For decades, the question of whether machines can truly “think” has sparked debate among scientists, philosophers, and engineers. The original Turing Test, proposed in 1950 by British mathematician Alan Turing, was designed to evaluate a machine’s ability to imitate human conversation convincingly. If a human judge could not distinguish between machine and person, the machine was said to exhibit intelligence. While historically influential, the Turing Test has been criticized for being too focused on imitation rather than true understanding.

In the second podcast conversation in this series on Mind Matters with host Robert J. Marks, Dr. Georgios Mappouras discussed his paper The Turing Test 2.0: The General Intelligence Threshold, which proposes a new framework for evaluating artificial intelligence. His approach shifts the focus away from surface-level mimicry and instead emphasizes creativity, learning, and the ability to derive new insights from given information.

The three rules of Turing Test 2.0

Mappouras outlines three conditions that must be met for a valid Turing Test 2.0 evaluation:

1. Functional Information: The AI system must be given a set of functional knowledge — capabilities or skills it can already perform. For instance, it may know how to generate images or search arrays in a dataset.
2. Non-Functional Information: The system must also receive information it has not yet learned how to use. This “raw” or unstructured data acts as the spark for innovation, challenging the AI to derive new insights from unfamiliar content.
3. No External Help: Crucially, the system cannot rely on additional outside training or human intervention. The test must reveal whether the AI can independently create new functionality from the information it already has.

Together, these rules form the foundation for assessing whether an AI can demonstrate genuine understanding rather than simply parroting back memorized patterns.

Flash of genius vs. pattern recognition

Marks suggested that central idea in Mappouras’s proposal is the notion of the “flash of genius.” Human creativity often involves sudden insight — moments when solutions appear seemingly out of nowhere. Nikola Tesla famously envisioned the design of the brushless motor in a flash while walking on a beach. Mathematician Carl Friedrich Gauss once awoke with a problem’s solution already formed in his mind. These moments reveal how humans can transform unstructured information into novel insights.

AI, however, struggles with this leap. While it can analyze massive datasets and generate plausible outputs, it rarely demonstrates the ability to derive new knowledge from scratch. Mappouras illustrates this through practical examples: when asked to draw a clock showing 6:30, many AI models fail, even though they can describe in words where the hands should be placed. This disconnect between description and application shows a lack of true understanding.

The hexagonal stop sign problem

Another revealing test involved generating images of a stop sign. Most AI systems easily produce the familiar red octagon with white letters. But when asked to render a stop sign in a hexagonal shape, the models stumbled. Even if they eventually succeeded, they could not consistently reproduce the result when asked again later. For Mappouras, this inconsistency highlights a critical weakness: AI often memorizes patterns rather than internalizing rules it can reuse flexibly.

This distinction matters. True intelligence, Mappouras argues, requires the ability to extract general principles from specific examples and then apply them across new contexts. Without that ability, AI remains limited to clever pattern-matching.

Comparisons to other tests of intelligence

Mappouras also contrasts his approach with other attempts to measure AI intelligence, such as the benchmark proposed by François Chollet, a researcher at Google. Chollet’s framework resembles an IQ test for machines, requiring them to detect patterns and predict outcomes. While valuable, Mappouras critiques these tests as being too narrow and too dependent on human standards. Intelligence, he suggests, should not be defined only by tasks humans excel at but should reflect the capacity to generate new knowledge from available information.

The debate recalls the early days of computing, when humans were called “computers” because they could perform high-precision arithmetic more accurately than machines. Over time, machines surpassed humans at calculation, but that alone did not make them intelligent. Similarly, passing a narrow benchmark today does not necessarily mean AI has reached general intelligence.

Toward a true measure of understanding

At the heart of Turing Test 2.0 is a simple but profound idea: intelligence is the ability to extract new knowledge from existing information and to apply it consistently across time and context. This mirrors the way good teachers test students — not by asking them to repeat memorized answers, but by requiring them to apply concepts in unfamiliar situations.

For now, Mappouras acknowledges that no AI system has passed his test. Current models excel at recognition and reproduction but fall short of the creative leaps that define human thought. Whether this gap can ever be bridged remains uncertain. Optimists like futurist Ray Kurzweil believe the Singularity, a point where machines surpass human intelligence, is just around the corner. Skeptics note that this prediction has been “just around the corner” for decades.

What is clear is that measuring intelligence in machines demands more than imitation. It requires probing for genuine understanding — the kind of flexible, creative reasoning that turns raw information into insight. The Turing Test 2.0 represents a promising step in that direction, challenging us to rethink not only what machines can do, but what it truly means to be intelligent.

Here’s Part 1: Measuring machine intelligence using Turing Test 2.0. Georgios Mappouras’s updated approach asks whether machines can go beyond imitation to produce new knowledge. Mappouras’s General Intelligence Threshold offers a test question: Can the system generate insights that were not directly programmed into it?

Additional Resources

• Listen to part 1 of this conversation: Turing Test 2.0: A Better Way to Test Machine Intelligence?
• Georgios Mappouras: “Turing Test 2.0: The General Intelligence Threshold”
• Robert J. Marks: “Is AI Truly Creative? Here Is the Ultimate Test” 
• Alan Turing: “Computing Machinery and Intelligence”
• François Chollet: “On the Measure of Intelligence”
• Handbook of Fourier Analysis & Its Applications by Robert J. Marks

Podcast Transcript Download