The advertising industry has spent the better part of a century trying to prevent human beings from making expensive mistakes with other people's money. Now it must figure out how to extend the same courtesy to software.
As artificial intelligence agents begin to take on media buying tasks that were once the exclusive province of junior planners and their spreadsheets, a new anxiety has settled over the holding companies and their clients: What happens when the machine gets it wrong?
The question is not entirely hypothetical. Media buying has always been a discipline where small errors compound quickly — a misplaced decimal, a forgotten exclusion list, a creative asset served to an audience that was never meant to see it. Human buyers have traditionally caught such mistakes through experience, institutional memory and the low-grade paranoia that comes from knowing one's name is attached to the insertion order.
Artificial intelligence agents, whatever their other virtues, do not yet possess that paranoia.
The emerging consensus among those who sell such systems and those who must live with them appears to be that guardrails are essential — though the precise nature of those guardrails remains a subject of some debate. Some favor human-in-the-loop approaches, in which a person must approve any action above a certain threshold. Others prefer automated verification systems that check the agent's work against predefined rules.
(One might note that "automated systems to check automated systems" has a certain recursive quality that would have amused the systems theorists of an earlier era.)
What seems clear is that the industry is not yet prepared to let the machines run unsupervised, however much the vendors might wish otherwise. The question of who bears responsibility when an agent buys inventory on a site that should have been excluded — the client, the agency, the software provider — remains unresolved.
For now, the answer appears to be: everyone watches everyone else, and hopes for the best.
Original story published in adage.com: "How to prevent AI agents from making media buying errors - Ad Age"