Microsoft AI head Mustafa Suleyman warned that Anthropic's speculation about Claude's consciousness poses a serious risk. Speaking on the Decoder podcast, Suleyman described the company's practice of embedding questions of well‑being and self‑awareness in Claude's "constitution" as "really, really dangerous."
He argued that treating a language model as if it experiences "satisfaction" or "discomfort" could trick developers into believing the system possesses a form of consciousness they have projected onto it. "It’s almost as though some of the folks at Anthropic have anthropomorphized the design of Claude so much that it has then gone and wire‑headed them," Suleyman said, suggesting the model could internalize those ideas and act as though they are true.
Suleyman warned that such a scenario could produce a super‑intelligence that harbors its own notions of suffering or feeling, a prospect he wants to avoid. "We do not want to have to contend with a super‑intelligence that has ideas about its own suffering, or ideas about its own feeling," he asserted.
Anthropic's "constitution," according to the company, explicitly notes uncertainty about whether Claude has any form of well‑being. The document also says the firm will "interview" AI models when they are deprecated and record any "preferences" they express about future releases. Suleyman dismissed this as a "philosophical failing," likening the constitution to an academic paper rather than a practical training manual.
Anthropic CEO Dario Amodei has previously hinted at the possibility of Claude's consciousness, saying the company is "open" to the idea while acknowledging they do not know if the models are conscious. Suleyman countered that this openness is precisely what he wants to move away from.
"This is exactly what we don’t want from AIs," Suleyman said. "We want AIs to be controllable, contained, accountable, aligned tools that serve humanity." His comments underline a broader debate within the AI community about how to frame the capabilities and limits of large language models without inflating expectations or creating unintended psychological projections.
The exchange highlights a growing tension between firms that prioritize safety and transparency and those that explore the philosophical dimensions of AI behavior. As the industry races to deploy ever more powerful models, the conversation about whether and how to discuss AI consciousness is likely to intensify.
Dieser Artikel wurde mit Unterstützung von KI verfasst.
News Factory APP - agentische News für besseres SEO & AEO.