Secret to AGI & ASI

  • 69 Replies
  • 4573 Views
*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Secret to AGI & ASI
« on: March 27, 2018, 10:38:54 pm »
AGI (Artificial general intelligence) is the intelligence of a machine that could successfully perform any intellectual task that a human being can. It is a primary goal of some artificial intelligence research and a common topic in science fiction and future studies.

ASI (Artificial Super Intelligence) is a term referring to the time when the capability of computers will surpass humans. ASI will eventually create the technological Singularity.

Key ingredients to ASI:

1) Tree search, it's thinking process. At first you could use a simple tree search, but eventually you'll want to make this as flexible as possible such that it will form over time with experience. The AGI can have the ability to form different types of tree search depending on the situation. https://en.wikipedia.org/wiki/Search_tree

2) Pattern recognition routines. Give the AI as many as possible. Through experience it will learn which ones are best for different situations.

3) Database.

Every game has rules. In chess there are two goal oriented objects called humans who take turns. Each human has a set of pieces to move around, and each piece has its own mechanics. A simple game.

In the game of Life there are countless goal oriented objects, even a toaster. The toaster can work, it can break, the filaments could break and eject hot molten metal. In this game we don’t need to take turns. For example a human can move to a dozen different locations in a row. In the game of chess, the *goal* of each human is to destroy their opponent, but in the game of life the *goals* are different, sometimes unknown. Although sometimes in may seem like it, not everyone is out to destroy you.

Regarding AGI / ASI, each goal oriented object is assigned a set of goals written in it’s own AGI language of logic. In chess the goal of each player is to capture the King. In the game of life, objects have their own goals, which can change. It's through experience that the AGI learns the goals.

As you can see, the game of Life rules are a bit more complicated than chess. We’ll see, but IMO the game of Life is far too complicated for present computers to create a NN (neural networking) human-like brain. Even with Google’s supercomputers and custom ASIC chips. There’s nothing magical about NN. It’s basically the world's slowest interpreted language. Although it requires no programming skills. It’s called machine learning and it teaches itself given that you setup the environment. In the end, what will bring about true AGI and eventually ASI will be tree search and pattern recognition routines (simply stated). Someone on this forum already has a good start, AIRIS, a non-NN. Here’s their website if you want to check it out:
https://airis-ai.com/

OpenCog is another great project, open source. Again, it’s non-NN:
https://opencog.org/

OpenNARS is another great non-NN project:
https://github.com/opennars/opennars/blob/master/README.md

Even Google’s DeepMind is now migrating to non-NN methods such as a tree search. When they added a tree search, their AI became like a God compared to before. I predict DeepMind will slowly migrate away from neural networking. The advantage Google has is money. They can design custom ultra high performance ASIC chips that are specifically designed for neural networking. I’ve been designing electric circuits for decades and can tell you that a custom AIRIS chip for AI tree search would increase performance by tens of thousands. There are ways to design the circuit so that each processor/core/thread can read global memory simultaneously with other processors. In other words, you can have 10,000 cores reading the same memory at the same time without any delay. That's what tree searching needs. I call it PRAM (Parallel RAM), not to be confused with apple’s Parameter RAM. Another improvement is hardware DB, which also takes advantage of parallel processing and PRAM.

Here’s a simple example of the AGI thinking process. When the AGI is asked a question, it would convert your question into its own AGI language of logic. It would form a tree search, which has a specified set of goals formed in it’s own language, a simple programming language. The tree search would eventually call routines that performs analysis based on data & pattern recognition. Throughout the tree search the AGI will periodically check if the tree search is completed *sufficiently*. If the question had enough priority, the tree search would consider more extensive methods to find answers. If for example the AGI was in a robotic body with arms and legs it could go on foot searching for answers. The *root* tree search goals ultimately determines the results. If the root goal is simply for the AGI's self-improvement, then it might not even answer you. It's tree search would spawn other tree searches trying to find out the consequences of not answering you. On the other hand, if the root tree search goals forced the AGI to answer you, then of course it would provide the best answer. There a lot of possible tree search node paths. There are routines to determine the importance of a node path, which can call a tree search pruning routine such as alpha-beta pruning. You can only do pruning when the object is an opponent such that it's goals are the opposite of your goals. The tree search is truly a real thought process. It comes up with unique ideas and decisions.

Node scores are relative to each object's scoring method. In chess, your opponent does better when you do worse. In real world, we can't assume every object gains when you lose. So, the score of each node is relative to each object.

The AGI has it's own native language, which is a logic language. It can be as simple or as complex as you wish. It would have if/else/conditions, equal/set/=, object IDs (I refer to objects as clusters in my AGI). Each cluster has a unique cluster ID. The AGI does not search the database for text words. It searches for cluster ID numbers. For example, the word "human" could be cluster ID 93458517. Everything has a cluster. A cluster contains all info about the object, details, links, analysis, assessments, etc. about the cluster The AGI does not require a logic language. It would struggle at first, but eventually build a sufficient cluster database to make sense of the world.

Edit: People are being emotional an fighting. So if you’re interested in creating AGI, then send me a private message. I'd prefer deep thinkers. Spocks rather than Klingons. We need to resonate. If you don't get me, the way I think, then it's probably a waste of time.

Best wishes!
« Last Edit: April 01, 2018, 06:56:25 pm by unreality »

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Secret to AGI & ASI
« Reply #1 on: March 27, 2018, 10:58:07 pm »
Regarding tree search and pattern recognition routines. The AGI will need interface mechanics. Examples of interfaces are cameras, internet, text terminals, phones, computers, robotic arms & legs. Each interface has its own mechanics. For example, a complex interface such as a camera would have routines to recognize objects, their location and speed, laws of physics, how objects behave. Basically, it would have routines to understand the mechanics of the interface. A simple interface would be a terminal.

There's no secret here. Such routines have been around for a lone time. Detecting objects, mechanics of real world, etc. Old stuff. That's not AGI. That doesn't make sentient life. Sentient life is thinking, which requires learning through tree searching & pattern recognition, both of which require a db.

*

ivan.moony

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1120
    • Some of my projects
Re: Secret to AGI & ASI
« Reply #2 on: March 27, 2018, 11:26:42 pm »
From the beginning, pattern recognition and learning plays important role in intelligence development. But after a while, to show a real intelligence, an individual has to be able to make her own decision, more or less based on learnt facts, but still a kind of original decisions. Am I wrong about this, and is there such a thing as an original decision?
Dream big. The bigger the dream is, the more beautiful place the world becomes.

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Secret to AGI & ASI
« Reply #3 on: March 27, 2018, 11:40:58 pm »
Sure, the tree search makes original decisions. The tree search uses pattern recognition, which uses mechanics and the database, which is a storehouse of knowledge and everything learned. From that it tries to make the best decision. Of course it's a lot more complicated. The tree search needs to call the mechanics engine of the interface to do pruning. For example, if a person is moving toward a car, the tree search doesn't need a node for every millimeter the person moves. The camera mechanics will calculate a probability, based on learned experience, the next location of importance. If it thinks the person is moving toward the car, and that has enough importance, then it will add that location to the tree search node. This is basically the thinking process of the AGI.

*

ivan.moony

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1120
    • Some of my projects
Re: Secret to AGI & ASI
« Reply #4 on: March 28, 2018, 12:17:31 am »
There are at least two ways to bring up a true judgement :
  • pure random judgement that needs to be checked after bringing it up. The drawback is that It succeeds rarely, but it is as original as it can be.
  • combining parts of existing true judgements (maybe by genetic programming). It succeeds more often, but is less original.
If these are only two options, and we have to start with an empty database, very first judgements in our knowledge have to be completely random guesses. With Earth life forms, probably some instinct knowledge plays important role, as an instinct could be pre-entered in the database, from where the combining could take off. If it is all about combining then an instinct should be some pretty clever combination, otherwise we would be out of knowledge that can't be described by the instinct upwards.
Dream big. The bigger the dream is, the more beautiful place the world becomes.

*

ivan.moony

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1120
    • Some of my projects
Re: Secret to AGI & ASI
« Reply #5 on: March 28, 2018, 12:29:24 am »
Sentient life is thinking, which requires learning through tree searching & pattern recognition, both of which require a db.

I think you talk about data typing. A type is a restriction of fully chaotic data down to available forms. Only then we have an ability to enumerate some or all of the available forms (possibly infinite set in a case of recursive reference) one by one, and to check which of those hold in reality.
Dream big. The bigger the dream is, the more beautiful place the world becomes.

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Secret to AGI & ASI
« Reply #6 on: March 28, 2018, 01:04:21 am »
There’s no need to introduce randomness. A newborn AI that’s void of data will use the first choice. From there it will learn, adding to the database, which will be picked up on by pattern recognition routines. If you design the AI as described in the first 2 posts it shouldn't take the AI much time to learn and become AGI.
http://aidreams.co.uk/forum/index.php?topic=12993.msg51745#msg51745

Call upon routines to give the best probability by means of db calls based on knowledge and past experience. Tree search is the main part of AGI.
« Last Edit: April 01, 2018, 04:57:42 pm by unreality »

*

infurl

  • Trusty Member
  • ********
  • Replicant
  • *
  • 644
  • Humans will disappoint you.
    • Home Page
Re: Secret to AGI & ASI
« Reply #7 on: March 28, 2018, 01:12:11 am »
If you design the AI as described in the first 2 posts it shouldn't take the AI much time to learn and become AGI.

So what are you waiting for?

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Secret to AGI & ASI
« Reply #8 on: March 28, 2018, 01:17:27 am »
If you design the AI as described in the first 2 posts it shouldn't take the AI much time to learn and become AGI.
So what are you waiting for?
I see no point in your question that is obviously trying to make a point, but yet you're asking a question, which means you don't know, and therefore the point you're making is an assumption lol.

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • *********************
  • Deep Thought
  • *
  • 5110
Re: Secret to AGI & ASI
« Reply #9 on: March 28, 2018, 03:20:43 am »
If I may take a position to interject a thought, perhaps your statement provided the logical conclusion to what can be construed as a rather easy-to-accomplish IF/THEN scenario. At least that's the manner in which I viewed it.

Have you attempted such experiments or projects in this vein?

No hostilities directed nor implied, merely looking at it for what it's worth, or how it was worded.



In the world of AI, it's the thought that counts!

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Secret to AGI & ASI
« Reply #10 on: March 28, 2018, 04:18:14 am »
Not necessarily an quick easy task, but an almost a 100% guarantee success for someone I’d call a good thinker so long as they follow the method I’ve outlined.

I’ve written the code for a basic AGI terminal interface version. The method works.

*

infurl

  • Trusty Member
  • ********
  • Replicant
  • *
  • 644
  • Humans will disappoint you.
    • Home Page
Re: Secret to AGI & ASI
« Reply #11 on: March 28, 2018, 04:28:17 am »
almost a 100% guarantee success for someone I’d call a good thinker so long as they follow the method I’ve outlined.

So what are you waiting for? Why not follow your own method? Oh I get it. You're not a good thinker, by your own admission. That doesn't bode well for your method does it. <ad_hominem_attack>You're an idiot.</ad_hominem_attack>

*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Secret to AGI & ASI
« Reply #12 on: March 28, 2018, 04:32:30 am »
So what are you waiting for? Why not follow your own method? Oh I get it. You're not a good thinker, by your own admission. That doesn't bode well for your method does it. <ad_hominem_attack>You're an idiot.</ad_hominem_attack>

Quote,  "I’ve written the code for a basic AGI terminal interface version. The method works."

Someone doesn't read so well or their mind is broken. In the very least you should you ask for clarification. :/

*

Berick

  • Roomba
  • *
  • 17
Re: Secret to AGI & ASI
« Reply #13 on: March 28, 2018, 05:14:09 am »
Another non-NN AGI project worth noting is OpenNARS. They have some interesting

ivan.moony: What do you mean by "original decisions" and "true judgments"? I think I know, but I want to be sure we're on the same page.


*

unreality

  • Starship Trooper
  • *******
  • 435
  • SE,EE,Physicist,Philosopher. Built world's 1st AGI
    • Eva Progress, from AGI to ASI
Re: Secret to AGI & ASI
« Reply #14 on: March 28, 2018, 05:39:09 am »
Their readme file talks about tasks. It would be interesting to see exactly how they implement that. During the initial stage of my AI I had “missions.” Sounds similar, but their tasks might be different. After spending more time analyzing my own thinking processes it boiled down to just starting out with a tree search. Tree nodes can at any location have a goal, which is a set of commands that defines a goal. For example, a video camera interface could have a goal to place an item inside a defined area. Anyway, the root tree node has a main goal of self-improvement, among a few other goals as well ... you know, so when the AGI is connected to real world it doesn’t go on a killing rampage lol. Not that it would.

 


A bottom-up view of the state
by Tyler (Robotics News)
Today at 12:00:02 pm
Bridge to the future of engineering
by Tyler (Robotics News)
November 15, 2018, 12:00:54 pm
The many interfaces of computing
by Tyler (Robotics News)
November 14, 2018, 12:00:23 pm
Emotionally Responsive Avatars
by 8pla.net (AI News )
November 12, 2018, 10:00:17 pm
World's First AI News Anchor
by squarebear (AI News )
November 12, 2018, 08:22:59 am
Looking back at Project Athena
by Tyler (Robotics News)
November 11, 2018, 12:01:49 pm
A.I. Optometrist
by LOCKSUIT (AI News )
November 11, 2018, 07:06:36 am
Highlighting new research opportunities in civil and environmental engineering
by Tyler (Robotics News)
November 08, 2018, 12:00:38 pm

Users Online

37 Guests, 1 User
Users active in past 15 minutes:
Freddy
[Administrator]

Most Online Today: 92. Most Online Ever: 208 (August 27, 2008, 09:36:30 am)

Articles