The Economic and Business Impacts of Artificial Intelligence: Reality, not Hype
The debate on Artificial Intelligence (AI) is characterized by hyperbole and hysteria.
The hyperbole is due to two effects: first, the promotion of AI by self-interested investors. It can be termed the “Google-effect”, after its CEO Sundar Pichai, who declared AI to be “probably the most important thing humanity has ever worked on”. He would say that. Second, the promotion of AI by tech-evangelists as a solution to humanity’s fundamental problems, even death. It can be termed the “Singularity-effect”, after Ray Kurzweil, who believes AI will cause a “Singularity” by 2045.
The hysteria similarly arises from two effects: first, from warnings that AI poses an existential threat. It can be termed the “Elon-Musk-effect” after the billionaire entrepreneur who tweeted that “Competition for AI superiority at national level most likely cause of WW3 imo”. Second, from warnings that AI could cause mass unemployment through job automation. This can be termed the “Robot-effect” after the bestseller of Martin Ford entitled “The Rise of the Robots: Technology and the Threat of Mass Unemployment”.
In a recent Discussion Paper I provide a critical survey of the Google-Singularity-Elon-Musk-and-Robot-effects,and argue that hard evidence to support hyperbole and hysteria is lacking.
Back in 2013 it was estimated that 47 percent of jobs could be automated in 10–20 years in the USA and even more in the EU and developing countries. Six years later, instead of mass unemployment, unemployment in advanced economies are in fact historically lows. It has been shown that the methods used to calculate potential job losses due to AI are sensitive to assumptions. Moreover, the evidence indicates that automation has created 1,5 million net new jobs between 1999 and 2010 in Europe.
At the same time, we have seen a continued decline in labor productivity growth. The UKs ten-year average labor productivity growth since 2007 was the lowest since 1761.Even global superstar firms, who may benefit most from AI, have not become more productive. This cast doubt on the claims that AI has and will enhance productivity significantly.
Why are the Robot-and-Google-effects not materializing? There are at least three reasons:
First, the diffusion of AI through the economy is slower than most people think. It is especially difficult for small firms to economically implement AI. Growing Pseudo-AI is a result of this. The Guardian points out “It’s hard to build a service powered by artificial intelligence. So hard, in fact, that some startups have worked out it’s cheaper and easier to get humans to behave like robots than it is to get machines to behave like humans”.
Second, AI innovation is getting harder; and it is mostly applied to fine-tune and disrupt existing products rather than introduce radically new products. It may be entertaining to play with Google’s Bach doodle, but it hardly raises productivity. The low-hanging fruits of applying Machine Learning (ML) may have been reaped, decreasing returns seem to have set in, and on top of this that ML is facing a reproducibility crisis. The end of Moore’s Law may be in sight.
Third, it is not profitable for businesses to invest in AI given slow growing consumer demand in most western countries. Most AI innovation is in visual systems for autonomous vehicles. Despite this, autonomous vehicles are most notable for their absence from our roads. This is likely to remain the case for a long time, for technical reasons, and due to sunk investments.
One may argue that just because AI’s impact has been small in the past this does not rule out that the massive impacts will still happen in the future. Maybe the robocalypse is inevitable because of progress in AI. Such an argument is based on misunderstanding AI. The term “Artificial Intelligence” itself is misleading. Current AI, using ML, is not intelligent. A joke about AI is: “When you’re fundraising, it’s AI. When you’re hiring, it’s ML. When you’re implementing, it’s logistic regression”. There are various reasons to be skeptical whether non-ML AI research will result in a super-intelligence soon, and not just one-trick ponies that ML applications currently are.
As a result of hype and hysteria many governments are scrambling to produce national “AI strategies”. Global governance organizations are rushing to be seen to take action. It has become fashionable to hold conferences and publish flagship reports on the “Future of Work”.
The United Nations’ Secretary-General has, for the first time in history, published a “Strategy on New Technologies, singling out certain technologies, including AI, for special attention, based on the belief that “automation, artificial intelligence and robotics promise enhanced economic growth, but they can also exacerbate inequality within and between nations and can contribute to unemployment”. Taking its cue from this strategy, The United Nations University’s Centre for Policy Research (CPR) goes even further in justifying the UN’s planned intervention in the field of AI by claiming that AI is “transforming the geopolitical order” and even more incredibly that “a shift in the balance of power between intelligent machines and humans is already visible”. Its blog have called for “an Intergovernmental Panel for Artificial Intelligence” and for a “UN-led multi-stakeholder global governance regime”. Yes, it sees AI as of the same complexity and magnitude as climate change. There are many other examples of AI hyperbole and hysteria leading to crazy proposals.
Singling out AI for control and regulation by the UN, governments, or even an intergovernmental panel on AI, unproductively shift the focus towards the technology and not the real problems. Technology is a “moving target”: imagine if during the second industrial revolution an intergovernmental panel was put together to “globally govern” electricity? The case of electricity in the late 19th century actually offers pertinent historical caution. As Carolyn Thomas de la Pena recounts (p.113), hysteria broke out in some quarters: “Doomsday predictions were made by those fearful of electricity’s deviation from their perceived natural order. Clergymen were particularly prone to this view…According to Bishop Turner…there was much to be feared from ‘the invention of the white man in controlling electricity’”.
Government regulation of technology and innovation is at the best of times fraught with difficulties and unintended consequences. When it is based on hysteria and hyperbole, the task may be particularly problematic. And when such a problematic task is taken up by global political bodies, whose decisions often are made in an “evidence-free zone”, caution is advised.
The upshot is that the hype and hysteria about AI has led to an “unhinged” debate about AI, and is now encouraging stifling regulations as well as AI “arms races”.
These consequences could hasten a premature AI-winter through inappropriate controls and a loss of public trust, unfortunately at a time when the world needs more, not less, technological innovation, and for this technology to diffuse much faster.
An earlier version of this piece was published as a Global Business School Network (GBSN)blog and is a based on my webinar for the GBSN with the same title.
April 2023 Update
Since this article was written in 2019, AI Hyperbole and Hysteria has risen to unprecedented heights through all four of the effects that I outline. The Singularity effect has become a bit subdued reflecting that public distrust in AI has grown, and that AI Doomerism has become much more prevalent. The latter has been given a huge boost by the growing number of Existential Risk Institutes (who would be out of business if there were no more existential risks) and the March 2023 Open Letter by a large number of AI scientists and others calling for a six month pause in development of large language models such as GTP-4.
The debate about AI has become even more unhinged, with one AI researcher even calling for all AI research to be shut down and warning that AI will “kill us all”.
As an antidote to this AI Doomerism, I have written the following articles, available on Medium:
- Airstrikes on Rogue AI Datacenters? Fears that Artificial Intelligence Pose an Existential Risk has Reached New Heights of Hysteria
- The Stop Button Society: Longing for the end of capitalism?
- Existential Risk and AI: Should we Fear Extraterrestrial Artificial Intelligence?
Further reading:
Gries, T. and Naudé, W. (2022). Artificial Intelligence in Economics, Journal for Labour Market Research, 56 (12).
Naudé, W. (2021). Artificial Intelligence: Neither Utopian nor Apocalyptic Impacts Soon, Economics of Innovation and New Technology, 30(1): 1–23.
Naudé, W. (2020). Artificial Intelligence vs COVID-19: Limitations, Constraints and Pitfalls, AI & Society, 35 (3): 761–765.
Naudé, W. (2019). The Race against the Robots and the Fallacy of the Giant Cheesecake: Immediate and Imagined Impacts of Artificial Intelligence, IZA Discussion Paper no. 12218. Bonn: IZA Institute of Labor Economics