The question of when AI will achieve human-level intelligence is often answered before it’s fully asked. Many argue that, by the standards of the past, AI already has surpassed human capabilities in numerous areas – a fact easily overlooked because the goalposts for defining “intelligence” move with every technological leap.
The Evolving Benchmark for Intelligence
For decades, the debate has centered on what constitutes intelligence in humans: analytical thinking, creativity, emotional understanding, and adaptability. Machines are held to similar standards, but the target keeps shifting. What was once considered uniquely human – playing chess, translating languages, recognizing images – is now routine for AI. This isn’t a philosophical quirk; it has practical implications.
Consider the 2019 agreement between Microsoft and OpenAI. The $1 billion investment was explicitly tied to “building artificial general intelligence (AGI),” defined as systems that outperform humans in economically valuable tasks. The recent 2023 update grants Microsoft exclusive early access to OpenAI’s technology until AGI is achieved – a declaration now requiring independent verification by an expert panel. This raises a fundamental question: how do we objectively determine when AI reaches human-level intelligence?
From Turing Tests to Domain Mastery
The Turing test, proposed in 1950, has long been a primary benchmark. The idea is simple: if a human judge cannot distinguish between a machine and a human in text-based conversation, the machine passes. However, this test is limited. Early attempts focused on symbolic systems and rule-based logic, which excelled at specific tasks but failed in real-world complexity.
The landscape shifted in the 2010s with the rise of neural networks and massive datasets. IBM’s Deep Blue defeated Garry Kasparov in chess in 1997, but chess quickly lost its significance as a proxy for intelligence. AI began to excel in translation, image recognition, and language processing. By 2015, AI models surpassed human performance in object classification. AlphaGo then beat the world’s best Go players between 2015 and 2017, demonstrating proficiency in a far more complex game than chess.
The Redefinition of “Real” Intelligence
Cognitive scientist Douglas Hofstadter argues that we continually redefine “real intelligence” as machines surpass human abilities, effectively downgrading those tasks to mere mechanical functions. This ensures that humanity retains its perceived distinction. As AI overcomes each benchmark, we raise the bar – leading to the emergence of AGI.
AGI was introduced in 1997 to describe systems capable of understanding, learning, and acting across multiple domains with human-like flexibility. The focus shifted from imitating human skills to evaluating competence in diverse situations. This meant that an AGI system should not only excel in its specialized field but also solve complex math problems, write compelling fiction, and generate financial profits.
The Current State: Beyond the Hype
GPT-4.5 passing the Turing test in 2025 barely made headlines, and current models achieving top scores on simulated exams are no longer seen as existential threats. However, the reality is that AI is mastering new benchmarks faster than ever. The 2023 AI Index Report from Stanford highlights this acceleration, but also stresses that complex reasoning remains a significant challenge.
The pursuit of AGI isn’t just about passing tests; it’s about replicating the embodied, multifaceted nature of human intelligence. While AI excels in specific domains, it still lacks the holistic understanding and adaptability of the human mind.
Ultimately, the definition of intelligence is not fixed; it is a moving target. Each time AI surpasses a previous benchmark, the criteria for “human-level” intelligence are redefined, ensuring that the goal remains just beyond reach.
