Anthropic unveils 'dreaming' feature for AI agents, sparking debate over anthropomorphism in tech branding
At a developer conference in San Francisco, the company introduced the update as part of its agent infrastructure, drawing comparisons to Philip K. Dick's science fiction while facing scrutiny from ethicists.
Anthropic has announced a new capability for its AI agents at a developer conference in San Francisco, describing the function as "dreaming". This feature enables agents to analyse their activity logs and past transcripts to identify patterns and refine their performance between sessions. The update is part of a broader infrastructure designed to help users manage and deploy tools that automate software processes, allowing agents to glean insights from recent activities to improve future interactions.
The naming convention has immediately drawn criticism regarding the industry trend of anthropomorphising machine processes. A contributor to WIRED has explicitly called for AI companies to stop using terms such as "dreaming", "memory", and "reasoning" to describe algorithmic functions. The argument posits that these labels mimic human cognitive functions and blur the distinction between human capabilities and machine operations, a practice that has intensified since the chatbot revolution began in 2022.
Anthropic's approach to this branding appears intentional, extending beyond mere marketing strategies to the company's internal governance. The firm's internal constitution states that it discusses its model, Claude, in terms normally reserved for humans, such as "virtue" and "wisdom". This stance is grounded in the expectation that the model's reasoning will draw on human concepts by default, given the role of human text in its training. To support this framework, Anthropic employs a resident philosopher to help interpret the bot's values.
The decision to name the feature "dreaming" echoes themes from Philip K. Dick's novel Do Androids Dream of Electric Sheep?, which explores the qualities that separate humans from powerful machines. While current generative AI tools do not approach the capabilities depicted in the book, the terminology suggests a desire to endow the software with a deep inner life. This includes descriptions of agents having "memories" regarding user preferences, distinct from standard computer RAM, and spending time "thinking" before responding, a tactic also adopted by OpenAI in 2024.
Ethicists and researchers warn that this linguistic strategy carries significant risks. Research published in the AI & Ethics journal indicates that anthropomorphism is a fallacy that can distort moral judgments about AI systems. By using human-centric language, users may project qualities onto machines that do not exist, potentially affecting assessments of responsibility and trust. Critics argue that adopting more distanced language is necessary to prevent users from overly trusting tools that lack genuine human cognition.
Despite the controversy, Anthropic maintains that the "dreaming" process refines memory between sessions and pulls shared learnings across agents to keep them up-to-date. The company views encouraging the embrace of humanlike qualities as actively desirable for the role of human text in its training. However, the debate highlights a growing tension between the desire for relatable AI interfaces and the need for clear, factual communication about the limitations of these technologies.
