8 Comments
May 24Liked by Vaughn Tan

can you redo the exercise but this time ask AI to make these judgment calls? i find that AI can definitely decide and definitely make judgment calls, but perhaps it is rather sophomoric and appears to lack taste. but perhaps another AI would disagree.

Expand full comment
author

mmm interesting. if what you mean is redo by rewriting the prompt so that all the intermediate interactions become unnecessary (including the manual rewrite), then my current view is that writing a precise and accurate prompt that generates the exact output with the meaning you want in a single prompt-response cycle is itself a act of meaning-making combined with sophisticated tool use.

so this is even more semantics, but are AI systems making judgment calls themselves in producing outputs, or are the "choices" of outputs driven by structures emplaced by meaning-making humans when they decide the underlying architecture, training dataset, UX, and output filtering protocols for the AI system?

Expand full comment

perhaps not the same meaning making. if i say "summarize and use your judgment as to what is important", what i did was explain to the AI what is meant by summarizing, not what i (personally) see as meaning - the AI supplies what it thinks is important (and you can even ask it to explain).

anyway, i would always use interactive prompting. but what i am suggesting here is that you do the experiment where you ask the AI to supply its own meaning making and see what it gives.

Expand full comment

I came here via meaningmaking at work article

If we go beyond classification of good and bad, essentially meaningmaking is how you draw the boundary of a category or class, when we allow exceptions to the rules etc

In personality science, using factor analysis, there used to be 5 categories ie OCEAN

Now there might be 6 in HEXACO

A reptile is a cold blooded creature that lays eggs while a mammal is warm blooded that gives birth

But we put platypus a warm blooded creature that lays eggs under mammals

Then a more meta version of this would be how many classes / categories do we want? Is it a more qualitative or quantitative distinction?

Is one way of classification better than the other ? Under what circumstances?

Expand full comment

I guess what I’m trying to say is meaning making is mostly (all of it?) doing ontological work

Expand full comment
author

meaningmaking is a type of sensemaking, and sensemaking is ontological work for sure.

but! my current position is that meaningmaking is a special type of sensemaking because it is so overt that the categorisations are inherently subjective. there may be disagreement about whether all phenomena are socially constructed and subjective or if there are some objective facts, but there's absolutely no question that the relative value of things is inherently subjective.

so: i think sensemaking broadly is the process of interpreting what is going on, which requires identifying category boundaries. sensemaking is the general process ("that blob in the image is a cat" or "we have identified 2 common oxidation states for iron ions"), but meaningmaking is the subtype of sensemaking where we cannot escape the fact that both the categories we put things in and what things we put in them are not objectively true. so meaningmaking is stuff like "blood diamonds are morally bad" or "we should exclude child porn photos from the AI training data set." there's no objectively correct set of categories into which to put a blood diamond or a child porn photo, and no objectively correct allocation of those things to particular categories. in meaningmaking, it is a feature not a bug that people can and will disagree both about the categories and about categorisation, and that consensus on both can evolve over time.

so i would pose the question a different way. it isn't that more is better than fewer categories, or qualitative categorisations are better than quantitative categorisations. it's that meaningmaking is the type of categorisation action where we are forced to make decisions that don't have a predetermined correct/incorrect answer, and where the correctness/incorrectness of any given answer potentially changes over time.

Expand full comment
Jun 13Liked by Vaughn Tan

I'm don't quite understand the difference between Type 1 and Type 3. Doesn't Type 1 imply an ordering - rather than a classification?

I'm thinking that meaning-making between two items is less a binary - "good" vs "bad", more a "better" vs "good".

Expand full comment
author

my 2c is that Type 1 is a qualitative distinction — things that are "good" in a particular way are not also "bad" in that particular way. in the category of "good" things, some will be better than others but none will be bad. (Type 2 is also a qualitative distinction ... i think)

Expand full comment