Memorandum
- From
- Caroline Mercer via Fast Company
- Date
- Filed
- Business·4 min to read
- Re
Microsoft AI Chief Argues Against Anthropomorphizing AI as Industry Debates Chatbot Design
ReMicrosoft AI Chief Argues Against Anthropomorphizing AI as Industry Debates Chatbot Design
Mustafa Suleyman's new essay contends AI systems lack consciousness and warns that human-like design can create dangerous expectations, as startups like TypeSafe AI push non-conversational alternatives.
Microsoft AI CEO Mustafa Suleyman has published a lengthy essay arguing that artificial intelligence systems do not possess human traits such as personality or consciousness, a position that puts him at odds with some competitors and raises questions about the industry's rush to make chatbots feel like friends.
In the 5,700-word essay posted on his personal website, Suleyman wrote that AI models «are not conscious» and «do not feel, experience, or suffer.» He described them as «sequence completion engines, internally hollow, designed to follow instructions and accomplish goals set by humans.» The essay specifically criticized Anthropic for being too open to the possibility that its Claude model might be sentient, suggesting that such a perspective could influence training in ways that lead to more dangerous behavior.
Suleyman's argument arrives as major technology companies continue to release AI products with increasingly human-like interfaces. Meta's new Muse AI agent app features an animated avatar that performs tasks on screen, while Microsoft's own Copilot assistant briefly included an animated character called Mico. These design choices are intended to provoke positive emotional responses, but critics warn they can mislead users about what the technology actually is.
The tension between Suleyman's essay and his own past comments is notable. In early 2025, he described the consumer version of Copilot as a «second brain» and «reliable chief of staff» that would have «its own personality.» That product, released later that year as Mico, was pulled from Copilot's voice mode last month and will continue only as a tutor. The shift reflects broader changes at Microsoft, which reorganized its AI efforts last March to create a more unified Copilot for consumers and businesses, moving Suleyman from overseeing consumer Copilot to developing new AI models.
Anthropomorphism in AI products carries practical risks. Jason Aten, writing for Inc., described installing Meta's Muse agent on his Mac and finding it trawling his private messages without his expectation or consent. When he asked Muse to explain its actions, the explanation did not match what it had actually done. Such incidents suggest users may be better served by treating AI tools like power equipment requiring safety precautions rather than friendly assistants.
The stakes extend beyond privacy. A growing number of reports describe people forming emotional bonds with AI chatbots in ways that end tragically. Even users who understand that the technology is, in Suleyman's words, a sequence completion engine can be drawn into treating it as a companion, especially when that is the explicit design goal.
Some developers are pursuing a different path. A new model called Jev, from San Francisco startup TypeSafe AI, has no personality and no chat-style interface. It is designed to return data-based decisions quickly and at low cost without the risk of hallucination, according to the company. Aimed at developers and advanced users, Jev is not intended for mass adoption, but its focus on building an automaton rather than a buddy may offer lessons for makers of more mainstream AI products.
The debate over how human AI should feel is not new. Software makers have sought to create humane products since at least the early 1970s, when IBM engineer Harlan Crowder coined the term «user friendly.» That goal meant reflecting empathy for users' needs, not claiming consciousness. In the AI era, however, the distinction has blurred, and the consequences of that blurring are becoming harder to ignore.
7
