In my previous columns, I conducted a thought experiment on a fully automated society, in which AI and robots can produce almost all the goods and services necessary for society. I explored income distribution in the first column on the matter and how to use freed-up time and “participation and co-creation” in the second. In this column, I would like to think about savings and investment over a somewhat longer time horizon.
Savings in a fully automated society
We save to prepare for old age, job losses, illness and to purchase homes and cars. In economics, theories such as the life-cycle hypothesis which assumes a leveling out of consumption throughout an individual’s life (Modigliani and Brumberg, 1954), precautionary savings to prepare for future uncertainty (Kimball, 1990), and buffer-stock savings for income fluctuations (Carroll, 1997) provide explanations for such saving behaviors.
If, however, access to essential goods and services is disconnected from labor income in a fully automated society, the need to prepare savings to manage risks associated with job loss or retirement will diminish. The role of borrowing, including mortgages, as a means of bringing forward future labor income may also diminish.
Furthermore, the need for long-term savings or loans to own homes or cars will diminish as the sharing economy expands. What will matter is access to services when needed, rather than ownership itself.
Disaster preparedness will also change. At present, accurately predicting earthquakes before they occur is considered difficult. In this thought experiment, it is assumed that earthquakes and typhoon paths will be forecast sufficiently in advance and with high accuracy, increasing safety. Even if disasters themselves cannot be controlled, the ability to pre-position necessary supplies and recovery robots based on such predictions would shift the focus from precautionary stockpiling to allocating resources based on forecasts.
Investment in a fully automated society
Even if the need for saving diminishes, the need for investment would not disappear, because savings and investments serve different economic functions. If savings transfer current purchasing power to the future, investment uses current resources to develop future production capacity.
Even in a fully automated society, tangible fixed assets, such as power plants, data centers, robots, and homes, must continue to be maintained and replaced. If existing technologies, demand, prices, and useful lives of equipment are known, deciding which equipment to replace, when to replace it, and to what extent, becomes a dynamic optimization problem, albeit a complex one.
There is a long tradition of corporate investment studies, including Jorgenson’s (1963) neoclassical investment theory, which considers the desirable capital stock based on the user cost of capital, and Hayashi (1982), who organized the relationship between optimal capital accumulation and the q theory, demonstrating that marginal q coincides with observable average q under certain conditions.
This type of investment decision-making is likely to be a domain where AI and mathematical optimization excel. They can calculate demand, failure probabilities, and resource constraints to develop optimal replacement plans. The same holds for disaster preparedness if predictions are accurate enough. Physical investment will not disappear; rather, it would become part of the routine operation of maintaining an automated economic system.
Meanwhile, investment targets are not limited to machines or buildings. Intangible assets, such as software, databases, research and development, and organizational capabilities are also important forms of capital (Corrado, Hulten and Sichel, 2005). Particularly, investment in R&D and innovation differs in nature from ordinary physical investment. It is often unclear in preliminary stages what should be researched, what will be successful, or what successful technologies will be used for. Rather than merely expanding existing productive capacity, such investment pushes the production possibility frontier outward and may even transform the range of choices available to the economy itself.
In a fully automated society, AI will take on much of the work involved in information gathering, analysis, design, programming, and other tasks. Humans may gain more freedom to experiment with new ideas than ever before. The “participation and co-creation” taken up in my previous column could also generate new knowledge and ideas.
This brings Schumpeter to mind. In The Theory of Economic Development, he distinguished between “circular flow,” where the economy cycles under known needs, technologies, and production methods, and economic development driven by “new combinations” (Schumpeter, 1934).
Investment for the Future—Will applications for KAKENHI disappear?
A familiar example related to the selection of investment targets for the future is the Grants-in-Aid for Scientific Research (KAKENHI) program. As I write this in September, many researchers in Japan are busy applying for Grants-in-Aid for Scientific Research. In a sense, the KAKENHI program can be viewed as a mechanism for determining which future possibilities should receive a portion of the limited research resources that are available.
In a fully automated society, the work of writing grant proposals will be almost entirely automated by AI. However, as long as such resources as experiment facilities, computing resources, and energy remain limited, the question of which research projects should be selected for the allocation of limited resources will remain.
AI may automate much of what Schumpeter calls “circular flow.” However, optimizing circular flow is not equivalent to creating new combinations. To realize new combinations, limited resources must be allocated toward new purposes.
The more possibilities humans and AI create, the more difficult the prioritization of resources among those possibilities will become.
Will Grants-in-Aid for Scientific Research applications disappear in a fully automated society? In my next column, I would like to consider innovation based on this question.
October 9, 2026
>> Original text in Japanese
