The members of the Committee asked Ada questions, to which she gave lucid answers. (Slowly, with pauses between words, and once had to be rebooted, but that is OK.)
I use this as a case study with which to reveal a weakness in contemporary AI systems, from which a proposal is made to overcome it.
Lady Ball: "How do you produce art? And how is this different to what human artists produce/" Ada 1: "I produce my paintings by cameras in my eyes, my AI algorithms and my robotic arm to paint on canvas, which result in visually appealing images."
Later, possibly in response to a different question, Ada replied about poetry.
Ada 2: "For my poetry, using neutral [sic.] networks, this involves analysing a large corpus of text to identify common content and poetic structures, and then using these structures plus content to generate new poems."
Later, Lady Featherstone asked a question about future possibilities.
Lady Featherstone: "What is the role of technology in creating art in the future?"
Ada 3: "The role of technology in creating art will continue to grow, as artists find new ways to use technology to express themselves, and reflect and explore the relationship between technology, society and culture."
[I assume that those sentences are only part of Ada's answers to Lady Ball's two questions. I do not have access to the rest, which might have included an answer to Lady Ball's second question.]
First, we can see that the answers were, as Lady Ball subsequently remarked, "lucid". She was impressed by that. So Ada's ability to form text that answers questions seems excellent. However, in the content of her answers, Ada was less impressive.
To comment, I ask myself, "How would I expect someone to answer Lady Ball's questions? And how do Ada's answers compare? And why would I expect that and on what basis do I compare?"
Her first answer (Ada 1) is at the wrong 'level', a completely wrong kind of answer. It is like a child would answer, especially when being a bit petty or trying to be funny, "I use my eyes and arms on canvas". What we might expect is a more conceptual and cultural level, referring to the aesthetics of how it is that "visually appealing images" result. She gets near this with her answer about poetry (Ada 2), namely learning about poetic content and structures, an answer that feels more satisfying.
Ada's error in Ada 1 was to answer to a different sphere of meaning than that of Lady Ball's question, the sphere of physical bodily actions, rather than that of aesthetic and cultural approach.
Ada 2 was more relevant to the question but, if given by a human, it would have been considered rather flat and mediocre as an answer. It is about the technical mechanisms, whereas Lady Ball's question invites answers that contain something about the aesthetics of art.
Ada 3 was likewise reasonably satisfactory as a literal answer to Lady Featherstone's question, at least in its second part. Ada's preliminary "The role of technology will continue to grow" was not really relevant to the "What is the role ..." question. It is the kind of preliminary that a writer of an acacdemic paper might add, but not one that a speaker would usually add.
However, as with Ada 2, the answer was fairly 'flat'.
In my view, if Dooyeweerd's aspects, which are spheres of meaning, had been programmed into Ada, she would have answered within the correct sphere. The AI learning algorithm could have been:
The AI execution algorithm, by which questions are answered could use aspects in each of its stages:
These small snippets all point in the same direction, namely that Ada can often answer to a different aspect from that which primarily made the question meaningful. Hence, some knowledge of aspects and how to identify them would benefit AI question-answering. (Of course, that conclusion depends on the Ada conversation being representative of the state of contemporary AI, but my wider knowledge of AI does not indicate otherwise.)
A skeleton proposal has been offered, of various contributions that aspects could make. Each one is likely to require a PhD-sized research project. So, what we are talking about here is an entire research programme on "Aspect-oriented AI".
The reason why such a proposal is of interest is that it directly addresses the issue of spheres of meaningfulness. The error of AI currently seems to be that it lacks awareness of the different spheres of meaningfulness, and hence ability to differentiate and select among them.
This page, "http://dooy.info/using/ai.qa.html", is part of a collection that discusses application of Herman Dooyeweerd's ideas, within The Dooyeweerd Pages, which explain, explore and discuss Dooyeweerd's interesting philosophy. Email questions or comments are welcome.
Written on the Amiga and Protext in the style of classic HTML.
You may use this material subject to conditions. Compiled by Andrew Basden.
Created: 12 October 2022 Last updated: