AI? Neither to Be Overestimated Nor Snubbed
Like all major technological revolutions, artificial intelligence has quickly triggered a division between doomers and boomers, that is, between the opposing attitudes of those who fear AI or exalt its virtues beyond any rational limit. AI, and above all generative AI, presents itself as such a radical revolution that everything suggests tackling it with the tools of rational analysis, measuring benefits and risks both in breadth, i.e. in the breadth of possible applications, and in length, that is, in the projection over the next decades. This is what Stefano da Empoli, president of the Institute for Competitiveness (I-Com) and Professor of Economics at the University of Rome III, does in The Economy of ChatGPT released by Egea, starting from the most discussed and famous chatbot of the new generation of AI.
In its short history, AI has already disappointed futuristic expectations several times. Why should ChatGPT be different and what are its economic consequences?
Chatbots like ChatGPT are an expression of generative AI, which is just one part of AI, but the one that is the most talked about. Presently, it represents a fifth of the AI market and in any case, according to growth estimates for this decade, it will never exceed a third of the total. The objective of the book, therefore, is not to discuss its greater or lesser relevance, but to refute the narrative that is made of it today. In fact, there are those who extol AI tools by overestimating them beyond all limits and therefore prefiguring a future in which machines will replace humans with catastrophic results. On the other side, there are those downplaying and trivializing AI, while highlighting its limitations and errors at every opportunity. The correct approach, however, is to start asking what kind of questions we need to ask generative AI and what kind of answers we can expect.
The ability to interrogate machines is the first requirement to obtain better answers. Do you feel like we're not asking AI the right questions?
We are too early for this to be the case. So far, generative AI has been questioned mainly by computer scientists, by scientists, by experts who know how to interact with machines. With ChatGPT this intermediation has been lost and we can all use AI, with consequences that today we are struggling to comprehend but which will be extremely far-reaching.
How do you build an economy, and therefore firms, investments, industries, training, on something that we don't yet know exactly where it will lead us to and according to what time frame?
As always, there will be industries invested first and others last, but I believe that the impact of generative AI will generally be faster than that previous industrial revolutions, such as electricity or information technology, for example, because it doesn't require huge infrastructural or technological investment. On the other hand, however, it requires a cultural and organizational rethinking of the way of we work and conceive employment. Therefore, everyone will try to implement generative AI in their activities, but only those who are quickest and most effective at changing their mindset will be able to put it to good use.
You dedicate a chapter to describing if and how AI increases productivity. Is it still a parameter that interests us, despite the growth in the share of immaterial products and services?
As an economist I feel like saying yes, productivity is still the basis of an efficient economy. We can discuss social costs, distribution, sustainability, but it still remains a fundamental parameter to assess. Even more so from an Italian perspective, since the decline of our country in the last thirty years is closely linked to its stagnation in productivity.
On a global scale, the US/China competition on automation and machine learning applications is increasingly evident. Could generative AI tilt the balance towards America?
Generative AI highlights the structural limits of the system built by Beijing. Although investment by Chinese companies is very high, some factors weigh on context, primarily internet censorship. Apart from the case of software code, for which the problem is less evident, for everything that concerns the generation of ideas, texts, images, videos and audio, the control of input data limits the results that chatbots can provide. Furthermore, most of the internet texts on which the models are trained are in English, and this is a great factor of competitive advantage for the US. On the other hand, outside the field of generative AI, the lesser attention to privacy in China makes applications that work with individuals' data simpler and more advantageous. However, there are also other issues than need to be considered: the US and their allies, starting with Taiwan, have quite clear control over the value chain of very advanced semiconductors and this greatly penalizes Chinese companies which have so far failed to attain the same standards with their domestic suppliers.
Between rules and investments, Europe has chosen to insist more on the former than on the latter, carving out the role of referee rather than player for herself. From your vantage point as member of the European AI Alliance promoted by the European Commission, what are the prospects for European companies?
In justification of Europe, I feel like saying that in Brussels it is certainly easier to regulate the economy than carry out industrial policy, because the latter is an area fiercely defended by national states. The so-called Brussels Effect arises from here, from the proven ability of the EU to provide well-structured legislative products to regulate various fields and industries, which are then adopted by other countries or multinationals, as occurred, for example, with the GDPR directive, the regulation that protects online privacy. For a long time, therefore, the idea prevailed that by regulating AI first, the same dynamic would be triggered, but generative AI has changed the picture. In recent months, for example, the debate has focused on how to regulate the foundational models underlying ChatGPT, because the text of the AI Act did not provide for them at all, limiting itself to regulating the uses of AI and its applications. The problem is that models like GPT-4 can be used for any activity, some considered high-risk activities according to the AI Act, others not at all. And while the EU is discussing, other countries are moving, the US first and foremost, taking other paths, such as exchanges with corporations and executive orders, the equivalent of our ministerial decrees, and they are taking this approach on a global scale through the G7. At this point, if the EU were to regulate these models too rigidly, it would paradoxically find itself slowing down innovation and damaging its companies on international markets.
However, your book ends on a note of optimism, also regarding the Italian situation. What makes you think our future could be bright?
As mentioned, AI requires little investment in technology, and this is a positive fact for Italian companies which are mostly undercapitalized and small in size. Even the theme of cultural change is easier to apply in less complex organizations, as SMEs typically are, in which fewer people decide, and tend to have a long-term vision, particularly if they are family companies in terms of ownership.
However, aging Italian entrepreneurs and the low quality of most management in small firms have inhibited much change in Italy...
That's true, and this is why I took the liberty of concluding my book with the proposal to allocate funds not only to investment in technology and training, but, even before that, to help companies evaluate the state of their technological preparedness and measure the divide that separates them from the industry benchmarks. There is a lot of focus on investment, but Italy needs to strengthen the preparatory phase, so that technology development programs, currently paid for mainly through Next Generation EU funds, can give better results.