Plato described writing as a pharmakon, a substance that can heal or harm depending on its use, in the dialogue Phaedrus. Today, the same dilemma resurfaces as generative AI tools flood newsrooms, law firms and research desks, prompting a fresh clash over whether these systems are a boon or a threat.
Ancient roots of the argument
The story begins with the Egyptian myth recounted by Plato, in which the god Theuth offers the gift of writing to King Thamus. Thamus warns that writing will foster forgetfulness, serving only as an aid to recollection rather than true memory. The philosopher Jacques Derrida revisited this passage in his 1968 essay "Plato's Pharmacy", arguing that the dual nature of the pharmakon cannot be fixed as either remedy or poison.
From Derrida to today's newsroom
That paradox sits at the heart of the AI-writing debate. When used by a knowledgeable person, AI can retrieve data, organise ideas and suggest revisions, a modern remedy. When it replaces human judgement, it can generate fluent prose without understanding, the poison. European media organisations have already tested the balance. In 2014 the Associated Press automated corporate earnings stories, expanding coverage from about 300 companies per quarter to 4,400. Bloomberg followed a similar path, using machine-learning to assist roughly a third of its content by 2019.
"AI is the deus coming out of the machina, more than the machina," said Drew Lichtenberg, dramaturg at Washington's Shakespeare Theatre Company.
These examples show that the technology can increase output while freeing journalists for deeper reporting, yet they also raise questions about attribution and the loss of the human touch.
The modern dilemma for content creators
Editors now face a choice reminiscent of the myth of Scylla and Charybdis. Steering closer to Scylla, the loss of a few skilled staff, may be preferable to the whirlpool of total obsolescence that comes from refusing AI altogether. The case of investor Stanley Druckenmiller illustrates the tension: his Wall Street Journal op-ed was flagged by the detection tool Pangram as fully AI-generated, a verdict he confirmed without public disclosure. The paper's editorial page editor, Paul Gigot, defended the silence, highlighting the growing acceptance of AI-assisted writing among influential figures.
What happens next?
European regulators are beginning to address the issue. The European Commission is consulting on transparency obligations for AI-generated content, and the EU's upcoming AI Act may require clear labelling of machine-authored text. For newsrooms, law firms and research teams, the challenge will be to define the acceptable level of machine assistance while preserving the credibility of their work. The debate is unlikely to end soon; instead, organisations will need to develop policies that balance efficiency with responsibility, ensuring that the tool remains a remedy rather than a poison.
In the end, the ancient warning still holds: every advance comes with a cost, and the task is to decide which losses can be endured without compromising the core mission of truth and insight.