The Peril of Humanizing AI: Why Naming Features After Human Processes Undermines Trust

At the recent Anthropic developer conference, the company unveiled “dreaming,” a new capability that lets AI agents sift through their stored “memories” to generate novel insights. This announcement joins a growing roster of features whose names echo human activities—“remembering,” “planning,” even “creativity.” While the technology itself is impressive, the linguistic choice invites closer scrutiny. The term evokes the human capacity to imagine, yet it rests on statistical pattern matching rather than conscious experience. Naming AI functions after human processes anthropomorphizes the system, suggesting agency and consciousness that the underlying algorithms do not possess. Such framing can inflate user expectations, blur the line between tool and partner, and ultimately erode trust when the gap between promise and performance becomes evident. It also obscures the mechanical nature of the technology, making it harder for developers and regulators to assess real capabilities and risks. Throughout the history of software, branding has shaped perception. The rise of “smart” assistants, “cognitive” services, and now “dreaming” mirrors a broader hype cycle that equates evocative language with breakthrough performance. This pattern dates back to early expert systems that marketed themselves as “learning” machines, a claim that often outstripped technical reality. Looking ahead, the industry would benefit from greater semantic honesty. Transparent naming would align public understanding with technical facts, facilitate clearer policy discussions, and reinforce the credibility of AI deployments. As AI systems become more embedded in critical domains, the cost of misleading terminology may prove far higher than the modest effort required to rename features in line with their actual function.