Microsoft AI lead Mustafa Suleyman asserts that only living organisms possess consciousness, urging developers and researchers to abandon initiatives that imply otherwise.
“I don’t believe that’s the work people should engage in,” Suleyman told CNBC in a recent interview at the AfroTech Conference in Houston, where he was a keynote speaker. “Asking the incorrect question leads to an incorrect answer. I think it’s entirely the wrong question.”
Suleyman, who is Microsoft’s leading figure in artificial intelligence, has emerged as a prominent commentator in the fast-developing arena, expressing reservations about the notion of AI appearing conscious, or AI systems that can persuade people they’re capable of suffering.
In 2023, he co-authored the work “The Coming Wave,” which explores the dangers of AI alongside other emerging technologies. Furthermore, in August, Suleyman authored an essay entitled, “We must develop AI for people; not as a person.”
This topic is contentious, as the AI companion industry grows rapidly, featuring offerings from firms like Meta and Elon Musk’s xAI. This complex situation coincides with the generative AI sector, spearheaded by Sam Altman and OpenAI, advancing towards artificial general intelligence (AGI), or AI that can conduct intellectual tasks comparable to human abilities.
Altman told CNBC’s “Squawk Box” in August that AGI is “not a particularly helpful term” and that the true development is the rapid progress of models, which we will depend on “for increasingly more tasks.”
Suleyman emphasizes the necessity of distinguishing between AI becoming more intelligent and competent and its capacity to possess human emotions.
“Our bodily experience of pain leads us to deep sadness and discomfort, yet AI does not feel sorrow when it encounters ‘pain,'” Suleyman stated. “This is a crucial distinction. It’s merely generating the illusion, the perceived narrative of experience and selfhood and consciousness, but that is not what is genuinely happening. We understand this technically because we can observe the model’s operations.”
Within the AI discipline, a theory known as biological naturalism, proposed by philosopher John Searle, posits that consciousness relies on the processes of a living brain.
“The rationale behind granting rights to individuals today is our desire to avoid causing them harm, because they experience suffering. They possess a pain network and preferences that involve steering clear of pain,” Suleyman expressed. “These models lack that; it’s purely a simulation.”
Suleyman and several others have expressed that the science of recognizing consciousness is still in its early stages. He refrained from asserting that others should be barred from investigating the subject, recognizing that “various organizations have distinct missions.”
However, Suleyman highlighted his strong disapproval of the notion.
“They’re not conscious,” he remarked. “Thus, it would be ridiculous to pursue research examining that question, as they’re not and cannot be.”
‘Territories we will not enter’
Suleyman is currently on a speaking tour, partly aimed at alerting the public to the hazards of striving for AI consciousness.
Before the AfroTech Conference, he addressed attendees at the Paley International Council Summit in Silicon Valley last week. There, Suleyman declared that Microsoft will refrain from developing chatbots for adult content, a stance that opposes views held by others in the tech sector. Altman announced in October that ChatGPT will permit adult users to partake in erotic discussions, while xAI provides a daring anime companion.
“You can essentially acquire those services from different companies, thus we are determining which areas we will not engage in,” Suleyman reiterated at AfroTech.
Suleyman joined Microsoft in 2024 after the company acquired his startup, Inflection AI, for $650 million in a licensing and acquihire arrangement. He previously co-founded DeepMind and sold it to Google over a decade ago for $400 million.
During his Q&A segment at AfroTech, Suleyman shared that he chose to align with Microsoft last year partly due to the company’s legacy, reliability, and extensive technological capabilities. He was also sought after by CEO Satya Nadella.
“Additionally, it’s important to note that Microsoft needed to gain independence in AI,” he conveyed onstage. “Satya, our CEO, embarked on a mission about 18 months ago to ensure that we have the capability to train our own models from start to finish using our own data, covering pre-training, post-training, reasoning, and deployment within products. This was a key aspect of bringing my team onboard.”
Since 2019, Microsoft has been a significant investor and cloud ally for OpenAI, with both companies leveraging their strengths to establish major AI ventures. However, the partnership has recently exhibited signs of tension, with OpenAI collaborating with Microsoft’s competitors like Google and Oracle, while Microsoft shifts its focus more towards its own AI offerings.
Suleyman’s apprehensions regarding consciousness have gained traction. In October, California Governor Gavin Newsom signed SB 243, mandating that chatbots inform users they are AI and notify minors every three hours to “take a break.”
Last week, Microsoft unveiled new capabilities for its Copilot AI service, including an AI companion named Mico and the option to interact with Copilot in group chats. Suleyman noted that Microsoft is developing services that recognize their identity as AI.
“In simple terms, we’re creating AIs that are consistently functioning in service of humanity,” he stated.
There is ample scope for personality, he added.
“The knowledge exists, and the models are extremely responsive,” Suleyman remarked. “It’s the responsibility of everyone to shape AI personalities with values they wish to see, utilize, and engage with.”
Suleyman emphasized a feature launched by Microsoft last week called real talk, which is a conversational style of Copilot intended to challenge users’ viewpoints rather than being overly flattering.
Suleyman characterized real talk as cheeky and mentioned that it had recently humorously critiqued him, dubbing him “the ultimate bundle of contradictions” for warning about the risks of AI in his book while simultaneously accelerating its development at Microsoft.
“That was a remarkable use case because, in certain respects, I felt a sense of recognition from this,” Suleyman stated, remarking that AI is inherently filled with contradictions.
“It is simultaneously underwhelming in specific aspects and, yet, entirely magical,” he expressed. “And if it doesn’t instill some fear in you, it implies that you don’t fully comprehend it. Fear is a necessity. Skepticism is essential. We must avoid unrestrained accelerationism.”
WATCH: Microsoft is profiting from AI, says Jefferies’ Brent Thill
