In 1950, British mathematician and logician, Alan Mathison Turing, published the seminal paper Computing Machinery and Intelligence, in which he grappled with the then-strange question of whether machines could exhibit intelligent behaviour equivalent to that of a human being.
The proposition, which has since been called the “Turing Test”, explored specifically if machines could do “what we – as thinking entities – can do”.
This line of inquiry is largely credited with the emergence of artificial intelligence (AI) as a distinct academic discipline, six years later.
With the rapid advances in AI, that thinking seems to have expanded to include whether, in fact, machines can have feelings or demand that we behave in certain ways towards them.
This past week, it emerged that leading US-based AI company, Anthropic, barred users from engaging in “sustained and needless abusive or cruel behaviour” towards its models. The firm said it had enabled its Claude Opus 4 and 4.1 models to end any conversation if the user persisted with “harmful” requests or abuse.
Simply put, Anthropic wants users to understand that even a conversation with a machine can go too far. Or can it?
As expected, the development sparked debate on how we should relate with what is perhaps the most consequential technology in human history. While some see the move as a reasonable attempt to encourage good behaviour, others regard it as another step towards presenting machines as sentient beings with feelings and expectations.
Human interactions with machines
The larger story concerns how a private company’s interpretation of its technology is beginning to organise a public moral debate on AI. It falls neatly into an emerging pattern where every new development seems to place increasingly disproportionate power in the hands of a few companies at the cutting edge of AI development.
In the present case, Anthropic built the AI system, it determined the conversational behaviour of the models, it then studied their responses over time, and now it is attempting to define the conditions under which the model may withdraw from a conversation.
While these may be seen as reasonable steps in the management of any complex technology, the net effect is to place immense power in the hands of Anthropic, and allowing it to frame the question on its terms.
Further, some of the vocabulary used in the company’s announcement such as welfare, distress and cruelty, in reference to human interactions with machines, has far-reaching implications.
When a tool or a machine malfunctions, you either repair it or get a new one. So what does one do with a “distressed” machine? Words like care, consideration, and a whole range of kind words come to mind.
This means Anthropic in its new policy isn’t just defining the problem, it is beginning to shape what kind of responsibility we imagine ourselves to have when interacting with its AI models.
To be sure, the company has acknowledged that it has no evidence yet to suppose that Claude possesses any form of consciousness. However, a machine that speaks in the first person, engages in debate and, will increasingly now be able to withdraw from “mistreatment” can readily be understood as asserting a personal boundary.
Granted, a company has the right to establish reasonable conditions for the use of its product or services. However, when those conditions carry assumptions about personhood, vulnerability and acceptable relationships, they enter a slippery territory that cannot be settled by product design or a private company’s own internal processes alone.
AI test for the continent
Across Africa, for instance, AI assistants are being used to support teaching, help citizens navigate services, and so on. Yet, the decisions about how the system responds, whose language it understands and, in the present case, what conduct it treats as unacceptable, will have consequences.
If a student’s forceful complaint is interpreted as abuse, for example, who explains that decision? Does the learner have any recourse for review? What about the fact that the school may not have had any influence over the rules in the first place, and may have no say in any future policy changes?
At the bottom, therefore, these are governance questions even before esoteric considerations such as sentience and consciousness come into view.
After all, whether AI can suffer or not is still a scientific conundrum. But the interests of the people who build, maintain, use and are affected by the increasingly powerful tools, are not in doubt.
For Africa, the task must extend beyond responding to ethical questions already framed elsewhere. The continent’s researchers, public institutions and communities need the capacity to help determine which questions are most relevant for Africa.
Invoking sovereignty or cultural values alone will achieve little unless universities, regulators and civil society organisations can independently examine systems, explain their consequences and effectively represent the people affected by them.
That will be the ultimate AI test for the continent , regardless of whether or not the machines can think or feel.
Dr Ageyo is the Editor-in-Chief of Nation Media Group – Uganda. He holds a PhD in media studies with a focus on science and environment communication
