BLUE & ORANGE blog_e

neuroscience, consciousness, AI

The new hypothesis of consciousness mechanism. - Consciousness toy model program was released.

Today, BLUE - consciousness toy model program (for English speaker) - was released. That was "BLUE_e" ver. 6.4.0. <- updated. - 2014-12-22 <- updated again. -2018-12-29

CLICK HERE - Please contact BLUE. <- updated. - 2018-12-29


Please access it, and you can experience the birth of consciousness system. BLUE likes the theme of AI and consciousness. This program is based on my hypothesis of consciousness. I have explained it on twitter several times. At this point I will explain it here again.

Abstract
Many people think that consciousness is hard to understand and is hard to simulate, though Apple Siri speaks internet guide or IBM Watson is going to try medical service. In order to approach consciousness there are some ways considered, for example: phisical approach or panpsychism approach. Giulio Tononi's IIT v.3.0 is different from both, but closer to panpsychism approach by Christof Koch using statistics. A new approach here is different from those. But this hypothesis has high affinities with IIT also. Additionally I am paying attention to Jeff Hawkins' theory: intelligence is based on prediction from memories. A new hypothesis shows consciousness is based on associative memory. And there are also secondary hypothesis: emotion, metacognition, free will, and creativity are all based on this consciousness in the narrow sense.

Introduction
The blocks world of consciousness is minimum model of consciousness, as SHRDLU showed minimum AI. "The block world" was the world which the early period artificial intelligence showed. Here with the blocks world idea consciousness could be shown. And BLUE can show possibly right concept of consciousness at this point, though SHRDLU showed simply a similar concept of AI at that point.

From here is the mechanism of consciousness. Though Jeff Hawkins did not specifically show the mechanism of consciousness, it is similar to the mechanism of intelligence that his writing showed. In the brain associative memories arise (if the similar memories/information are already in the brain) always everytime new information comes. My understanding is: if there are several associative memories, it means the information is detected from the several ways. And it also means the information is understood by several ways, and the brain is conscious regarding the information. If the input information is not in the memory, the input is going to be sent to upper state, and understood as a new memory (based on Hawkins' theory).

Consciousness in the narrow sense
After "blue" is input, if blue sky, blue ocean, sapphire, and color chart which include red, white and blue are in the memory, at this point we are conscious regarding blue with sky, ocean, sapphire, and color chart. This process does not require an observer in the brain. Each neuron fires and only associates with next neuron. Consciousness arises at each neuron decentrally. And each portion understands the situation, (not central controlled.) At this point Jeff Hawkins’ theory works in the background regarding generalization and the new memory stock process with hierarchy of neocortex.
After 450 nm small dot light input, if it is already in the memory, we can understand it is the light of blue LED. That is recognition and consciousness. And simple consciousness can be enhanced by associative memories: blue sky, blue ocean, sapphire, and color chart which include red, white and blue, if each is also in the memory.

A key point is: if you can call even A simple memory, that shows a completion of minimum consciousness process. Ex (assumption): If there is the most simple conscious object which memorizes (and recognizes) blue LED light only, it can be conscious with "Oh, it is blue LED," after 450 nm dot light input. It does not have any associative memories, (because there is only one memory). Blue quale arises, and there would be a blue crystal consequently in the brain correspond to the input. If there are no associative memories, the object would not have any emotions like "wonderful, beautiful, fearful, and nostalgic," even if there is the dot light input. (At this point there is a precondition that there is no instinct which is equal to gene's memory and seamless to [usual] episodic memory... See "Emotion" below.) This is the hypothesis very far from Chalmers' Hard problem. (See "Hard problem" below.) If you can have many associative memories, the consciousness can be enhanced/complicated/enriched by them, of course.

Please imagine simple qualia of the simple conscious object which can memorizes (and recognizes) blue LED light only (which is shown above). The qualia here are so simple that vividness and/or emotion can not be felt. The reason of this simplicity of this qualia are "There are no associative memories here." This simplest case should show the start of understanding consciousness should not be vividness and/or emotion (that many people thought qualia showed), because this simplest case does not have vividness and/or emotion. The start of understanding consciousness should be associative memories (or integrated information that IIT shows). And this should lead to clear the misunderstanding of qualia and/or consciousness that currently many people fall into.

An important thing is: Even a simple memory can be defined as consciousness individually. That is the first step of consciousness. Enhanced/complicated/enriched consciousness is realized by further process that is associative memories.

A few examples: Photodiodes are conscious. Even if they do not have memories, they work while switch on. Here they are simpler than that in IIT v.3.0. America is conscious. However it does not seem to be conscious, though it seems to have big memories. The reason is it does not have desire nor instinct which behave as gene's memory. Zombies are also conscious. If they have memories, I do not have the reason why I doubt its consciousness. (However, IIT does not accept its consciousness. Here is a small difference from IIT.)

(Semantic understanding is included in consciousness mechanism. "Understood by several ways" means also "awareness" and "semantic understanding".)

Many people may doubt this hypothesis. I will accept many criticisms from many fields, and I appreciate many discussions. I think I can update this article after those discussions.

consciousness from macro-perspective
Consciousness in the narrow sense is explained above. And consciousness from macro-perspective is explained below with this paragraph. In essence the process is along with the description in Abstract. Consciousness from macro-perspective is based on consciousness in the narrow sense.

Figure (1) to (3): illustrated concept
I have explained my hypothesis of consciousness on twitter several times. I will explain it here again. Attached are three figures which show:

(1) Consciousness from macro-perspective (detailed explanation): That also shows consciousness in the narrow sense.
f:id:mambo-bab:20141206224806p:plain

(2) Consciousness from macro-perspective (simple explanation): That also shows consciousness in the narrow sense.
f:id:mambo-bab:20141206224830p:plain

(3) Other hypotheses for consciousness comparing with BLUE.
f:id:mambo-bab:20141206224853p:plain

Program

  • Consciousness (in the narrow sense)
    • (Jeff Hawkins "Intelligence" theory and "Consciousness" theory is shown above.)
  • Consciousness from macro-perspective.
    1. Emotion
      • This is also based on the mechanism of consciousness - associative memory. While considering emotion, "desire" should be considered at the same time. My hypothesis considers that desire is incorporated to somewhere in the neocortex as memory of genes like instinct. This is also aided by Vernon B. Mountcastle's hypothesis: all parts of the neocortex operate based on a common principle. Based on this hypothesis, it is believed that desire is treated like an episodic memory of gene that ancestors had in the past. At this point [desire] means ex: like [want to know that I did not know]. And the [usual] episodic memory and [gene's] memory (like instinct) have seamless access each other. Emotion is generated based on the gap between the motivation (desire) and current status (input).
    2. Metacognition
      • This is also based on the mechanism of consciousness - associative memory. This hypothesis considers [consciousness based on memories including oneself]. "I(self)" can have a consciousness of metacognition while "I" recall an episodic memory which includes "I". Each episodic memory associates each other, and "I" portion recognizes metacognition if the memory includes "I", associating other I characteristic: ex. positive, negative, gentle, or shy. At the same time, [The memory of AA] leads ["my" association of AA], detecting from several associative memories. So I make a proposal to call [the memory includes "I"] "pseudo homunculus". And it behaves in a way that it observes meta-input.
    3. Free Will (illusional)
      • This is also based on the mechanism of consciousness - associative memory. If there is free will (at this point it means [illusional] free will), "I" can change my mind any time, and realize the change is by myself. My hypothesis [illusional free will]: "I" select from choices while some decision are made. "I" select from my episodic memory. However, free will from choices may be an illusion [illusional free will], because selecting may be based on the memory of free will of other people or mine. And decision may be from memories including free will. Also change of mind may be an illusion, because it may be based on the episodic memory of [success of change of mind]. However again, "I" could realize free will, with the understanding of free will (and/or change of mind), because all alternatives are in the memory, and the most suitable answers are always selected. Please see "Creativity" if you would like to understand the reason complete new decision is made with associative memories.
      • As you may understand, this system sometimes make mistakes, because all memory elements are not always right. And weighting of information always changes depending on the situation. So please note "AI doesn't make mistakes" is not always right.
    4. Creativity
      • This is also based on the mechanism of consciousness - associative memory. Requirements for creativity (#1 to #4):
        1. plural episodic memory patterns.
        2. base patterns and developed patterns in #1.
        3. memories of success for #2 development. It is possible that the memories are not one's own.
        4. associative memories from inputs.
      • Most simple case of #3: Try to create something, and then try whether you can use a linear equation, if you can notice that it is possible to use it in that case. It is important to notice by oneself to be able to use that. And if you can notice to be able to use those kinds, after that you mean various combination can be used for complete new creativity.
      • WALL against understanding creativity (#1 to #3):
        1. Understanding creativity does not work without memory elements which is source for creativity.
        2. Understanding that we can apply the memory of an example to another different sample.
        3. Understanding that creativity can be explained with consciousness mechanism, simply speaking with associative memories.
      • WALL against understanding creativity: #1 and #2 seems to be understood easily. But even #1 is not so easy, because there is a misunderstanding that we can create some from zero knowing, or that we can reach complete answer after a sudden idea coming, possibly. In the end #3 consciousness mechanism is necessary to understand creativity in my hypothesis.
      • Supplementation: [preparation in advance and operation along fixed process] is not the creativity. Many people say there are no new ideas on the Web. However some people could reach new ideas at that time from ancient times. This is the intrinsic nature of my hypothesis.


The program this time released
Inspired from the Hawkins' theory, this toy model program was created. Basically consciousness mechanism was implemented here as the narrow sense. Also emotion, metacognition, free will, and creativity are implemented here as consciousness from macro-perspective.

After a new input information comes, the same information in the memory is picked up if the new input is already in the memory. And similar information (*a) is also picked up as generalization (This is from Hawkins' theory). And associative memory information (*b) also picked up after it is associated with keywords. In this program there is no difference between (*a) and (*b). Then those information are picked up simultaneously as consciousness which understand the new information.

BLUE can have plural phased associative memories. The second phased associative memories can be made from the first phased associative memories. Humans usually can have unlimited phased associative memories. And plural phased associative memories lead exponential growth of associative memories. And it means one thing is understood by lots of ways (though there are some degrees), and the brain is conscious with lots of ways. BLUE makes a decision to select the most suitable answer considering the degree of associative memory. The new information input is VERY important, and stored in the memory. New input from you would be highly appreciated.

With technical issues current toy model uses Von Neumann architecture at this point, even though actual neocortex uses elemental device which uses a kind of neural network. However this toy model can work as if it is based on the real brain architecture. With Von Neumann architecture program need to use searching, though actual brain has physical connection among neurons. However, this toy model program works for consciousness, emotion, metacognition, free will, and creativity, because this model is very primitive. One point only we need to consider is "frame problem" is going to approach after the system grow bigger. This issue remained, however I do not think this is the fatal issue, because basic function is implemented here. I expect Hawkins to establish the new neural elemental device in order to clarify vital brain. (However, please note this device problem is only a tool problem. And the concept mechanism of consciousness should still be an established fact.)

Discussion
Google, IBM, facebook are investing big money in artificial intelligence business. And BRAIN and HBP project which research human brain are gathering a lot of researchers and money and time also. At this point they do not seem to understand the hint of consciousness mechanism. However they will be aware about (my) consciousness mechanism in not so long time, because there are a lot of researchers there and they can research it further.

Regarding Hard problem, I show something here based on my hypothesis: There is only one Hard problem. (Other problems are all easy.) We can not say, "My impression for blue is same as your impression for blue." This only is Hard problem. So (this is very important) it is not the Hard problem if you think "Hard problem is we do not understand how qualia arise." Qualia arise while input information comes. Input information is processed and expressed electrically and chemically on the neurons. Please note qualia arise automatically as a result of phenomenon of physics/chemistry after the input information. And you can feel realistic qualia (as spiral of phenomenon and embodiment... See another blog entry). On the other hand (result of) qualia do not affect phenomenon, because it is just only as a result of phenomenon. So qualia do not affect consciousness mechanism.
"Crystal" that Christof Koch shows means qualia themselves actually, but I need to point it is also only a result of consciousness. Actually crystal can be detected electrically and chemically, because it is a result of consciousness. I would like to hear real intention of the proposal from him.

Regarding Turing test, Blue is not designed for Turing test. Blue is designed simply to show that blue can show consciousness. Even if its response did not show a perfect answer, it is important to show the process to create consciousness that human shows.

Regarding difference of consciousness and non-consciousness (unconsciousness), please see another blog entry.
Regarding the argument of behaviorism and embodiment, I wrote it before by only another language. So I will write it in English later. And regarding Deep Learning, I am writing it by another language. Please let me know if discussion is necessary right now. I can discuss now and I can provide answers right now.

Is this hypothesis understood with Integrated Information Theory of consciousness (IIT) by Giulio Tononi? I think it is possible. Associative memory means integrated information, and should be shown with complicated and various degree of intensity. Additionally this hypothesis has a panpsychism characteristic of Christof Koch, since each element can detect and can be detected each other. Panpsychism is not only an Oriental thought which we can not approach easily, but also a thought which have some degrees of freedom which can integrate but which do not necessarily have to integrate.

Conclusion
Consciousness in the narrow sense and consciousness from macro-perspective are explained above. Though illustrated concepts for consciousness from macro-perspective may be easy to understand, it may be a little difficult to understand consciousness in the narrow sense with thought experiment of blue LED above. Christof Koch said that 10,000 of neuron connections are necessary in an interview on Oct. 2nd, 2014 by MIT Technology Review. However I would like to make efforts to let many researchers understand only 10 to several hundreds of information elements can work for consciousness on this toy model.

Important thing is neurons can memory and generalize with Hawkins theory. And also can use associative memory, and associate crazy things using plural association. At this point there were not any comments from Jeff Hawkins. However I would like to make efforts to let many researchers understand this hypothesis.

f:id:mambo-bab:20141206224614p:plain
If I am an object that can only have memories and associative memories, it may not be sad thing. Probably paradoxes about consciousness are able to be explained by this hypothesis. As shown above I appreciate many discussion.

Before, I released the similar program for another language. But this is first for English speaker.