The word cat is not different from the word dog, or a image of a dog, or liking
women or men, all things are equal, there is no right or wrong or
attractiveness, it is the physics that has rules, you can set a machine to
think snails are sexy, it'll just stalk them and try to get "close" to them,
and be in heaven when it "does" get near them, but physics makes humans end up
being better machines, you can set walls to look sexy but setting AI to look
sexy is going to be what lasts in the end, other race cars don't make it to the
line at the end of the day's test of machines. Back to words and relationships
and images and bodies of atoms in space, there's no use programming it to say
hi or...just let it store what comes with what, and co-occurrences...look up
tree tries and markov chains, it just learns the frequency of 1, 2, 3 etc
letter strings....and look up word2vec and seq2vec, it just learns the
probability of 2 things having similar predictions......this is learning....the
patterns. There's lots of variations of things in space, you don't compute
every atom in a simulation or brute force it all, you take snapshots of the
world using various sensors in the spectrum with sparsely placed cone and rod
cells, and approximately compute the world. You're not supposed to tell it
things really....rather when you talk to it you tell it things in the data, but
not the code...Like it'll pick up what word follows "zebras" more, ans when you
talk to it you can affect what it learns here. You can also get it to make
rules up by talking to it in fact, like Cyc approach...but it is more a memory
than a hardcode when you "install" Cyc in a brain.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T2e6144b49eda1e2e-Mc474f6cdf9bb0e44855e6719
Delivery options: https://agi.topicbox.com/groups/agi/subscription
