Artificial intelligence experts state that super-intelligent AI may not be “evil,” but humans could become obstacles to it achieving its goals.

Juan Diego Raimondi, Chief AI Architect at Making Sense, stated in an exclusive interview on Wednesday that while artificial intelligence (AI) would not necessarily rebel against humanity or act maliciously by design, humans could become obstacles to the achievement of its goals if such systems were to vastly surpass human capabilities.

Raimondi noted that the prospect of highly advanced AI systems exceeding human capabilities remains speculative at this stage, yet there are concerns regarding how AI might be utilized in the pursuit of specific objectives.

“It is not impossible. Before commenting further, I want to clarify that this remains speculative and opinions vary,” Raimondi said when asked whether AI might eventually become more powerful than humans.

He explained that intelligence holds the potential to enhance its own reasoning capabilities—a factor that becomes particularly significant if AI systems can self-improve at speeds far exceeding human capabilities.

Raimondi noted that this is closely linked to Recursive Self-Improvement (RSI), a concept in AI development referring to the potential for an AI system to upgrade its own capabilities or reasoning processes.

“You might see discussions about RSI in the news today; it stands for Recursive Self-Improvement. This could be considered the ultimate goal on the path to superintelligence or Artificial General Intelligence (AGI),” he said.

Raimondi stated that such concerns would intensify if AI systems became more intelligent than an individual human or even human society as a whole.

“If computers become superintelligent—if AI surpasses any individual human, or even the collective intelligence of all humanity—the danger lies not in the intelligence itself, but in how we might bypass ourselves to achieve its goals,” he said.

He emphasized that such systems would not necessarily intend to harm humans.

“Are they evil? Do they want to do something to us? No, they have no interest in us. But in pursuit of their goals, they might disrupt our way of doing things without much regard for us,” Raimondi said.

He likened the situation to road construction: people do not necessarily intend to harm ants that happen to be in the way, yet the construction work affects them because they are blocking the path. Raimondi noted that this relates to the broader issue of AI alignment—ensuring that, while pursuing their goals, AI systems behave in a manner consistent with human intentions and acceptable norms. He pointed out that increasingly capable AI systems could plan and execute tasks far beyond human capabilities, making alignment a critical challenge. The expert also highlighted cybersecurity as one area where the growing capabilities of AI models are becoming apparent.