AI and the economy: where do we begin?

“AI is the ability of a machine to display human-like capabilities such as reasoning, learning, planning and creativity” (European Commission). Developments in recent years are moving us closer to a world in which we live alongside these “machines”, undoubtedly bringing vast changes and improvements, not to mention unexpected impacts and uncertainty. Against this backdrop, we’re kicking off a series of reports on how AI will shape our economies.

Since its birth in the 1950s, AI has evolved through advances in machine learning during the 1990s, the development of deep learning in the 2010s (which uses neural networks to work with unstructured data), and the invention of large language models (LLMs) that enable neural networks to understand the context of words. LLMs have become the leading light in generative AI, whose main feature is the ability to create texts, images, music and other content based on natural language prompts. 

More traditional machine learning models are also effective and will continue to find new uses. They help boost companies’ results by performing numerical, optimization and predictive model-related tasks that reduce errors and increase speed and quality. The novelty of generative AI is that it enables us to move away from systems that perform certain tasks through code and extensive and painstaking data collection to train neural networks to systems that can perform a wide range of tasks easily and cheaply. 

Companies´ studies are reporting a positive impact on productivity and in the anatomy of work

Company research shows that generative AI increases employees’ productivity in tasks that require cognitive abilities (i.e. the brain's capacity to receive, process and produce information). For instance: (i) Customer support. In a survey of 5,000 agents, those who use AI were able to handle 13.8% more queries per hour and increase the quality of their work; (ii) Drafting. One study asked professionals in several fields to draft two documents. For the second document, half of the participants were asked to use ChatGPT. Those who did were almost 60% faster and produced more polished writing; (iii) Programming. Software developers who use LLMs were able to code more than double the number of projects per week. 

Some studies also point to a positive impact on the bottom line and productivity in companies that invest in and use AI. AI patents and the use of AI in the companies subject to these studies link to faster growth in employment, output, revenue and market valuation, which is mainly driven by product innovation. And this will all have come about in just one to two years. 

Though research is yet to provide evidence, AI could also boost productivity by promoting innovation, which arises in jobs that require cognitive work. If we make this type of work more efficient, innovation and productivity could also increase.  

McKinsey estimates that generative AI and the more traditional machine learning models have the potential to automate tasks that currently take up 60-70% of workers' time. It also estimates that half of all tasks could be automated between 2030 and 2060, depending on the pace of adoption. 

AI will likely be the next general-purpose technology and will bring many benefits...

But despite the signs, people are divided on how far it can go.

On the less optimistic side are researchers who argue that we are not in an era of unprecedented innovation, and that past inventions such as electricity or toilets had a greater impact on living conditions than AI will. On the opposite side are those who say that AI will reach levels of intelligence with human-like cognitive capabilities, known as general AI.

Without going that far, AI can become a transformative technology of profound impact if it becomes — as it appears to be — the next general-purpose technology (GPT).  GPTs such as engines, electricity, computers and the internet eventually reach widespread and versatile use, with ripple effects that enhance other technologies, affect jobs and sectors and, ultimately, bolster productivity. 

GPTs usually impact on economies in three stages. First, for a recent innovation that is yet to be widely used, the overall productivity benefits are not so obvious. Second, as the technology evolves and gets better, its costs fall and its wider roll-out starts to nudge productivity upwards. And third, amid widespread use of the technology and a slowdown in enhancements, the law of diminishing returns starts to kick in and productivity gains begin to tail off. 

IA is currently in the first stage. As it evolves, its costs fall and companies adopt it, AI will transform tasks, organization, business models, strategies, efficiency, the basis of competition in several sectors and, over time, accelerate productivity at the macroeconomic level, which will impact on economic growth and living standards (which, in turn, will be affected by the investment needed to implement the technology). 

...but also questions and uncertainty... 

As with all technological revolutions, AI will bring uncertainty and questions, at least for some periods. 

It will have an impact on the job market. The type of jobs on offer will change. And the effects will vary between countries.

Everything will depend on the policies adopted to drive and mitigate potentially adverse effects and how AI and these policies line up with demographic shifts, climate change, the transformation of globalization, increased geopolitical tension, and other major trends that shape the world. 

We don’t know the exact potential of AI in the long term

Current LLMs are unable to perform elementary logical reasoning tasks and also suffer from “hallucinations”. Their limitations stem from being based entirely on statistical information contained in text. They lack the tacit, non-linguistic reasoning that comes from interacting with the world. They don’t learn abstract concepts. In view of technology developments as we know them, the potential impact on the economy is considerable. If AI comes to perform tasks that require logical reasoning, the implications for its long-term economic impact would go beyond that.

So, while the potential of AI is clear, how this transformation will play out is highly uncertain