Dusseldorf Thomas Edison lit the lightbulb in 1879. Anyone who thinks that a wave of inventions and changes followed is wrong. Two decades later, only 3 percent of US households were connected to electricity, and the figure for factories was hardly higher. It took four decades for electricity to catch on. Why so long?
This is the question Ajay Agrawal, Joshua Gans and Avi Goldfarb ask themselves in their book Power and Prediction. Their analysis is worth reading, if only because of their expertise: The three Canadian economists and AI experts became known in 2018 for their book “Prediction Machines”. The bestseller described the advances in artificial intelligence (AI).
“This is the best book about the best technology that’s coming our way,” said Lawrence Summers, professor and former chief economist at the World Bank. Currently everyone is talking about ChatGPT and AI. The advances are astounding, due to larger data sets and computer capacities, neural networks and machine learning models recognize objects or people better and better, write texts and analyze chemical components or X-ray images.
However: Even five years ago, the enthusiasm for AI was great – only to fall asleep afterwards. And that could happen again. Because we are in the “in the meantime”, as the authors describe it. By this they mean the period of time that elapses between the introduction of a fundamental innovation and its economic consequences.
In the case of business consultancies, this is called the “Hype Cycle”: First, a new technology is greeted with euphoria and its consequences are overestimated. Then follows a phase of disillusionment and disappointment – in which the consequences are underestimated.
Ajay Agrawal, Joshua Gans, Avi Goldfarb: Power and Prediction
Harvard Business Review Press
Brighton 2022
288 pages
26.99 euros
It’s refreshing to read the authors accusing themselves of exuberance. “Our speculations have been proven wrong — dead wrong,” it says of their previous predictions about the “commercialization of AI.” They had predicted a rapid triumph.
They quote Geoffrey Hinton somewhat apologetically. In 2019, the professor and “godfather of deep learning” predicted that the profession of radiologist would soon come to an end – because AI can evaluate X-ray images better. Technologically, Hinton was right, the economists write in the book, but there would by no means be fewer radiologists.
How could they be so wrong? The authors pursue the question on more than 250 pages. Using electricity as an example, they clearly describe why key innovations take time.
Their power of change is almost too great, at first they are hardly used. At the end of the 19th century, for example, factories were operated with steam engines, and switching to electricity was out of the question for most owners. After all, new machines and turbines cost money.
If you start early, you will reap the benefits faster
Only over time did the real advantage of electricity crystallize: unlike in steam operation, the machines do not have to be close to the energy source. This enables a different production structure, such as Henry Ford put into practice with assembly line production – but only in 1914, a full 35 years after the lightbulb lit up.
In particular, industries that were emerging at the time, such as the automotive and chemical industries, were the first to rely on electricity. Just like today, companies like Amazon, Google or Meta are heavily using AI with their digital business models.
The “innovator’s dilemma” prevents long-established industries from repositioning themselves, the authors write: Although companies see the benefits of AI, they fear changing their well-running business. Tech giants like Google are also concerned. The search engine provider fears the impact of AI on business, which is why, according to US media, its own AI Lamda was withheld.
As a way out, companies choose a path of compromises: They introduce AI selectively. How banks are using AI to detect financial fraud. It’s like the lightbulb on the factory floor: The application does not disrupt the previous business model and can be seamlessly integrated.
However, the true power of AI lies elsewhere – in system change. How the electric motor eventually replaced the steam engine. But that takes time, there is a lot of resistance. Not just for employees or managers who shy away from change. According to the authors, the problem goes deeper: in the “standard procedures” that have become anchored in the regulations, hierarchies and rules of a company.
More about AI:
This ensures that the individual parts of the company and the employees can work together as efficiently as possible. Rules are the “glue” that rules out mistakes and guarantees reliability.
However, rules also prevent reforms. The potential of AI cannot be fully exploited: namely to make precise predictions that cost next to nothing. This changes the decision-making process in a company. So far, predictions and decisions have gone hand in hand; AI solves this and enables new business ideas.
An example would be the delivery of goods before customers have ordered them. Amazon and other companies are experimenting with it. AI could make it possible for us to have a supply of goods on our doorstep every day, from which we take those we need. Shopping would be abolished.
However, the idea encounters many problems: what to do with the returned goods? Transport costs would also have to be much lower than they are today. This example shows the thinking of the authors: the AI can predict our shopping behavior quite precisely, but the implementation comes up against system limits. But it would be conceivable to convert the system in such a way that the idea works: with efficient transport and reuse of returned goods.
According to the thesis, AI will have the greatest impact in industries and companies in which there are many rules and “hidden uncertainties”. In airports, for example, the many shops are an expression of the passenger’s uncertainty as to when exactly he has to leave home to catch the plane. An AI-driven app could change that — and threaten store revenue.
Personalized learning with AI
Another example in the book is schools. There are a lot of rules to ensure that all students get an adequate education. It is clear that children learn at different speeds and have different talents. But they are still pushed into classes with the same learning content.
Personalized learning could ease the tension, the authors argue. AI trained for specific learning content conveys mathematics or grammar better and more directly to individual students. Teachers can focus on group work or specific students to solve social or other problems.
The conclusion of the authors: Even if it will still be some time before AI will revolutionize our economy like electricity once did – don’t sit back and relax. There are huge benefits to using AI earlier: “The sooner it’s implemented, the better the prediction.” If one AI is just a little better than another, more customers will use it. “With more users, the AI gets more data, with more customer data, the AI makes better predictions. That attracts new customers.”
More: These are the best-selling business books in January