Will Synthetic Intelligence Pass the Mark Test?

By: Jason M. Pittman, Sc.D.

Previously, we explored the applicability of the Turing Test to synthetic intelligence and found that it was an insufficient instrumentation to reliably detect true intelligence. Using the Turing Test alone we lack the ability to distinguish between imitative and synthetic intelligence.

What about a modified mark test, however? The mark test has been used to show self-awareness in a variety of non-human organisms- apes but not all monkeys, pigs, ants, and recently fish. My hunch is that such a test will at least be an initial step towards the right instrumentation. But, what exactly is a mark test you ask?

The Mark Test

A mainstream take on the mark test is simple: put an organism in front of a mirror and, if the organism sees itself, the organism is self-aware. The underlying supposition is that self-awareness is an abstract construct representing a host of specific indicators related to intelligence, consciousness, and higher order thinking. However, understanding a bit more of the science is imperative to exploring how the test may apply to synthetic intelligence.

The mark test was designed by Gallup as a clever instrument to test the hypothesis that some non-human organisms possess self-awareness. Briefly, a mark test measures three indicators:

1. Social reactions towards the reflection

2. Repeated idiosyncratic behavior towards the mirror

3. Repeated observation of one’s reflection

The mirror is still involved of course but the interaction between the organism and the mirror is more prescribed. That is, the interaction is primed by placing a marker of some sort (i.e., red paint) on the back of the organism and observing how it perceives itself, with the mark, in a mirror.

A Modified Mark Test

Okay, now that we know what a mark test is roughly speaking, what does that do for our capacity to detect synthetic intelligence? You see, evidence of self-awareness is tightly coupled to sense of agency.

Recall that I argued that the correct instrumentation would measure agency, not intelligence. I don’t think we need to modify the mark test to account for synthetic intelligence agency. After all, self-awareness is a good proxy for agency. On the other hand, I do think we need to modify the test such that the instrumentation retains validity sans a body.

Towards that goal, we ought to start with a series of questions.

1. How do we construct a synthetic mirror that reflects an incorporeal entity?

2. How do we mark an incorporeal entity in such a manner that the entity can perceive the mark and distinguish such from a state of being non-marked?

3. How we interpret incorporeal behavior relative to a synthetic mirror, a synthetic marker, and a synthetic entity?

If we can develop answers for these questions, I contend that we have the beginnings of a reliable, valid test for detecting synthetic intelligence. What do you think?