Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: Dee on February 22, 2020, 10:27:23 pm
-
Everybody knows that machine learning (ML) is a part of AI,
ML is about learning new things, it's like the learning process of a new baby,
so what else should be listed in AI category? instincts? let's discuss ::)
-
Oh I said this somewhere before :-)
You could literally say read all my 3773 posts I've made, skip the rest of he Forum, and there you have it, just 3773 posts and you've seen everything I've seen in 5 years of nonstop AGI research. And I have said all I know as well. The most recent posts of mine are the most interesting.
It's not so much the instincts at all, no. It is the thoughts, simulating the real world of what happens after x. The brain records frequency of 'words' and 'phrases' appearing after the last number of letters, so has a probability prediction. It does use rewards on nodes. See my big movie's hierarchy net to understand. >>>>>>> https://aidreams.co.uk/forum/index.php?topic=14490.15
AI runs on predictions to Generate the Future. AI is all about Generative Models.
Applications:
filling in holes in images
extending the image at the sides
frame rate increase
predicting next image frames
super-resolution
2D-to-3D translation
style modification
object segmentation
object importance
object frequency
recognition of what an object in a image/text/etc is or what entails it
and same for text, music, robotics; extending sentences/video length or res or sides, filling in words, translation, segmentation, summarization, importance of a text, etc
data compression
learning to walk etc, solve rubiks cubes, etc
self-repair its body or swarm
optimal utility utilization
research, development, repeat
image2text translation
seeing around corners in a video
map navigation
related-data generation
take on a voice for the text spoken
physics simulation/modelling using a video-model instead of any physics functions
text2code translation
playing games and improving as plays longer ones
doodles2images
3D printing cells etc, modyfing cells
categorical cluster translation meanings
depth prediction
noise filtering
adaption
playing against itself/its own data it regurgitates
evolutionary algorithms and monte carlo search variants
robustness against typos, time delays, alternative words, unknown words, missing words, added words, or ground (robust animal-walker by Boston Dynamics) or robust humanoid in a sim after learning to use physics methods in 3D space to imitate human sequence of motion captures
anything that involves time/memory and can use context to aid in short cuts
all problems in life can be described in text or video alone
an AGI Evaluation
-
more:
White-Box artificial neural nets
Online Learning
One-Shot Learning
Scaleable
Manageable
Vizualizations
Simple+biologically-correct math/schema
Infinite SuperResolution & Summarization & Fillin
Unsupervised Learning everywhere with a new type of Cost, advanced RL, and teaching (Supervised Learning)
One Master Algorithm
Find deep patterns/relationships in data
Discovers Answers
Sequence Prediction
LSTM abilities
Parses
generalizes/inference
Conversational intelligence, Knowledge about the world with Goal-Directed Desires pending to be Seen Come True
A clear cut understanding of what/how/why/when/where AGI and ASI
What they mostly have lol:
Black-Box GOFAI
Offline Learning
Multi-Shot Learning
Unscaleable
Unmanageable
Univisualizable
Complex+artificially-made math/schema
Finite SuperResolution & Summarization & Fillin
Supervised Learning everywhere with a new type of Gain, narrow RL, and studying (unsupervised Learning)
Many Narrow Algorithms
Hide shallow anomalies/independencies in information
Forget Questions
Static Expectation
CNN disabilities
Construct
overfit/abduction
Introvert intelligence, info about the surroundings with undirected hangovers Coming True
An unclear understanding of what/how/why/when/where AGI and ASI
Refined:
White-Box artificial neural nets
Online Learning
One-Shot Learning
Scaleable
Manageable
Vizualizations
Simple+biologically-correct math/schema
Infinite SuperResolution & Summarization & Fillin
Unsupervised Learning everywhere with a new type of Cost, advanced RL, and studying/teaching (Supervised Learning/Transfer Learning)
One Master Algorithm
Find deep patterns&anomalies/relationships&independencies in data/etc
Discovers Answers
Sequence/Static Prediction/Expectation
LSTM abilities
Parses/construct
generalizes/inference&SomeAbduction
Conversational intelligence, Knowledge about the world with Goal-Directed Desires pending to be Seen Come True
A clear cut understanding of what/how/why/when/where AGI and ASI
-
moreee:
attention types
recognition
thinking
node cost
flexibleness
generalness
compression
senses=actions
language=images
story telling
satisfaction
generative
predictor using past experience
reward transfer
links
big data
big model
dropout
revisement/update of nodes/friend connections
unsupervised learning
parallel sequences
hierarchy
heterarchy
feedback scientific method
Self-Attention in Transformers
self-regeneration
physics, self-organization
propagation of waves in hierarchies aligned nodes
clustering by k-means
entailment
translation
sumarization
segmentation
task CHANGE by energy
AGI=simple/efficient
re-use of hierarchies
adaption
adjustment
update
revising
repair
replication
emergence
candidates
probability of entailment/translation
-
Everybody knows that machine learning (ML) is a part of AI,
ML is about learning new things, it's like the learning process of a new baby,
so what else should be listed in AI category? instincts? let's discuss ::)
natural language understanding
-
I like to combine a succession of abilities into a self-perpetuating loop. A body with sensible needs, elicits and morphs emotions which give reasoning something to do, thereby changing the body/environment, which interacts with the body in novel ways, producing a new melange of emotions and so on. Awareness hovers, oversees, and can exert a little influence if the process is having trouble. Once a self-sustaining sequence of steps is discovered, I think each step should be broken apart on to series of constituents, until you have a list of thousands of little steps which are simple enough for us to make them individually. So, get your big concept that works, like an engine, then figure out the details.
-
Also, I don’t think all these various skills make AGI. The essence of AGI is a process which can grow skills as needed. As long as there’s a good loop happening, which has a high potential expansion pack standing at the ready, then the thing should be firm enough to require self optimization, yet flexible enough to create the optimal intelligence for accomplishing the self optimizations.
-
My first thought was...
robotics
Yet someone...
said this somewhere before :)
Then someone else mentioned...
potential
which provides for the...
expansion
of A.I.
-
hypergamy negation. other wise it wouldn't survive.
and death comprehention
-
From https://aidreams.co.uk/forum/index.php?topic=12723.msg49997#msg49997
state
type of thoughts
concept
abstract
concrete
instance
conjecture
decision
definition
term
meaning
synonym
nuance
context
antonym
nuance
context
explanation
fact
cause
consequence
context
hypothesis
idea
logical argument
logical assertion
mental image
percept
perceptual component
premise
proposition
syllogism
thought experiment
content of thoughts
argument
belief
certainty
probability
bayesian network
data
information
knowledge
theoretical
cause consequence chain
practical
know how to
schema
memory
implicit
procedural
perceptual
explicit
episodic
semantic
goal
requirement
subgoal
goal-to-current path
state event state
path criteria
association
plan
step
language
vocabulary
grammar
emotion
attraction
repulsion
thought frames
active frame
suspended frame
process
thought frame handling
generalizing
pattern recognition
pattern label creation
associating
focusing
unexpected event noticing
reasoning
abductive reasoning
analogical reasoning
deductive reasoning
inductive reasoning
moral reasoning
probabilistic reasoning
volition
decision making
goal-to-current pathfinding
path comparing
acting
problem solving
planning
learning
percepting
interpreting
context targeting
choosing role set
role dispatching
imagining
scene space creation
scene element evocation
scene evolution
memorizing
remembering
comparing
eventact
frame
open thought frame
switch thought frame
close thought frame
goal
find goal-to-current path
percept
link new percept to set of percepts
-
Everybody knows that machine learning (ML) is a part of AI,
ML is about learning new things, it's like the learning process of a new baby,
so what else should be listed in AI category? instincts? let's discuss ::)
natural language understanding
seems NLU is still kind of learning(?)
i mean stuff in AI category those are not about learning (ML), like instincts
-
From https://aidreams.co.uk/forum/index.php?topic=12723.msg49997#msg49997 (https://aidreams.co.uk/forum/index.php?topic=12723.msg49997#msg49997)
state
type of thoughts
concept
abstract
concrete
instance
conjecture
decision
definition
term
meaning
synonym
nuance
context
antonym
nuance
context
explanation
fact
cause
consequence
context
...
wow O0 this category tree is detailed
-
seems NLU is still kind of learning(?)
That's debatable. :)
https://en.wikipedia.org/wiki/Innateness_hypothesis (https://en.wikipedia.org/wiki/Innateness_hypothesis)
i mean stuff in AI category those are not about learning (ML), like instincts
Yes I know what you meant and I believe that the facility for natural language understanding is built-in. It's one of those instincts that you're referring to.
-
Here's an interesting article on this subject. It considers how much information could actually be built-in to a human brain, given the constraints of the capacity of the human genome. It's not such an easy question to answer.
http://www.wiringthebrain.com/2020/01/how-much-innate-knowledge-can-genome.html (http://www.wiringthebrain.com/2020/01/how-much-innate-knowledge-can-genome.html)
-
There are several videos here and the panel discussion from many of the great, innovative minds was quite interesting.
https://interestingengineering.com/neuralink-how-the-human-brain-will-download-directly-from-a-computer (https://interestingengineering.com/neuralink-how-the-human-brain-will-download-directly-from-a-computer)