Unlock Editor's Digest without cost
FT editor Roula Khalaf selects her favourite stories on this weekly newsletter.
The National Bureau of Economic Research has published a brand new Paper by Daron Acemoglu, the superstar MIT economist who tries to dismiss AI dreams corresponding to a productivity renaissance, accelerated growth, and reduced inequality.
At this point, it almost looks like heresy to say that AI is not going to revolutionize every part. A yr ago Economists from Goldman Sachs estimates that AI would increase annual global GDP by 7 percent inside 10 years – or nearly $7 trillion in dollar terms.
Since then, Goldman’s forecast has turn out to be almost sober. Even the IMF predicts that AI “has the potential to remodel the worldwide economy“FTAV’s personal favorite is ARK’s prediction that AI will help speed up global GDP growth to 7 percent per yr. 🕺
Professor Acemoglu – a possible future Nobel Prize winner – is take the opposite side. Alphaville's emphasis below:
I estimate that the impact (on total factor productivity) of AI advances over the subsequent decade will likely be modest – an upper sure that doesn’t have in mind the excellence between difficult and straightforward tasks can be a complete increase of about 0.66% over ten years, or a rise in annual TFP growth of about 0.064%. If one takes into consideration that there will likely be difficult tasks among the many tasks that will likely be exposed to AI, this upper sure drops to about 0.53%. The impact on GDP will likely be somewhat larger because automation and task complementarity can even result in higher investment. But my calculations suggest that GDP growth over the subsequent 10 years must also be modest, within the range of 0.93% − 1.16% over 10 years.assuming that the rise in investment on account of AI is moderate and, within the event of a serious investment boom, will likely be within the order of 1.4-1.56% overall.
As Acemoglu says, that is “modest but still removed from trivial.” But as he notes, we also need to contemplate the indisputable fact that a few of the commonest AI use cases are bad – e.g., deepfakes, etc.
Tackling these problems may stimulate growth in the identical way that rebuilding a city devastated by a hurricane stimulates growth, nevertheless it still comes on the expense of overall prosperity. Alphaville's emphasis below.
. . . If we have in mind the likelihood that latest AI-generated tasks may very well be manipulative, the impact on welfare could also be even smaller. Based on figures from Bursztyn et al. (2023) referring to the negative impact of AI-powered social media, I provide an illustrative calculation for spending on social media, digital ads, and IT defense attacks. These could increase GDP by as much as 2%, but when we apply the figures from Bursztyn et al. (2023), their impact on welfare may very well be -0.72%. This discussion suggests that it is vital to contemplate the potential negative welfare impacts of the brand new tasks and products generated by AI.
Acemoglu can be skeptical that AI can have a serious impact on inequality—neither a big worsening nor an improvement. But overall, his work suggests that “women with low levels of education might even see slight declines in wages, overall inequality between groups may increase barely, and the gap between capital and labor income is more likely to proceed to grow.”
The skepticism is interesting because Acemoglu is one third of an influential trio of MIT economists who’ve developed the clumsily named Shaping the longer term of labor Initiative.
The professor emphasizes that the potential of generative AI is great, but only whether it is used primarily to offer individuals with higher and more reliable information, fairly than Chatbots susceptible to hallucinations and mechanically restored images.
In my view, there are literally much greater advantages to be gained from generative AI, a promising technology. However, these advantages is not going to materialize with no fundamental re-orientation of the industry. This may include a fundamental change within the architecture of essentially the most common generative AI models, corresponding to LLMs, to concentrate on reliable information that may increase the marginal productivity of several types of employees, fairly than prioritizing the event of general, human-like conversational tools. The overall purpose of the present generative AI approach could also be ill-suited to providing such reliable information.
To put it simply, it stays an open query whether we’d like foundational models (or the present breed of LLMs) that may hold human-like conversations and write Shakespearean sonnets if we would like reliable information useful to educators, health professionals, electricians, plumbers, and other tradespeople.
Further reading:
— The Manicure Industry (FTAV)
— Investment outlook for the approaching yr or ChatGPT? Take the quiz (FTAV)
— Generative AI will likely be great for generative AI consultants (FTAV)
