In recent years, artificial intelligence (AI) has achieved remarkable milestones, consistently surpassing benchmarks that were once considered definitive tests of humanlike intelligence. This rapid progression, however, has led to a continual redefinition of what it means for AI to be “intelligent” in a human sense. As AI systems grow more capable, the standards by which we judge their intelligence keep evolving, raising profound questions about the nature of intelligence itself and how we recognize it in machines.
The question of when AI might achieve humanlike intelligence is no longer straightforward. A conversation with a friend revealed a compelling perspective: AI has arguably already reached that milestone. From the standpoint of 1995, today’s AI would likely appear not just humanlike but even superhuman in its capabilities. This shifting perspective highlights a key issue—the “goalposts” for what constitutes humanlike intelligence seem to move every time AI technology advances.
Defining intelligence has always been a challenge, whether in humans or machines. Human intelligence is multifaceted, encompassing analytical, creative, and emotional dimensions, as well as a balance between following instructions and exercising autonomy. These complexities have made it difficult to pin down a universal definition. Similarly, AI researchers have struggled to set fixed standards, as machines evolve and perform increasingly complex tasks that were once thought to require uniquely human cognition.
This evolving landscape is not just philosophical—it has concrete implications for the development and deployment of AI technologies. For instance, when Microsoft invested $1 billion in OpenAI in 2019, the partnership was framed around the pursuit of artificial general intelligence (AGI). OpenAI’s charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” This ambitious goal underscores the stakes involved in defining and achieving human-level machine intelligence.
Recently, in October 2025, Microsoft and OpenAI updated their partnership agreement. Microsoft retains special early access to OpenAI’s technologies and exclusive rights to use them in products until OpenAI declares it has reached AGI. Importantly, any such declaration will now be subject to independent verification by an expert panel. This development raises a critical question: How will this panel decide when AI has truly achieved human-level intelligence?
Since the mid-20th century, the primary benchmark for machine intelligence has been the Turing test, proposed by computer scientist Alan Turing in 1950. The test involves a human judge communicating via text with both a human and a machine, without knowing which is which. If the judge cannot reliably distinguish the machine from the human, the machine is said to have passed the test. While elegant in concept, the Turing test has long been debated and critiqued for its limitations in capturing the full scope of intelligence.
The evolution of AI over the decades reflects changing approaches to machine intelligence. Early AI systems in the latter half of the 20th century relied on symbolic reasoning and explicit rules to mimic human problem-solving. These “expert systems” could solve puzzles and play games within narrow domains but were inadequate for dealing with the complexities of the real world. They functioned well under constrained conditions but lacked general adaptability.
The modern AI era began in earnest in the 2010s, fueled by advances in neural networks and the availability of large datasets. This shift enabled machines to learn from data patterns rather than relying solely on pre-programmed rules. Landmark achievements included IBM’s Deep Blue defeating chess grandmaster Garry Kasparov in 1997—a watershed moment that diminished chess’s role as a proxy for human intelligence. Subsequently, AI models excelled in language translation, image recognition, and reasoning tasks. In 2015, a vision model surpassed human performance in object classification, and by the late 2010s, AI had made strides in language understanding and complex reasoning. Between 2015 and 2017, AlphaGo, an AI designed to play the complex game of Go, defeated the world’s top players, further redefining expectations.
Cognitive scientist Douglas Hofstadter has argued that each time AI achieves a capability once considered uniquely human, society tends to downgrade that ability to a mere mechanical skill. This psychological defense preserves the notion of human uniqueness by continually redrawing the boundaries of “real intelligence.” In other words, as AI passes new milestones, the definition of intelligence shifts upward, maintaining a moving target.
This dynamic is why the concept of AGI emerged—to describe machines capable of understanding, learning, and acting flexibly across many domains, akin to human cognition. Coined