Andrew Ng Proposes New ‘Turing AGI’ Test to Cut Through AI Hype

The ongoing debate about artificial general intelligence (AGI) has intensified as experts grapple with defining what AGI truly entails and whether current large language models (LLMs) qualify. In this context, Andrew Ng, co-founder of Coursera and a prominent AI researcher, has proposed a new benchmark called the ‘Turing AGI’ test. This test aims to provide a more practical and rigorous way to evaluate whether AI systems have achieved general intelligence.

Unlike traditional Turing tests that focus on conversational ability, Ng’s ‘Turing AGI’ test involves an AI system and a skilled human professional performing real-world work tasks over multiple days. Both participants have access to a computer equipped with internet connectivity and common software tools such as web browsers and video conferencing platforms like Zoom. The goal is to assess the AI’s capability to handle complex, multi-step tasks that require reasoning, learning, and adaptation in a dynamic environment.

This approach reflects a shift from purely theoretical or linguistic benchmarks toward evaluating AI’s functional performance in practical scenarios. By allowing the AI to use external resources and tools, the test acknowledges the evolving nature of AI systems that increasingly integrate with software ecosystems to augment their capabilities. It also mirrors how human professionals operate in real work settings, making the comparison more relevant and grounded.

Ng’s proposal comes amid widespread skepticism about claims that current LLMs have achieved AGI. Many experts argue that while these models excel at generating coherent text and answering questions, they lack true understanding, long-term planning, and autonomous learning abilities. The ‘Turing AGI’ test could serve as a clearer litmus test to distinguish between advanced narrow AI and genuine general intelligence.

If adopted, this test might influence AI research priorities by encouraging the development of systems that can perform diverse tasks over extended periods, rather than optimizing for short-term benchmarks or single-domain proficiency. It could also impact how AI capabilities are communicated to the public, helping to temper exaggerated expectations and clarify the technology’s actual progress.

Overall, Andrew Ng’s ‘Turing AGI’ test represents a significant step toward establishing more meaningful standards for evaluating AI. By focusing on practical work performance and real-world adaptability, it seeks to cut through the hype and provide a transparent framework for measuring progress toward true artificial general intelligence.