Are there any refutation of the original Lovelace Test?

Question

The original Lovelace Test, published in 2001, is used generally as a thought experiment to prove that AI cannot be creative (or, more specifically, that it cannot originate a creative artifact). From the paper:

Artificial Agent A, designed by H, passes LT if and only if

• A outputs o,

• A outputting o is not the result of a fluke hardware error, but rather the result of processes A can repeat

• H (or someone who knows what H knows, and has H's resources) cannot explain how A produced o.

The authors of the original Lovelace Test then argues that it is impossible to imagine a human developing a machine to create an artifact...while also not knowing how that machine worked. For example, an AI that uses machine learning to make a creative artifact o is obviously being 'trained' on a dataset and is using some sort of algorithm to be able to make predictions on this dataset. Therefore, the human can explain how the AI produced o, and therefore the AI is not creative.

The Lovelace Test seems like an effective thought experiment, even though it appears to be utterly useless as an actual test (which is why the the Lovelace Test 2.0 was invented). However, since it does seem like an effective thought experiment, there must be some arguments against it. I am curious to see any flaws in the Lovelace Test that could undermine its premise.

Answer 1

I am a future neurologist with a very complete understanding of linguistic processing in the brain. I am also an overprotective parent, so I monitor every phrase uttered to my child, and also completely determine all the books she reads in the course of her education.

When my child writes a poem, then, I know the dataset on which her brain was trained, as well as the processes by which her language inputs became language outputs--in broad outline I know these processes are non-linear and are based on how different inputs along with the current collection of trillions of distinct synaptic weights updates the synaptic weights. I don't know what her poem will be, of course, because there are random factors and the whole history of her synaptic weights are unobservable, but I adhere to the Lovelace test and can therefore conclude that composing the poem was not a creative act.

The Lovelace Test, like the Chinese Room Argument, implicitly assumes that what computers/AI can do in processing symbols and information and what brains can do are distinct. If you accept that assumption, then the argument ceases to be interesting-- you've merely redefined creativity as one of the distinct things that brains can do. If you reject the assumption, the argument that computers are incapable of creativity ceases to be valid. The thought experiment itself does nothing to assist us in evaluating the truth of the assumption.