points correspond to selected AI-related developments that received extensive news coverage. The sample spans M1:2016 to M4:2025.

Focusing on the US economy, Figure 4 shows how key economic indicators respond when there is a sudden rise in uncertainty about AI. Specifically, it looks at stock prices (the S&P 500), hours worked, wages, employment, and industrial production following a typical increase in AI-related uncertainty. 

Figure 4: Impulse responses to AI uncertainty shock

Note: Figure 4 displays the impulse responses to a one standard deviation AI uncertainty shock (red line) identified using SVAR-IV. The residual-based instrument isolates distinct components of AI-related coverage, separating discussions that frame AI as a source of uncertainty from broader narratives linked to productivity and innovation. The identified shock, therefore, captures adverse assessments related to labour displacement, regulatory challenges, and sectoral disruption, among others. Shaded areas denote 68% confidence bands based on 1,000 wild bootstrap replications.

Equity market

I find that the equity market responds immediately and persistently. S&P 500 fall sharply (approximately 1.10%) as soon as AI uncertainty rises and stays below its previous level for an extended period, suggesting lasting downward pressure on equity prices.

Labour market

The analysis provides evidence that labour market adjustment occurs primarily through hours worked rather than employment. Hours worked decline initially (around 0.15%) but recover within around ten months, indicating a temporary effect. Wages, by contrast, show a more persistent decline. They fall initially and continue to decrease over time (nearly 0.20%), with no sign of recovery during the period studied. Employment levels remain largely stable throughout.

Overall economic output

The results also indicate that the impact on overall economic output is smaller. Industrial production declines briefly (approximately 0.20%) following an increase in AI uncertainty but then gradually returns to its previous level. This drop is both less severe and less long-lasting than the effects seen in stock prices and wages.

How AI uncertainty looks different

When uncertainty comes from more familiar sources such as recessions, financial stress, or policy uncertainty, it tends to cause widespread but mostly temporary slowdowns. The literature shows that employment, hours worked, and output decline together, equity prices fall sharply, and these effects gradually fade over time. (Bloom, 2009; Jurado et al., 2015; Baker et al., 2016).

Uncertainty linked specifically to AI follows a different pattern. Employment levels remain broadly stable, while hours worked and wages decline persistently. This means that firms adjust mainly by reducing working time and pay, rather than by laying off workers. Output falls only slightly, while equity prices show a larger and longer-lasting decline.

At first glance, this behaviour may resemble what happens during normal downturns when workers are kept on because firing and rehiring staff is costly.⁶ The responses to AI uncertainty show a similar split between stable employment and falling hours. However, there is an important difference: wages also decline and do not recover quickly, whereas in standard downturns, wages tend to be more stable. This may be because AI is perceived as a possible alternative to labour, which weakens worker bargaining power. Even if automation is not immediately adopted, the possibility that it could be adopted means workers have less power to push for higher wages.⁷

Overall, these responses point to a distinct adjustment pattern. Uncertainty about AI mainly affects how intensively people work and how much they are paid, rather than whether they remain employed at all. This distinguishes AI uncertainty from more familiar sources of economic uncertainty and suggests that uncertainty related to technological change may operate through different labour market margins than uncertainty associated with the business cycle or government policy.

Why does this matter?

AI-related uncertainty is economically meaningful. As AI continues to evolve, the uncertainty it generates becomes increasingly important to monitor. The results show that uncertainty around AI can affect the wider economy well before productivity gains are seen, changing how the economy adjusts overall.

Understanding how uncertainty linked to technological change differs from more familiar sources of economic uncertainty can help policymakers, firms and workers better anticipate and respond to the economic challenges of rapid innovation.

Summary

This blog introduces a new AI Uncertainty (AIU) Index to track uncertainty about the economic effects of artificial intelligence. It finds that AI uncertainty is distinct from traditional economic uncertainty and is associated with falling equity prices and wages, and temporary declines in hours worked, even when employment and output remain stable. The results show how uncertainty around AI can influence the economy before productivity gains materialise.

ESCoE blogs are published to further debate.  Any views expressed are solely those of the author(s) and so cannot be taken to represent those of ESCoE, its partner institutions, the Office for National Statistics, or the UK Endorsement Board.

References

¹  Bloom, N. (2009). The Impact of Uncertainty Shocks. Econometrica, 77(3), 623–685. https://onlinelibrary.wiley.com/doi/abs/10.3982/ECTA6248

² Baker, S. R., Bloom, N., and Davis, S. J. (2016). Measuring Economic Policy Uncertainty. Quarterly Journal of Economics, 131(4), 1593–1636. https://academic.oup.com/qje/article/131/4/1593/2468873

³ Jurado, K., Ludvigson, S. C., and Ng, S. (2015). Measuring Uncertainty. American Economic Review, 105(3), 1177–1216. https://www.aeaweb.org/articles?id=10.1257/aer.20131193

⁴ Piffer, M. and Podstawski, M. (2018). Identifying Uncertainty Shocks Using the Price of Gold. The Economic Journal, 128(616), 3266–3284. https://academic.oup.com/ej/article/128/616/3266/5251693

⁵ Cascaldi-Garcia, D. and Galvao, A. B. (2021). News and Uncertainty Shocks. Journal of Money, Credit and Banking, 53(4), 779–811. https://ideas.repec.org/a/wly/jmoncb/v53y2021i4p779-811.html

⁶ Oi, W. Y. (1962). Labor as a quasi-fixed factor. Journal of Political Economy, 70, 538–555. https://econpapers.repec.org/article/ucpjpolec/v_3a70_3ay_3a1962_3ap_3a538.htm

⁷ Leduc, S. and Liu, Z. (2024). Automation, bargaining power, and labor market fluctuations. American Economic Journal: Macroeconomics, 16, 311–349. https://www.aeaweb.org/articles?id=10.1257/mac.20220181