Will the Turing Test Detect Synthetic Intelligence?

We know that synthetic intelligence is not artificial intelligence. The latter is an imitation whereas the former is authentic. Further, we know that agency is an emergent expression of intelligence, synthetic or organic. With those points established, we can begin to explore more concrete areas. Thus, where our previous discussions focused on speculative (what) questions, I would like to concentrate on something more practical this time. Specifically, can we use the Turing Test to detect synthetic intelligence?

Previously, I suggested that measuring synthetic intelligence would be desirable. Stated specifically, if we’re going to create a synthetic intelligence we ought to have the ability to detect whether our work was successful and to what degree we are successful. As well, we want to know when a synthetic entity develops intelligence so that we interact with it appropriately. We do this now, implicitly, with our ourselves and with other humans. We also have several measures that we believe ought to function adequately for artificial intelligence. But, what about synthetic intelligence?

Human Intelligence

When examining human intelligence, we have instruments such as observation, introspection, developmental tests, and of course g factor. Further, we have a deep understanding of how human intelligence develops from birth until maturity. More precisely, we know what a human intelligence ought to be capable of certain developmental milestones. Whilst we are unique individuals, when it comes to possessing intelligence and detecting such, we humans seem to be rather similar. Thus, apart from a specific detective instrument, we ascribe intelligence to another human when we perceive them to act as we do. This seems rational insofar as we understand agency to be an embodiment of intelligence. Further, when we need to empirically measure intelligence we have those instruments too.

Artificial Intelligence

Broadly speaking, the same is true for artificial intelligence. That is, we rely on a combination of observation and empirical testing. Here, the Turing Test is the de facto standard for testing for the presence of intelligence in artificial agents. Before we consider if the Turing Test is a viable measure for synthetic intelligence we ought to ensure that we grasp the Turing Test correctly.

The Turing Test is not an interviewer measuring or detecting intelligence by posing questions to an artificial intelligence. Rather, the Turing Test detection mechanism relies on an interviewer posing questions blindly to both an artificial intelligence and a human. The measurable output then is the degree of similarity between responses to the inquiries. We must take note here that there is an implicit presupposition that there can only be a difference in degree between artificial intelligence and human, not a difference in kind. The distinction is critical.

As well, there is a salient point in the power of a test being bound by the level of intelligence creating that test. Furthermore, the Turing Test (or any detective control for artificial intelligence for that matter) can only serve to detect intelligence that is perceivable by the intelligence creating and administering the instrument to begin with. This creates quite a conundrum for us and for apply the Turing Test to synthetic intelligence.

Synthetic Intelligence

Accordingly, I suggest that the Turing Test is not a good instrument for detecting synthetic intelligence. I think that the Turing Test is insufficiently equipped to interact with synthetic intelligence because synthetic intelligence will certainly differ from human intelligence in kind, not just degree. What’s more, synthetic intelligence may not manifest in a modality we are accustomed to considering as harboring intelligence. Plants are a great example.

Therefore, I offer that the appropriate instrumentation for measuring or detecting synthetic intelligence ought to be situated in agency. That seems like meaningless words; I get it. However, even a plant intelligence embodies recognizable aspects of intelligence in the same world we inhabit. Additionally, synthetic intelligence may embody aspects that we do not yet recognize. The specific aspect is less relevant though than detecting that an aspect is present. Thus, the synthetic intelligence test should align, for now, to that degree and kind only. As an example, I don’t need to know the precise wavelength of light shining outside of my window to know that there is a type of light present. I can simply detect that some form of light is there. Yet, this brings up a more interesting line of inquiry for us.

Okay, so what if we do measure and detect that a synthetic intelligence is present? Can we trust that intelligence and can that intelligence trust us? Check back in two weeks for my idea as to how we might be able to demonstrate both!

By Dr. Jason M. Pittman, Sc.D