From “Can Machines Think?” to “Can Machines Be Governed?”

Reading time: 5 min

Note: This article reflects the author’s personal opinions and analysis and does not represent the views of any organization with which the author is or has been affiliated.

In 1950, Alan Turing published Computing Machinery and Intelligence. The article began with a question that, more than seven decades later, remains an unavoidable reference when we talk about artificial intelligence:

“I propose to consider the question, ‘Can machines think?’”

Turing wondered whether machines could think.

Today, machines generate text and images, analyze information, recommend decisions, write code, and are even beginning to execute actions within business processes.

Perhaps it is time to ask another question: Can artificial intelligence be governed?

From capability to governance

In recent years, much of the conversation around AI has understandably focused on its capabilities. What can it do? Which model should we use? Which process should we automate? How should we integrate it? How should we build a copilot or an agent? These are necessary questions.

But as AI moves from the laboratory and experimental projects into the everyday work of organizations, another set of questions emerges.

Evidence on AI adoption, implementation, risk, and governance suggests that the challenges accompanying its use are not exclusively technological. Issues also arise around accountability, authority, oversight, risk, data, security, third parties, and organizational capabilities. And that changes the conversation.

Because enabling an artificial intelligence system to do something is one thing, but deciding whether it should do it, under what conditions, and within what limits is something very different.

Questions that begin to emerge

Imagine an organization that begins by using an AI tool to help employees draft documents; then it uses another to analyze customer information; later, it introduces a system that recommends decisions; finally, it connects an agent to enterprise applications and allows it to execute certain actions.

At each step, technological capability increases, but different questions also emerge: Who can authorize an AI system to make a particular decision? Who is accountable when that decision has consequences? What happens when an AI tool appears in the company without authorization? Who decides what level of risk is acceptable? What changes when AI stops recommending and starts acting? Who can stop it?

None of these questions can be answered simply by choosing a better model. Nor do they disappear because the technology works correctly.

So, what does it mean to govern AI?

Here an additional difficulty appears. Talking about AI Governance may immediately bring to mind regulation, compliance, policies, committees, or new controls, but the issue may begin much earlier. It may begin by asking how an organization makes decisions about AI, who has authority, who assumes accountability, what must be known before a use is authorized, what happens when something moves beyond the intended boundaries. And what evidence of those decisions should remain.

Different standards, risk management frameworks, international principles, regulations, and business practices have begun to address parts of this problem. They do not all do so in the same way. Nor is there a single organizational structure that can be considered the right answer for every company.

For example, a large regulated corporation and a midsize company probably do not need exactly the same mechanisms. An internal chatbot and a system capable of taking actions that affect customers do not necessarily create the same requirements either. But behind those differences lies a common idea: the capability of an AI system and the authority an organization decides to grant it are not the same thing.

An increasingly relevant question

This distinction may become especially important as technology evolves because for years, much of enterprise AI was oriented toward analyzing, predicting, or recommending. Generative AI greatly expanded the possibilities for interaction and creation. Now, the development of agents introduces another possibility: systems that not only produce a response, but can also use tools, interact with other systems, and execute actions to achieve a goal.

Then the question changes again; it is no longer only about what information an AI system can produce, but also what we allow it to do with that information. And we inevitably return to the earlier questions: Who authorizes? Who is accountable? Who oversees? Who sets the limits? Who can stop it?

Can artificial intelligence be governed?

In 1950, Turing posed an extraordinarily difficult question. More than seven decades later, we are still discussing its implications while building machines with capabilities that at the time belonged largely to the realm of imagination.

The question of AI Governance is much more recent, and perhaps we are still learning how to formulate it correctly. Because governing AI does not appear to be solely a technology problem, nor merely a matter of regulation, nor exclusively one of risk. At its core, it seems to lead us toward something far more familiar to organizations: deciding who can do what, under what conditions, with what accountability, and within what limits.

Technology will continue to advance, capabilities will continue to change, and organizations will have to decide what to do with them. That is why, seventy-six years after Turing posed his question, and echoing his words:

“I propose to consider the question: ’Can machines be governed?’”