Historical context: this article was published on August 25, 2023. It discusses a contemporary argument by Thomas Ramge and Viktor Mayer-Schönberger, not current product capabilities or a verified business outcome.
Can ChatGPT support executive decision-making? This August 2023 discussion considers the possibilities and limitations of using a language model as a conversational aid. Illustrative responses should not be mistaken for a verified product test: managers remain responsible for checking evidence and making decisions.
This dose of humility notwithstanding, language models like ChatGPT have the potential to become indispensable decision-making aids for both executives and organizations. Their true value lies not in supplying definitive answers but in guiding us through a more methodical decision-making process—something that’s often lacking, even in critical management decisions.
Informed decisions typically unfold in three phases. First and foremost, we must define our objectives and the context surrounding them. What precisely is the decision’s scope, and how does it align with our goals, values, and preferences? This initial phase establishes the decision-making problem and the framework within which it operates. The second step involves generating alternatives. What decision-making choices are at our disposal? The objective here is to cultivate a diverse array of possibilities, rather than fixating solely on the obvious ones. Only when we’ve extracted a sufficient pool of options from our decision-making framework can we effectively evaluate them and arrive at a well-informed decision—the third and final phase.
In the 2023 argument, ChatGPT could help structure discussion in all three phases of business decision-making. Practically speaking, this means we can engage in exploratory dialogues with the system throughout these phases. When assessing decision alternatives, for instance, we can inquire about the common missteps made by managing directors in the mechanical engineering sector when expanding into new markets. What were the key success factors for those who undertook successful expansions?
ChatGPT won’t provide us with a one-size-fits-all template for evaluating options in our unique context. However, it can aid in uncovering our inherent biases and prompting us to question preconceived notions. Prompts informed by the work of Daniel Kahneman and Amos Tversky can invite alternative perspectives, but the model can itself reproduce biases or agree too readily with a user. It encourages us to contemplate how we can assess options from a more well-informed perspective.
Moreover, ChatGPT may help us discover additional options that may elude our own creative thinking. This, in turn, broadens our decision-making horizons and reveals the multitude of possibilities, often more far-reaching than we initially perceive.
For instance, consider the challenge of reducing dependence on China and diversifying a supply chain—a decision unfamiliar to many managing directors and their teams. Here, ChatGPT could draw upon strategies documented online by companies in similar predicaments, potentially offering more innovative solutions beyond simply relocating production to Vietnam. Training on past text is not the same as current web access or verified company-specific evidence. Check supply-chain suggestions against current, reliable sources.
ChatGPT can also be prompted to help in setting goals and preferences, assessing decision-making circumstances, and selecting the appropriate framework. Dialogue remains pivotal. By asking the right questions, we gain deeper insights into the context of a decision. With ChatGPT, we can request suggestions for typical goals that other companies have pursued in similar decision-making scenarios. For instance, a query might read: “Hello ChatGPT, I lead a successful mid-sized tooling manufacturer near Columbus, Ohio, struggling to attract engineering talent. What could be the underlying reasons, and what strategies have similar manufacturing companies employed to address talent shortages?”
In summary, ChatGPT can act as a conversational sparring companion, but it does not remove responsibility for defining the problem, generating options and checking them. It has no human personal agenda, yet this does not make it objective or immune to groupthink: the GPT-4 technical report describes sycophancy, hallucination and overreliance risks. Its cost advantage over consultants or internal strategy teams depends on the work and the human review required. Smaller companies may find it useful for preliminary decision preparation, not as a replacement for accountable expertise.
Aspiring managers at business schools often indirectly acquire decision-making prowess through numerous case studies. The goal is to cultivate a repertoire of decision-making models by crafting and evaluating actionable options within a decision-making framework. Naturally, case studies don’t proffer perfect solutions to specific decision-making scenarios; instead, they pose questions, present frameworks, and outline potential choices. Case studies may also be useful for model research, subject to access, copyright and licensing; this article does not establish which collections were included in training.
The original discussion proposed cooperation with case-study providers such as Harvard Business Publishing and the Case Centre. The precise proportion of their material in ChatGPT training was not established, and GPT-4’s published technical report does not disclose a complete training-data inventory. Properly licensed domain material could be explored for decision-support tools, alongside programming, copywriting and customer-service uses; it does not guarantee reliable business advice.
Moreover, future advancements in learning algorithms may enable more efficient models, paving the way for “medium-sized language models” that no longer require ingestion of vast swathes of the internet and complete libraries. Such models could focus primarily on texts and documents relevant to specific domains. This was a forecast, not an inevitable or verified transition. “DecisionGPT” was a speculative label for a possible domain-focused assistant, not an announced OpenAI product.
A proposed use of ChatGPT, and that of similar systems, lies in their ability to juxtapose and contrast analogous situations—a critical need in many management decisions. Rarely are these decisions wholly unique; countless managers before us have grappled with similar choices. The better we articulate these decisions in human language, describing how the decision-making framework was established, options were weighed, and choices were made, the more useful a carefully evaluated decision-support system might become for fostering well-informed decisions.
In due course, numerous management decisions could become ripe for automation. Robo-managers might be deployed more frequently and sooner than many corner office executives currently anticipate. This is a historical forecast, not evidence that autonomous systems should make consequential management decisions without oversight.
However, for now, the advantage lies with those managers who harness available tools to refine their decision-making processes. Rather than seeking definitive answers from models like ChatGPT, we should engage them throughout each stage of the decision-making journey.
Sources: Original August 2023 management argument · March 2023 GPT-4 technical report and limitations
