Comment by ajani | Hacker News Reader

Comment by ajani | Hacker News Reader

Parent Full thread

The misunderstanding that leads to belief in strong AI is that meaning is somehow embedded in the symbols used to communicate it. Meaning is a natural process that occurs inside each of us, speech is just a symbol of that meaning, text is a symbol of that speech.

Further, meaning is an ever-evolving, ever-mutating process much like the universe.

Training on symbols cannot arrive at meaning, since the meaning isn’t contained in those symbols. Using past symbols, also means no room for evolution.

Machine learning does work though in areas where the needs of the end goal are densely present in the symbols being used for training.

Like recognizing text. We learn to recognize those marks from just the marks, and nothing else. And so those marks contain all that is needed to recognize them. This can be encoded/learned.

But what they mean isn’t encoded in them, nor is it in words, in sounds, in facial expressions, in tones, in body gestures. It may even lie in between us, rather than in us.

Thank you for expressing this nuanced idea so well.

It ties in with deBord's "Society of the Spectacle" [0], a theory that societies evolve from Being, to Having, and ultimately devolve into merely the Appearance of Having.

Your point about the symbols being tools for communication (as opposed to the ineffable ideas being communicated) is also echoed in Lockhart's Lament [1].

[0] https://en.wikipedia.org/wiki/The_Society_of_the_Spectacle

[1] https://www.maa.org/external_archive/devlin/LockhartsLament....

Edited to add that this was extra thought-provoking:

> "It may even lie in between us, rather than in us."

...what? How is this different than saying you can't learn from reading a book because its just symbols.

The best I can decipher is that your argument begs the question, ie you're saying machine learning can't derive meaning from symbols because machine learning doesn't derive meaning from symbols.

You can't learn from "just" reading books. You need to experience the real world to know what "The girl jumped into the stream and wet her shoes, her mum was not amused" means.

Your brain probably made a simulation of girl wearing shoes, jumping, water splashing, wet shoes. You need to know water makes things wet, people can jump, there is gravity. The motivation behind jumping was fun. Humans like to have fun. Wet shoes is not healthy. A mum is the mother and care taker for a girl.

When we're comprehending reading, we're building 3d simulations in our mind and deducing a ton of other info. AIs can't really do that just yet. They don't understand the world like we do.

One of my physics professors used to make a joke every year about how we could read every athletics book written, but we would not start to win races until we hit the track (at some point he made a fun video of him studying “how to run” to show the first day of class).

I interpret this in a similar way. Reading F=ma means nothing, even if you know what each letter stands for, is understanding the consequences what gives you meaning.

> The misunderstanding that leads to belief in strong AI is that meaning is somehow embedded in the symbols used to communicate it

yet you have Roget's thesaurus, where words are grouped into sets of words that have a similar meaning; it might make a difference when dealing in terms of these categories, instead of dealing with word instances; i mean if you have a dependency graph of a sentence, and the nodes are the corresponding sets in the thesaurus, then you might get a similar, but more general interpretation of that sentence.

(i once had a project for parsing/representing Roget's thesaurus in python: https://github.com/MoserMichael/roget-thesaurus-parser There are probably other projects like that around)

Reply on news.ycombinator.com