A Framework for Technology Decisions in the Age of Generative AI
By Frank A. Schmid, CTO, Gen Re
Generative AI is a general-purpose technology (GPT). The arrival of a GPT is a rare event, even in modern times. Among the GPTs that have emerged since the Industrial Revolution, electricity and the semiconductor are notable examples. Generative AI may be the most general of GPTs because it rivals human intelligence and cognition.
A defining characteristic of a GPT is its pervasiveness, meaning it finds its way into a wide range of applications in both business and consumer sectors. Two other key properties of a GPT are its capability for ongoing improvement and its ability to spawn a co‑invention process in the application sector. Businesses co-invent by redesigning workflows, adopting novel organizational designs, and developing new products and distribution channels.
In the Generative AI adoption process, upstream technological advances and downstream co‑invention form a feedback loop that persists over a prolonged period. The adoption process may extend over decades, as was the case with electricity and the semiconductor. Viewing the release of ChatGPT in November 2022 as the arrival of Generative AI in the application sector, we can conclude that we are in the initial stages of the Generative AI adoption process. Importantly, the feedback loop’s velocity may be higher for Generative AI than for earlier GPTs.
The adoption of Generative AI is a discovery process where learning is central. The adoption process is expected to be prolonged, iterative, and potentially nonlinear.
Organizations differ in their costs of adopting a new GPT, and these costs are significantly influenced by the existing technology stack. What are the potential implications for how we make technology decisions in this rare event of adopting a new general-purpose technology?
Adopting Generative AI comes with a high degree of uncertainty. There is uncertainty about the pace of technological advances, the direction of these advances, and the trajectory of costs for various functionalities. As time progresses, uncertainty resolves, and organizations learn. Indeed, learning is a central feature of a GPT adoption process.
Considering that the adoption of Generative AI is a prolonged process accompanied by learning, it is crucial that today’s technology decisions deliver a high degree of reversibility. Reversibility is associated with creating optionality and avoiding the stranding of newly created assets.
In a technology decision, we may commit an error of commission by moving forward with a project only to regret its execution as circumstances change. Such changes may be related to technological advances favoring an alternate approach to adoption or changes in business conditions related to competition or macroeconomic factors. Conversely, we may commit an error of omission by deciding not to execute a project and then regretting not having created optionality for follow‑on technology projects or newly arriving business opportunities.
If the future were certain, there would be no learning, and reversibility would not be a factor in technology decision-making. Also, if projects were reversible at no cost, there would be no stranded assets if we decided to abandon the project or change its course. Hence, it is the concurrence of learning and reversibility that creates a challenge for decision‑makers. Interestingly, the decision not to execute a project is always irreversible.
The traditional net present value rule advises executing a project if and only if the net present value is positive. Where learning and irreversibility exist, real options complement the traditional net present value rule.
Real options come in two forms. The potential for an error of commission creates value in waiting, which subtracts from the traditional net present value of the project. Conversely, the potential for an error of omission creates a cost of waiting, which adds to the traditional net present value of the project, favoring execution.
Real options valuation can be challenging to implement as the parameters that enter the calculation may come with a high degree of uncertainty. In a pragmatic approach, decision-makers can consider real options by following two principles. First, the value of waiting favors technology decisions that create a highly reversible technology stack. High reversibility implies that assets created in a technology project are portable to an altered or alternate project. Second, the cost of waiting favors technology decisions that increase the options for future technology choices. This aspect is key where technology is sequential in nature. As the cloud is a prerequisite for a modern data architecture, so is a cloud-based data foundation for low costs of Generative AI adoption.
In summary, the adoption of Generative AI is a discovery process where learning is central. The adoption process is expected to be prolonged, iterative, and potentially nonlinear. In the technology decisions driving this adoption process, the value of waiting and the cost of waiting are significant factors. Keeping the value of waiting low allows the organization to move forward with little regret. Increasing the cost of waiting allows the organization to move quickly.
Author Bio:
Frank Schmid is Chief Technology Officer at Gen Re and a member of the executive board of Gen Re AG in Cologne. Previously, Frank was affiliated with AIG, the National Council of Compensation Insurance, and the Federal Reserve Bank of St. Louis. Frank holds a doctorate in economics (Dr. rer. pol.) and a post-doc in finance (Dr. habil.) from Leuphana University of Lüneburg, Germany, where he held an appointment as an extraordinary professor of business administration. Frank taught finance at the University of Vienna, Goethe University Frankfurt, Free University of Berlin, Leuphana University of Lüneburg, and other academic institutions. In 2006, the European Economic Association awarded Frank the Hicks-Tinbergen medal, jointly with Gary Gorton of Yale University.