Artificial intelligence: incrementum ex machina

Technology has always been tied with growth. Since Adam Smith, economists have been interested in how innovations raise living standards. The rollout of textile machinery in 19^th century Britain helped kickstart the industrial revolution, with railways, electricity, automobiles and computers ushering in the modern age.

Amid developments in generative AI, speculation about its effects on growth is rife among economists and technologists alike.

Evangelists believe AI will have a greater impact than previous technologies, seeing the emergence of artificial general intelligence (AGI) assisting scientific innovation, alongside product design, and easing supply impediments to growth (discussed in a UK context here).

If generalised AI is sufficiently powerful, it could be recursively applied to boost its own intelligence, potentially expanding into domains previously unexplored by humans. A future of limitless growth, optimists believe, would surely await.

Triumph of experience over hope

Economists are more sceptical. Even if we get generalised AI (a very big if), changes in relative prices (Baumol effects) mean the economy will gravitate towards tasks that cannot be done by machines. This process is similar to how the economy pivoted towards manufacturing after the agricultural revolution, and to how services became pre-eminent in advanced economies as manufacturing productivity increased.

While AI might encourage a growth burst for some time and may even persist longer than previous waves (owing to recursive improvement), growth will eventually ebb.

AI also has to push against structural factors slowing productivity growth. An ageing population, the passing of ‘peak education’, climate change and animosity in global trade all crimp growth. Agency is also important to consider. Humans may not blindly follow AI guidance on reforms that raise growth, either on safety grounds or owing to rent-seeking behaviours from affected firms and workers.

The winner’s curse (and blessing)

AI will also affect countries at different speeds. Owing to the technology’s impact on skilled work, the need for large reserves of data and greater capital availability, we believe developed markets will likely see more immediate impacts. The IMF agrees. In their survey of the economic impacts of AI, in advanced economies (AEs) just under 60% of employment was deemed to have high AI exposure. Even adjusting for complementarity (the extent to which AI and humans can co-exist within a role) exposed jobs comprise over 20%.

For these economies, AI is both an opportunity and a curse. If AI is freely implemented, these countries will see the greatest short-term bounce in productivity. However, the costs associated with either delays in implementation or capture by affected firms in time is also greater.

AI might also generate fiscal headaches. If automation of extensive areas of work is possible, transfers to the unemployed may dwarf revenue accrued from profitable companies with smaller (albeit more productive) workforces. While calls to adjust the tax system are present in every technology rollout (anyone remember the email tax?) if AI produces substantial dislocation, tax policy will need more radical reform.

Back to the present

In recent months, speculation about AI has powered a large part of equity gains. Some have made unflattering comparisons with the dotcom bubble of the late-1990s, suggesting we are seeing something similar in equity prices today. Evidence on this score is mixed. It is certainly true valuations are increasingly stretched among recent winners, as much AI optimism is priced into major stocks. However, we have not yet seen a widening of the bubble in unprofitable companies. We don’t see anything on the scale of Pets.com or Egghead Software in the current boom.

AI adds a further dimension to relations between the west and China, with competition over the technology likened to the nuclear race in the 1940s. While China’s manufacturing-oriented economy is less exposed to AI than the west’s service model, anxiety around who has the edge in any AI race will spur continued government interest in the sector.

If productivity effects can be sustained, rates are likely to remain elevated, as the balance of projects to savings moves against its historical trend. We remain long AI-exposed stocks, and as the technology develops we will be watchful for breakthroughs that can broaden the rally into non-technology sectors.

Current AI may not produce the singularity, but its development makes for an interesting time for investors.

This concludes a series of blog posts covering improvements in generative AI; previous instalments examined the technical advances in AI and reviewed its effects on productivity and unemployment.