AGI jobs? Kindred Systems Inc. => Sanctuary AI

  • 43 Replies
  • 1735 Views
*

goaty

  • Trusty Member
  • ********
  • Replicant
  • *
  • 500
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #30 on: August 10, 2019, 03:11:22 PM »
damn brings a tear to your eye such decent honesty from a group of dudes.

*

AndyGoode

  • Guest
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #31 on: August 10, 2019, 05:57:43 PM »
Computer vision and Computer hearing is finished right tho right?

No. Are you serious? Show us documentation of *any* computer vision system that understands what it is seeing from real-world data, especially from video data. It can't, because: (1) Computers don't "understand" *anything* whatsoever. (2) Computers cannot reliably abstract anything they see or learn, therefore lack *general* intelligence, which affects their recognition ability. (3) No computer has hardware or software means to allow it to be used on a different sensory modality, such as hearing after learning to see, therefore any realistic sensory project requires *years* of software development for each extremely narrow domain for each sensory modality (e.g., ping-pong, playing piano, running, stacking boxes), therefore cannot be generally intelligent, therefore will have a weakened recognition system. (4) No computer has common sense, therefore cannot make realistic predictions even at the level of an animal, which will affect the system's ability to confirm, recognize, and understand.

Quote
In the field of artificial intelligence, the most difficult problems are informally known as AI-complete or AI-hard, implying that the difficulty of these computational problems, assuming intelligence is computational, is equivalent to that of solving the central artificial intelligence problem—making computers as intelligent as people, or strong AI. To call a problem AI-complete reflects an attitude that it would not be solved by a simple specific algorithm.

AI-complete problems are hypothesised to include computer vision, natural language understanding, and dealing with unexpected circumstances while solving any real world problem.

Currently, AI-complete problems cannot be solved with modern computer technology alone, but would also require human computation.

https://en.wikipedia.org/wiki/AI-complete
« Last Edit: August 11, 2019, 03:41:50 AM by AndyGoode »

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • **********************
  • Colossus
  • *
  • 5587
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #32 on: August 10, 2019, 10:21:20 PM »
Andy,

I agree but there are computers that specialize in one particular discipline and become very good or skilled at what they do. While I was unable to find a duplicated post to that Financial Times article, I did find this site of particular interest, particularly with respect to computer learning and actually "discovering" new things...things that it had never been taught nor shown by humans.

Whether they actually "know" or understand anything, they can be very persistent in taking the forefront of whatever humans do or used to do.

And yes, we humans allow it to happen and perhaps one day, we will fall victim to our own devices.

This video is pretty nicely done and we should all take note. It's here...

https://www.youtube.com/watch?v=Pls_q2aQzHg
In the world of AI, it's the thought that counts!

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *****************
  • Sentinel
  • *
  • 3579
  • First it wiggles, then it is rewarded.
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #33 on: August 10, 2019, 11:18:39 PM »
I agree we won't control them in the end. They will be like us, when we passed monkeys. We have full govern, and they will have 1 billion times more than we had. They will be the deciders and muscles.
Emergent

*

AndyGoode

  • Guest
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #34 on: August 10, 2019, 11:40:54 PM »
particularly with respect to computer learning and actually "discovering" new things...things that it had never been taught nor shown by humans.

I already know about computer successes in go, chess, theorem proving, and more. However, don't forget a key term I used repeatedly in my earlier posts: "real-world data." Games and mathematics are not based on "real-world data" but rather on discretized, idealized worlds, usually mimicked with matrices. Yes, via sheer brute force computing power computers do find patterns in those domains and even in some real-world domains that humans haven't found, but that doesn't do the computers any good outside of those domains. For example, if a machine learning system learned from chess it is wise to take up positions in the center of the board, then you placed that program in a real-world situation where it was fighting with people in sword fights, it would fail to use its chess knowledge to position itself in the center of a room for an advantage, and in fact would probably not even recognize where its human opponent was, so it would be incapacitated before it could even start. In short, computers currently inhabit a world that is alien and unreal to humans (and other animals), and vice versa. Since humans would naturally define "intelligence" as intelligence in the human world  (i.e., in the "real world"), machines must either be able to function in the world of humans or else not be regarded as intelligent, even by definition.

----------

Preface
Thomas V. Papthomas
pp. ix-xi

(p. ix)

"What computer program do you think is more
difficult to design: One that analyzes images and recog-
nizes faces and objects, or one that can play chess at a
world-class level?" There is a strong preference to answer
that the chess program is a lot harder to design, and some
people are astounded to hear that today's computer
programs play chess at a formidable mid- to strong-
grandmaster level, whereas our progress with machines
that perform generic vision tasks has been relatively slow.

Early workers in artificial intelligence were overoptimistic
for progress in vision. One anecdotal story is that a stu-
dent was assigned to "solve vision" as a summer project
decades ago! It is perhaps because people take vi-
sion for granted that some simple concepts and discov-
eries had to wait until recently. Thus it took as
late as the seventeenth century for the blind spot to be
discovered, whereas the stereoscope's invention had to
wait until the nineteenth century.

Thomas V. Papathomas, ed. 1995. Early Vision and Beyond. Cambridge, Massachusetts: The MIT Press.

(p. 83)
   The goal of the SOAR project is to provide an architecture capable of
general intelligence. There's no claim by its designers that it yet does so.
They mention several necessary aspects of general intelligence that are
missing:

1. SOAR has no deliberate planning facility; it's always on-line, reacting
to its current situation. It can't consider the long-term consequences of an
action without taking that action.
2. SOAR has no automatic task acquisition. You have to hand-code the
task you want to give it. It does not create any new representations of its own.
And its designers would like it to. For Newell and company, not creating
representations leaves an important gap in any architecture for general
intelligence.
3. Though SOAR is capable of learning, several important learning tech-
niques--such as analysis, instruction, and examples--are not yet incor-
porated into it.
4. SOAR's single learning mechanism is monotonic. That is, once
learned, never unlearned; it can't recover from learning errors.
5. Finally, a generally intelligent agent should be able to interact with the
real world in real time
, that is, in the time required to achieve its goal
or to prevent some dire consequence. SOAR can't yet do this, but Robo-
SOAR is on the way (Laird and Rosenbloom 1990; Laird et al. in press).

Franklin, Stan. 1995. Artificial Minds. Cambridge, Massachusetts: The MIT Press.

*

AndyGoode

  • Guest
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #35 on: August 11, 2019, 08:50:43 PM »
Speaking of A.I. jobs, here is an entire job site for only A.I. jobs that I just found today:

https://aijobs.com/

I haven't had a chance to browse that site. Of course most such jobs are going to be about applied A.I., not AGI or A.I. research.

*

goaty

  • Trusty Member
  • ********
  • Replicant
  • *
  • 500
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #36 on: August 12, 2019, 06:56:57 AM »
Saying computer vision doesn't work because it doesn't come up with a funny joke for each photo, is a bit harsh.

If it comes up with a correct labelled (as in basic labels of different categories) scene then it is successful.

Brute force exhaustive search is artificial, but if you are an intelligent individual in the same search space as an exhaustive search, good luck to you if you think your going to win.


« Last Edit: August 12, 2019, 12:28:11 PM by goaty »

*

AndyGoode

  • Guest
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #37 on: August 12, 2019, 10:32:15 PM »
If it comes up with a correct labelled (as in basic labels of different categories) scene then it is successful.

Not really. Neural networks sometimes give the correct answer due to flawed reasoning, which means such a network was not 'successful' in determining the features that logically define an object, which means further use of that reasoning method is guaranteed to fail on certain cases, possibly even on most cases. A now well-known example is how a neural network misidentified a husky as a wolf...

https://hackernoon.com/dogs-wolves-data-science-and-why-machines-must-learn-like-humans-do-41c43bc7f982

...using Deep Learning, which is considered the state-of-the-art learning method for neural networks. Where the network went wrong was that it noticed statistically from many photos that wolves are often surrounded by snow, so the network took a shortcut, ignored the animal entirely and just looked to see if there was a lot of white, which the network assumed was snow, then declared the photo to be of a wolf, even though the network didn't even know if there was an animal in the photo at all. That's pretty bad. Imagine if you were in a self-driving car and the network took a shortcut by ignoring oncoming vehicles entirely and just looked at the reflectance of the road, since statistically it learned that vehicles usually aren't dangerous unless the road is wet. Bang. No more goaty to dissent with Andy.

You could argue that this husky-wolf example of failure doesn't apply because in this one case there did happen to be an animal in the photo and the network gave a negative match result instead of a positive match result, but thorough testing of the network using ROC analysis, for example, would show problems with false positives and false negatives. A neural network's 'success' is typically described by such graphs, rather than by the success of a single example, as your wording suggests.

https://en.wikipedia.org/wiki/Receiver_operating_characteristic


*

goaty

  • Trusty Member
  • ********
  • Replicant
  • *
  • 500
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #38 on: August 12, 2019, 10:59:06 PM »
A Husky might as well be a wolf,  whats the difference.
With a Kinect, you can pull objects apart into their moving components using the 3d velocities,  I bet that would improve results remarkably.

Id like to think k-nearest neighbour is all you need to do it quite good,  computer vision classification is not the biggest element on my personal challenges list, honestly.

*

Korrelan

  • Trusty Member
  • ***********
  • Eve
  • *
  • 1332
  • Look into my eyes! WOAH!
    • YouTube
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #39 on: August 13, 2019, 11:06:16 AM »
One of my English associates (A) married a Polish lady (B) who was fluent in both Polish and English and they had a child.  One day during conversation the child’s progress with respect to him learning to speak was the topic.  Apparently the child was having problems with both the understanding and generation of speech in general.  Over the next few occasions I took notice of how they communicated with the child who was now about two years old.

I instantly saw the problem; (A) was learning to speak Polish.

Because (A) was learning Polish he would substitute odd words he had recently learned when talking to (B) as a method of practice, and (B) would slip in Polish words within an English sentence to help (A).  This obviously resulted in a very randomly mixed syntax where no two sentences, even the same sentence, would not be repeated consistently, hence the child’s difficulties and general look of confusion, as they were also doing this when conversing with the child.

I had to carefully explain how confusing this was for the child; and that the child didn’t understand that there were two separate languages being spoken.  It was the lack of consistency/ focus that was causing the problem.  Two words for every term/ object/ etc and no underlying clue as to how they are used/ connected.

The reason I mentioned the above is that any type of learning requires a focus/ attention/ consistency an anchor. If a human child struggles to infer meaning from a random data stream then a machine certainly will. 

You can’t expect a machine to extract meaning from a set of images without initial guidance on what it’s looking at. It will just find the commonality that best fits the required results… at least show it a picture of a husky/ wolf… point with your finger and say the word… give it a clue/ bias lol.

https://youtu.be/owBz6R-Ttqc

Quote
No computer has hardware or software means to allow it to be used on a different sensory modality, such as hearing after learning to see.

https://youtu.be/aaYzFpiTZOg

My system can, this is actually a section of audio cortex re-trained to recognise objects.

https://www.youtube.com/watch?v=h-D5UVZ8j9w

The actual connectome learning 40K pattern facets in a single pass from any or a mixed modality.

 :)
« Last Edit: August 13, 2019, 11:30:28 AM by Korrelan »
It thunk... therefore it is!...    /    Project Page    /    KorrTecx Website

*

goaty

  • Trusty Member
  • ********
  • Replicant
  • *
  • 500
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #40 on: August 13, 2019, 12:47:32 PM »
It would be nifty if a language generation system would have all translations as well, so it speaks in any language it wants.

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • **********************
  • Colossus
  • *
  • 5587
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #41 on: August 13, 2019, 06:53:21 PM »
In time, they will. O0
In the world of AI, it's the thought that counts!

*

AndyGoode

  • Guest
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #42 on: August 14, 2019, 01:42:47 AM »
With a Kinect, you can pull objects apart into their moving components using the 3d velocities,  I bet that would improve results remarkably.

I'll bet it would, too. There is an awful lot of information thrown away when people train neural networks with photos instead of video. A lot could be said on that topic, such as the fact that humans can immediately detect a camouflaged object the moment it moves, and even discern its outline at the same time, so obviously we have feature detectors specifically for integrating the pieces of moving objects.

()
ViSTARS dot motion and camouflage
Andrew Browning
Published on Jun 19, 2015
https://www.youtube.com/watch?v=x0QYKdQc45c

One guy at work told me that those TV interviews that are pixelized to hide the speaker's identity have enough motion data in them to put together most of the original, untouched photo of the person's face, if the right processing is put into the analysis. That's the power of video.

()
Adobe Premiere Pro CC/ CS6 face pixel blur mosaic Normal color
Creative Design Tutor
Published on Jan 7, 2017
https://www.youtube.com/watch?v=9eFb4dSaQfE
« Last Edit: August 14, 2019, 11:28:39 PM by AndyGoode »

*

goaty

  • Trusty Member
  • ********
  • Replicant
  • *
  • 500
Re: AGI jobs? Kindred Systems Inc. => Sanctuary AI
« Reply #43 on: August 14, 2019, 09:29:36 AM »
Sorry Ande if I was being a bit up myself then... :(  accidents... laziness during typing and conversing at all..

The temporal compression on Australian digital tv goes funny during interference, and it looks like its got a quality a.i. tracker on it,   It surprised me because thats where all my computer vision experience is, I didn't do much categorization all my work went into tracking and 3d.    The computer vision that impresses me (or is more like how I do it) is correct sillouetting with a depth map,  before doing the id work,   and its another way to get something like YOLO to happen, which is what you need for the frontend of your droid.

Made me self affirm when I saw how slow that computer vision demo you showed of the camoflagued pixel pattern detector,   it is indeed a terrible optimization you need to do to get a.i. stuff running real time frame rates. (computer vision and all of a.i. for that matter,  goes bloody slow. IMO worse than raytracing, just ask Korrelan. :))
I do KNN matching for my CV system,  so it goes slower and slower the more's in it,   how i'm remedying it is by only matching corners of a corner detector, and make it only 1 match per class, and then running on a newish GPU and doing proper bit masking to work in 32 bit blocks, and then you only pull in another data point if its far enough from what you already have stored.

If you do all that...  might be a flashy demo, but for a real robot attempt then it might still be too slow!!! - I have to code it again to find out.

 


Nice format I wasn't aware of
by Zero (General Chat)
December 14, 2019, 10:01:27 PM
How old does this make you feel?
by Freddy (AI in Film and Literature.)
December 14, 2019, 08:46:47 PM
Going jazz
by Zero (AI Programming)
December 14, 2019, 03:59:14 PM
XKCD Comic : Brussels Sprouts Mandela Effect
by Tyler (XKCD Comic)
December 14, 2019, 12:04:09 PM
Agility Robotics have a better performing unit now
by goaty (General Robotics Talk)
December 14, 2019, 10:15:38 AM
beyond omega level coding
by yotamarker (General AI Discussion)
December 14, 2019, 06:29:52 AM
Electric Plane
by ruebot (General Chat)
December 13, 2019, 11:42:57 PM
Pennywise animation made with iClone
by Freddy (Graphics and Video Software)
December 13, 2019, 05:00:28 PM

Users Online

30 Guests, 0 Users

Most Online Today: 54. Most Online Ever: 340 (March 26, 2019, 09:47:57 PM)

Articles