Questions as Objects in AI Object Handling
2023.07.25,08.14 gg.ashbrook
Questions
A few years ago when I first started thinking about what became “object relationship spaces for AI architecture and operating systems,” if memory serves (and it often does not…) one of the first types of objects that I suspected would be important was questions: the ability for AI to recognize questions to answer them, to answer questions, to ask questions about the objects that were being categorized or identified, etc. Though the role of LLM-GPT in handling questions was not something that I predicted at all, and in general the topic of questions mostly fell by the way-size, only being left in the list of object-relationship-categories out of sentimental stubbornness.
Object Skepticism
As a side note, perhaps the best analogy I have found for what ‘object’ means in object-relationship-spaces, is objects from object oriented programming…which I find puzzling since I remain very skeptical of the wisdom of the OOP programming paradigm that, if coincidentally, corresponds with a catastrophic dark-age of software creation (in a bewildering perfect storm of respects that may have nothing inherently to do with OOP…somehow).
Types of Questions
The topic of questions re-emerged with early experiments and versions creating externalizing frameworks in which the AI operates. In short, the initial task and test seemed nicely narrowly defined: What does the system do when asked a math question (in particular the type of math question that raw-generative-ai famously gets wrong)? And what resulted from trying to collect and test a spectrum of questions was the stubborn-re-emergence of quagmires and controversies from both my own experience teaching STEM and language and from the STEM history books I try to learn from: people (H.sapiens-humans) are themselves very messy language generators.
“Math” Question Subcategories:
- Invalid Questions:
- Missing Reference Questions:
- Compound Questions:
- Incomplete / Complete (self-contained) Questions:
- Set STEM Value Questions:
- Set STEM Function Questions:
- Open STEM Function Questions:
- Open Working-Solution Questions:
- Variably Interpretable Questions:
- Telepathy Questions: (arbitrary ‘what is the teacher thinking’ questions)
- Mirage Pastiche Questions: (incoherent garbage questions)
Categories of types of questions from categories of types of systems:
- Logic-Math Tautology Questions
- Statistics/Probability Questions
- Linear Physical STEM Questions (logic + statistics + raw measurements)
- Dynamical & Nonlinear Physical STEM Questions
- Undefined Questions
An overall space of question-spaces?
Some analytical problem question are a clean intersection between:
- STEM
- questions
- deterministic functions
- analysis
- problems
- signals
- interactions
- learning
- etc.
But not all problem-questions are deterministic analytical problems, and not all questions are about well defined problems and tasks. The questions that I was thinking of original for an AI to navigate with were along the lines of:
- Is this input problematic?
- What category of sub-object is this?
- Is there information I need but do not have?
- Is X a question?
- Is this testable/measurable?
- How can I test/verify this?
Questions can be key tools in navigation, course-correction, and exploration. ‘Questions,’ in a context of music and in a context of improvisational musical dialogues as in some classical music from India, may also be a part of signals and signal protocols between participants where the context is far from analytical problem solving.
Use and Mis-Use of Questions: Group Monologues, Cargo Cults, and Bell-Curve-ish Fake-Meritocracy
One of life’s tragic or comedic lessons in humiliation comes from the unlearnable-ness of project-management and project-failure. People imagine themselves as STEM individuals, rationally solving problems and working together. The reality outside of human delusion is, depending on the resulting level of harm and damage caused, a tragic or comedic display of people paying apparently no attention to anything anyone else says or does, merely ‘believing’ they are engaged in anything resembling a project in which they are a participant.
There is a kind of triangle of death (in reality a myriad pointed star-nexus of countless feedback-factors of destruction, but to start out let’s look at just three) that comes from this toxic brew of ‘group monologue’ behavior (language generation ignoring context), cargo cults (belief ignoring observed data), and Bell-Curve-ish Fake-Meritocracy (institutionally disenfranchising most people for no apparent reason, and giving unchecked executive authority to unaccountable morons on a whim).
If a bitter pill, this is part of the context of the well documented history of STEM where through hard-won (and often two steps forward one step back) STEM has slowly incremented itself towards better clarity despite the constant pressure to erase the boundary between alien-fact and comfortable-fiction. And this can be seen entirely literally within institutions of education where utter nonsense and random-number generation is passed on by bon-vivant personalities who (despite that there remains, as Shakespeare observed, ‘no art to find the mind’s construction in the face) apparently a believed-in mirage (though complete sociopathic fraud is an option too, as in the not infrequent cases where nonsense is rubber stamped as “rigor” openly and “whistleblowers” are vitriolically ousted from polite society).
(Note: Some people, of course, decide to use this historical material to generate a conspiracy theory that STEM does not exist, and that technology is merely another product of the human imagination and the power of social reality construction. While I will defend effective traditions, practices, and open-ness to new data, such extremist conspiracy generation is not something I am arguing for or advocating, directly or passive-aggressively).
Humanity, Patience, and Understanding
Occasional colorful antics of ne’er do wells aside, even when people are trying to do something constructive it still takes a long time and is quite a process.
While it is not a good idea to allow or encourage, for example, a STEM teacher to have openly fake exams, fake evaluations, fake grades, and lots of parties, it would be no better an idea to inflexibly demand impossible levels of accuracy at impossibly fast speeds and all boiled down into impossibly small and convenient simplifications.
About The Series
This mini-article is part of a series to support clear discussions about Artificial Intelligence (AI-ML). A more in-depth discussion and framework proposal is available in this github repo:
https://github.com/lineality/object_relationship_spaces_ai_ml
