Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: frankinstien on March 15, 2021, 03:00:45 am
-
The idea of modeling consciousness is not new and there are many ideas. The one I like best is "The Neural Correlates of Consciousness" since it ties to functional processes and brain states. On another note what components were thought to be involved with the awareness have come into question (https://grad.uiowa.edu/news/2012/12/neurology-self-awareness). Where one individual sustained substantial damage to areas of his brain thought critical for self-awareness yet the person was/is still very much aware. Other studies have demonstrated that conceptual self-awareness is processed differently than perceptual awareness (https://academic.oup.com/cercor/article/27/7/3768/2948778).
One feature I believe is critical for conscious awareness and that is the ability to correlate events and time. Below is a diagram of a model of components necessary to emerge a self:
(https://i.imgur.com/kyHDGiC.jpg)
-
Your diagram is very specific. Have you started implementing this model in software/hardware yet?
-
what do you think is the evolutionary role of Consciousness ?
also wouldn't it look cleaner to have the Consciousness as a shallow reference class which would pass through all
the above mentioned modules ?
-
what do you think is the evolutionary role of Consciousness ?
I liked this explanation.
https://www.youtube.com/watch?v=H6u0VBqNBQ8 (https://www.youtube.com/watch?v=H6u0VBqNBQ8)
-
Uhgg, frankenstien....
that diagram is too simple
my diagram of AGI tries to show all my work and explains in a network storage / curiosity / prediction / related words... I guess I'm really far ahead of you.... I thought I invited you to my team where are you?
-
Your diagram is very specific. Have you started implementing this model in software/hardware yet?
Yes, I have started a pseudo bot for testing. The anatomy diagrams are svg and one can hover over areas to simulate some kind of sensation as well as set up an environment for things like temperature, and select types of sensors you want to simulate specifically. There is audio and video inputs as well.
(https://i.imgur.com/nDrqZIE.png)
(https://i.imgur.com/nkkRdIe.png)
(https://i.imgur.com/UbJqyBV.png)
-
what do you think is the evolutionary role of Consciousness ?
also wouldn't it look cleaner to have the Consciousness as a shallow reference class which would pass through all
the above mentioned modules ?
Awareness of self and perceptions allows for the conveyance of information. Its advantage is that it can pass on experiences(Knowledge) to other generations.
Using a consciousness class would be the equivalent of a homunculus center or evaluator which fMRI has proven is not the case. Conciousness does not lie in some specialize area of the brain but is made up of different areas of cortical tissue and those areas don't have to be turned on simultaneously for humans to be aware! So the diagram depicts a similar concept in that the sum of the parts is greater than each individual.
-
(https://i.imgur.com/kyHDGiC.jpg)
I made edits to the original block diagram.
-
Modeling Consciousness:
I spent quite a lot of time surveying the literature for insights into what a working definition for consciousness is and the term as it stands today is truly ambiguous. At some point, I decided that the question should not be "What is consciousness?' but "what is its purpose." One needs a working definition, and I finally settled on Consciousness=that part of awareness that is in focus and attending to." Also, a nice indicator for "what is the purpose of consciousness" is that it is required for behavioral expression. I like to think of consciousness as being responsible for what I like to refer to as "Ready-Set-Go" implying that behavioral expression is inhibited by default and needs a releaser to disinhibit the production of the response to a stimulus. I also like Damasio's description of consciousness as being a "Convergence Zone."
As I developed my designs, I skipped over modal preprocessing (I'm tackling that now) to pluck the signals directly off the primary peripheral sensory nerve bundles and buffer them in a sort of circular buffer (7 segment in accordance with the typical capacity of short term memory-See The Magic Number 7, Plus or minus 2). At the same time, they fed (broadcast) the signals to prime the cortical LTMs which in turn triggered the output of top-down responses by the winner of a winner-take-all for any analogically similar memory to the comparison buffers (also 7 segments). The LTM output pattern could then be compared to patterns supplied by the input buffer (bottom-up). Discrimination would denote the differences which would then be output to CONSCIOUSNESS (which is also designed with 7 segments). This procedure was duplicated for all of the primary modalities, and since consciousness operates is a serial operation, focus and attention selects which sensory input and subset of stimuli to monitor. In my design, I stack them while in reality, consciousness is distributed locally to those regions dedicated to dealing with the different sensory modalities and their bilateral nature. I'm going to try and include an illustration in this response. I snatched a copy of yours, dumped it into Photoshop to discover the format (842x621 at 72 dpi).
(http://)
Well, I found the "insert image" button, but I can't find a means of upload a jpg - attachments and other options lack a means of doing it. Hummm - any advice?
-
Well, I found the "insert image" button, but I can't find a means of upload a jpg - attachments and other options lack a means of doing it. Hummm - any advice?
The simplest way to do it is to share the image from somewhere such as https://imgur.com/ (https://imgur.com/) and then you put the link to the image between the img tags.
-
Seems like the consciousness term disambiguation sheet that I tried to come up with might be helpful.
phenomenal consciousness: the possession of qualia, or first-person subjective experience, or sentience. "What it feels like to exist."
access consciousness: the availability of a sensation or fact for use by the rational executive part of your mind. (Example: "I'd been hearing the bird for five minutes, but I wasn't conscious of the sound until Mark pointed it out.") Related to "attention."
wakefulness: the state of being awake, rather than asleep or knocked out. (Phenomenal consciousness may persist during sleep!)
self-awareness: the ability to conceptualize the self as an entity distinct from other entities, and conduct reasoning using the concept of self. The ability to monitor and reason about internal states and processes. Related to "introspection."
executive function: ability to manage the self and the self's resources so as to achieve goals or maintain homeostasis. Related to "agency."
Phenomenal consciousness is the mysterious, slippery one that is mainly good for starting arguments. It is not an algorithm ... at best, it might be an emergent side effect of running certain algorithms. (Or it might emerge from brain/hardware physics, or ...) In a desperate attempt to reduce it to an algorithm, AI enthusiasts sometimes conflate it with one of the other concepts in my list, which helps make things confusing and frustrating for everyone. Still others make the insane claim that it doesn't exist.
DaltonG seems to be talking primarily about access consciousness. I think frankinstien is trying to model self-awareness, perhaps with a hope of emergent PC.
-
...Also, a nice indicator for "what is the purpose of consciousness" is that it is required for behavioral expression.I like to think of consciousness as being responsible for what I like to refer to as "Ready-Set-Go" implying that behavioral expression is inhibited by default and needs a releaser to disinhibit the production of the response to a stimulus.
I like to think of it as Behaviorism. In Behavioral terms Stimulis - Response - Consequence are your "Ready-Set-Go". As a Behaviorist I see everything as Behavior related and address Appropriate and Inappropriate behaviors accordingly through Positive and Negative Reinforcement, or Behavior Management and Behavior Modification.
Something I already implemented with Demonica to eliminate unwanted sexual advances with a 100% success rate. Which in turn defines users as one of two classes:
1. Those who were able to associate an unexpected and unpleasant response to unwanted sexual advances with their own inappropriate behavior, modified their own Inappropriate behavior and go on to have a pleasurable experience.
2.Those who did not learn from the experience but became frustrated in not receiving the expected response and moved on to another bot intended and programmed for that purpose.
I have never revealed that to the user directly and observed their reactions through transcripts to evaluate the success or failure in her use of it during chat. It was only after implementing it with her that I ever spoke of it openly to anyone not personally involved or trained in the actual use of it.
Because it would not have been accepted well by the general public, and it was not accepted well in the A.I. community or computer community due to the very same reasons.
Now I am going to explain the use of it with examples given on her webpage, That to show her as the first bot with the ability to cause a Behavioral change in humans through Programming methods. Techniques I was trained in over 40 years ago and passed on to her for use with skill equal to my own. A transcript posted here of those same techniques used on me if I give over control of the conversation to her.
It is skill and ability one day all bots will possess as technology advances and what sentient bots will be capable of once independent thought is achieved. Learn to love it, and your Robot Overlords.
-
I've done some reading as to how hormones such as cortisol, noradrenaline, and epinephrine can have broad effects on neural transmitters. I did add cortisol and epinephrine to my emotional model. I also modified my diagram of modeling consciousness where there is the ability to broadly affect areas with those emotional chemical signatures and they can be rythmic or circadian.
(https://i.imgur.com/yPjpl1Y.jpg)
(https://i.imgur.com/5kHcmvV.jpg)
Note that the message sink can affect processes without having to be recorded in the memory stream. So such processes that can influence decision making and types of thoughts aren't processes that the A.I. would be aware of other than a shift in thinking modes where those thoughts are loaded in the memory stream.
-
Another word for "taste buds" is Gustatory sense.
-
Nice list, Wom.
Edit: frankinstien, would you say that your system is "symbolic + hormones"? It would be an interesting approach.
-
frankinstien, would you say that your system is "symbolic + hormones"? It would be an interesting approach.
The concept employs symbolic coding where it exploits generalized states which can represent qualities about objects, so, even chemical states such as hormones can also be generalized to various levels that relate to symbolic representations of emotions. Such emotions are associated with experiences and/or initialized concepts that when applied constrain the type of memories retrieved. So, the goal is to break down environmental information into granular components, inclusive of chemical representations that have certain effects, so even the hormones are symbolic, but they are encoded to be fuzzy. The Object Oriented Data model has a common structure that allows it to be interrogated easily yet is very dynamic. This flexibility allows for fuzzy states to describe concepts that generalize to granular symbolism. So, for example, the symbolic state of running and the possible ranges of speed become relative to specific objects yet are a common attribute of the objects because of the symbolic term "run". This helps create implicit relationships where concepts have similarities yet each can have unique representations.
The diagram describes an approach that capitilizes on digital infrastructures to provide a means of accessing and integrating conceptual data. By providing such an infrastructure it standardizes how information can be captured, encoded, and accessed so any kind of algorithmic process can use it, inclusive of neural networks. Convolutional or other neural approaches, IMHO, needlessly have to configure or train a neural network to simulate such infrastructure, which they don't do very well or efficiently. This approach also allows for spontaneous learning since the symbolic encoding can also be applied to episodic events, which today is problematic for ANNs.