In recent months, there have been murmurs about the failure of artificial intelligence to live up to the hype. As AI company valuations and investments swell, onlookers wonder whether AI will be able to deliver its promises and recent history offers plenty of reasons to be sceptical.
Technology hype waves, such as those surrounding cryptocurrencies and the metaverse, have come and gone. Legitimate questions remain about whether the current wave of AI offers real value or instead serves as another example of zealous speculative investments, driven perhaps in part by the lingering impacts of a protracted period of low interest rates.
AI’s core contribution today takes at least three forms: improvements in productivity, new products and the creation of artificial general intelligence. Despite several reasons to believe the AI wave is over-hyped in many domains, there are good reasons to believe AI will buck the trend of ‘revolutionary’ technology failing to deliver its promises.
Labour productivity
Perhaps the most widely accepted value proposition for AI comes from its potential to bolster labour productivity. Productivity appears to have lagged for decades in many rich democracies and there are competing perspectives on whether the digitalisation of the early 21st century was productivity-boosting.
AI has been touted by economists, consultants, bankers and technologists as the application of ‘digitalisation’ that would finally generate higher productivity growth as well as improvements to global standards of living. These predictions, echoed throughout the last 20 years of AI discourse, have had a central problem: AI has been almost entirely absent from macroeconomic data. Due to this, AI’s productivity effects have been – and continue to be – relegated to fiction or speculation. This remains true for all ‘AI’ systems, except for, perhaps, industrial robots, where there may be a small discernible impact on macroeconomic figures for some rich democracies.
Yet the proliferation and popularisation of large language models mark a decidedly new stage in the history of AI. Preliminary evidence – through both experimental and observational methods – suggests that LLMs do boost productivity across many major industries. These effects appear particularly pronounced when some element of ‘content creation’ is involved, such as in writing and coding. Many major companies are introducing LLMs and other AI tools as part of capital-intensive strategies to reduce costs, increase efficiency and improve labour output. It will take some time for these shifts to manifest in national-level macroeconomic data, but top economists model that the current wave of LLMs will boost productivity.
These effects may not be especially large, yet the technology is improving at breakneck paces, with the biggest technology companies in the world racing for AI primacy. It is not particularly radical to assume those predicted productivity effects will increase as the technology becomes more advanced and applicable across more diverse and complex tasks. For instance, LLMs are increasingly used in diverse sectors, including banking, where they assist with customer service and fraud detection, and healthcare, where they provide patient support and aid in the analysis of medical data.
Novel practical applications
Emerging evidence that AI is boosting productivity is accompanied by the continued development of useful AI-enabled products. Much of this predates the public reckoning with LLMs and includes AI inventions in robotics. In manufacturing, AI-driven automation has been used to streamline production processes, improve quality control and optimise the supply chain. In waste management, AI is used to support the recycling process. In healthcare, AI is supporting diagnostics and treatment where machine learning algorithms can, for instance, analyse medical images with human-level or greater accuracy, helping detect diseases at early stages. AI-powered robots are assisting surgeons and developing personalised medicine, tailored to individual genetic profiles. DeepMind’s AlphaFold famously cracked the code of protein structures, allowing scientists to accelerate drug discovery and understand diseases at a molecular level.
In the financial sector, AI is enhancing fraud detection and risk assessment, enabling institutions to make more informed decisions and protect themselves from financial losses. AI-powered chatbots and virtual assistants are providing personalised support and guidance around the clock. Companies have emerged primarily offering AI products that can operationalise unstructured or qualitative data to improve financial benchmarking of key variables. This includes, for example, initiatives by Banque de France to use AI to better gauge consumer confidence through application programming interface web scraping.
Artificial general intelligence
Finally, there is the more theoretical promise of AGI – machines with the ability to understand, learn and apply knowledge across a wide range of tasks as flexibly as a human. The idea of AGI has clearly been a topic of fascination, not just within the AI research community but also in popular culture and there is perhaps no goal greater than AGI that seems to animate leading AI companies today. Yet, AGI remains a generally ill-defined and contentious term, as does the debate about how far along humanity is in developing the technology. OpenAI, perhaps the world’s leading AI company, seems to believe in a ‘scalability’ hypothesis for achieving AI. That is, if OpenAI can muster up the capital investments in compute, data and talent, the improvements in existing LLMs should continue to ‘scale’ to reach AGI.
This slightly crude ‘brute-force’ method is a disputable way of achieving AGI. Beyond potential environmental impacts, critics point out that LLMs, despite their impressive capabilities, are fundamentally statistical models that lack true understanding or consciousness. They excel at pattern recognition and can mimic human-like responses, but they do not possess awareness or the ability to reason beyond their training data. Critics argue that achieving AGI would require more than just scaling up existing models; it would necessitate breakthroughs in areas such as reasoning, learning from minimal data and the development of common sense – abilities that current AI lacks. Furthermore, the human brain operates on principles vastly different from those of artificial neural networks, suggesting that mimicking the brain’s functionality may not simply be a matter of increasing computational power.
Despite reasons to remain sceptical about the imminent promise of AGI, it does seem that AI is likely to mostly live up to its promises. Investors and business owners may be disappointed by perceived lagging gains, amid soaring AI costs. We may witness another instance of a hype wave that is followed shortly by a period of protracted disappointment and AI has been through several ‘winter’ periods before. Yet the bend of the technology appears to be one destined to be – perhaps imminently – transformative.
Julian Jacobs is Senior Economist, Digital Monetary Institute at OMFIF.
