Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: HS on July 18, 2018, 04:33:52 am
-
Humans seem to posses a few innate abilities. We use these with varying degrees of skill, and this inbuilt hardware, these tools, can be rickety 5 dollar pliers or expensive seimens surgical pliers. But they are both limited to grasping concepts in an identical way. The only difference is that it's less work for the good quality pliers.
You could imagine an intelligence as a toolbox. Each species could be defined by having the same tools in it's starter kit among healthy individuals. Then they can then mod them (up to a point) to do better in their specific circumstances. Humans and dogs presumably have an overlapping ven diagram of basic tools. The ven diagram of two dogs would be superimposed. Each species would posses the required tools to survive the universe in their unique way.
Some of them are probably super simple and super clever, tools that wouldn't occur to human inventors. As demonstrated by the fact that we are forced to call these things "tools" or "algorithms" because we are very into the whole tool using thing, and as such can only grasp this concept using a "tool" metaphor.
Get this though! Maybe we don't have to invent a lot of this stuff. We could study the behavior of any organism and copy basic traits with code or machine learning. With enough examples we could compile our own starter kits. I suspect it would be like building a steam engine and lighting a fire. If you do the setup correctly the rest will become self propelling. You don't have to make this crazy smart skilled robot, you just have to make something whole. A very broad foundation builds the tallest tower, yes? no? What do you think?
-
Good ideas keep em coming. Hm, so let's take a organism ability to react to being touched or its ability to transform into a long shape...The best thing I'd say here is we must build AGI, a human body, with what we know we do/have ourselves, and that learning many facts is the groundwork our smarts in any kid is based upon. Each fact is a mental tool, in ajunction to your thread title lol.
-
"Hopefully maybe something", u may not know it, but u might like this idea, because your idea is neural network based.
My opinion will never change from now on, until i proove it false... that, if you have a rule (could be a tool? if its like what ur saying), combined with a neural network or any kind of search algorythm based method, this will bind with the environment in a way to produce all sorts of novel ways to achieve it. the environment will complicate the basic rule, into complex behaviours.
But thinking that, because its doubtish, takes belief, it may be true, it may not be true, until someone prooves it for real.
Some people say, all the robot needs to do is "reach harmony with the environment" so it is "reduce all creatures pain and its own" and if it has this one simple law (which u still have detect somehow. and might be tricky, and it is scored spacially every update frame of the ai system.) then this involves the robot doing all sorts of complex things to achieve it. (even building a car, if you were optimistic!)
If im right about this, we will behold the new computer robot consciousness/awareness, as the algorythm.
-
Each fact is a mental tool
I think we need existing structures to channel facts in productive ways. The nature of these structures could be found by studying broad behavior patterns. Facts are good, but you need both fuel and oxygen to keep a fire going. Pure info makes a hard drive, (not sentient on it's own). In a hard drive all this information has the potential to react with something, but additional fact trees will not bring it to life, probably. At the ground level we need some basic drivers whose job is to be influenced and slowly adjusted by incoming facts.
-
"Hopefully maybe something", u may not know it, but u might like my idea, because your idea is neural network based.
My opinion will never change from now on, until i proove it false... that, if you have a rule (could be a tool? if its like what ur saying), combined with a neural network or any kind of search algorythm based method, this will bind with the environment in a way to produce all sorts of novel ways to achieve it. the environment will complicate the basic rule, into complex behaviours.
But thinking that, because its doubtish, takes belief, it may be true, it may not be true, until someone prooves it for real.
Some people say, all the robot needs to do is "reach harmony with the environment" so it "reduces creatures pain and its own" and if it has this one simple law (which u still have detect somehow. and might be tricky, and it scored spacially every update frame of the ai system.) then this involve the robot doing all sorts of complex things to achieve it. (even building a car, if you were optimistic!)
If im right about this, we will behold the new computer robot consciousness/awareness, as the algorythm.
YES !!! This is also what I think we need.
-
Theres a really easy and practical rule for a safety robot, and its just avoid moments of harsh accelleration, and that is easily measurable with simple calculus derivatives.
If it could see bullets fire out of guns, it would hate them a high scoring amount! :2funny:
-
Theres a really easy rule for a safety robot, and its just avoid moments of harsh accelleration, and that is easily measurable with calculus.
If it could see bullets fire out of guns, it would hate them a high scoring amount! :2funny:
Makes sense. I'll try to remember that, famous last worlds lol. If I build a robot it will be Wall-E shaped, such a good simple design. Very expressive too, that thing can communicate things with just a few movements.
-
OK
Can we come up with some rules/motivations to direct a neural network?
The neural net can learn facts but we need rules to steer it in helpful directions for an organism. Help it to develop a general functionality. Later if a need is detected, (as it should be if good rule/motivator mechanisms are added at the beginning) these drivers can begin constructing more esoteric skills. Maybe the drivers would work best and be easiest to make in computer form, and the net would work best as a physical structure.
-
Rules can be for internal structure development as well, because the robots memory of facts needs to be infallible if its a good one. So in other words there could be rules to split the objects apart from each other at the sensor base, and it converts into a different set of semantics/at least groups. and also to find objects that have equal effect in a situation, become the same "cell". Its no good if its database is higglety put together because it wont be able to represent states as well.
Really neat structure, is a good idea to give you the best chance.
As far as the machine/body goes, a simple design thats similar to a electromechanical pin ball machine could be as complicated as it needs to be. And there is pinball machine designs that are simpler than others. So I like REALLY simple, when I go to do it, no arduino, (look mum no computer. :))
I have a personal solid-state, electronic design paradigm, involving oscillating capacitors alone, its just an rc circuit (resisor capacitor) i dont even use diodes, and my oscillator doesnt even use a transistor, but i have to get it going first. I actually have only lazily completed a mechanical oscillator only. (they are easy and fun to make, but they only hit 20hz tops usually... because they end up breaking themselves if they hz too quick because they are physical.)
but thats all crappy details. its how few parts come together to make something work, the concept is the most important thing, a robot comes down to so many automatic faders, and easiest way to get it, do it.
-
Sorry if I sound like im trivializing the situation, im just a prankster on the internet, i dont ever show my true self hardly ever... i just joke around in such a fashion.
But seriously...
If u want to go down my alley, make markov chain video. I recommend you do it serially against the rendering surface, because then you get less spurious states. if you do it parallel, it lasts 10 frames before it disintigrates... even serially it holds together better but itll drop out at random.
And even if you get to that point, you need 1000000 simulations going at once for your population, which it then rehillclimbs off its last point, learning off the video memory response.
So its hard to scrum up the processing power required in your "designated design space", being your light bulb mesh, mines different its more of an ultra parallel ternary cam router. (better than a gpu)
After you get over that, its all downhill from there.
Thats how I plan on doing it. Its not easy at all, but broadly conceptually its simple.
So do I expect people to give me respect for doing this stuff... hell no, i expect complete ignorance, so then i start stuffing around completely. :uglystupid2:
-
No worries, too complex if anything. I'm having to go research stuff on Wikipedia. I guess I prefer messy systems where probability is on your side. Lets keep working on both our approaches, I'll post again if I have some inspiration or (god forbid) make actual progress :)
-
OK
Can we come up with some rules/motivations to direct a neural network?
The neural net can learn facts but we need rules to steer it in helpful directions for an organism. Help it to develop a general functionality. Later if a need is detected, (as it should be if good rule/motivator mechanisms are added at the beginning) these drivers can begin constructing more esoteric skills. Maybe the drivers would work best and be easiest to make in computer form, and the net would work best as a physical structure.
It can be interesting "biasing";
The approaches and reasoning behind biasing is often not truly explained... When a neural network has been fully trained .... adding biasing after is like an organic thought. one day you think this idea is correct and others not so much.... fluctuating biasing based on "Emotion?" could give a more accurate "human type response" where we actually make mistakes or slight errors even though our understanding and muscle memory is intact... too many people talking can biased your thinking ... yet when surrounded by silence your thinking is focused.
Mistakes are a human traite as a robot could not make mistakes if fully trained... there fore it could not pass the Turin test.... as mistakes should be a part of determining if "one is speaking to a human"
-
I tend to agree with the original post, intelligence as a toolbox. Human mind has evolved against prehistoric challenges: finding food, hunting, escaping predators, ...etc. These challenges have shaped the way we see things.
It's so easy to set up a chatbot nowadays (notice I didn't say a good chatbot...), I think that's the way to go. Thinkbots. I'm working on a simple system currently, that brings together several excellent "low-tech" existing libraries: a chatbot engine (Rivescript), an in-memory search engine (Lunr), a logic engine (Logicjs), a limited NLP library (Compromise), a behavior/utility tree (homebrew), and an ECS-based game engine (Crafty), in a wrapper of Goldenlayout and probably T3js. I'll try to integrate these simple tools in a single easy framework that would allow users to create offline modular Ais, with sharable parts. I have a nice avatar already, see pic.
-
@Hopefully Something,
read this book by Jeff Hawkins - On Intelligence
It is a true AGI book
-
@Hopefully Something,
read this book by Jeff Hawkins - On Intelligence
It is a true AGI book
Cool, thanks!
-
Ranch will like this one, the leafs adjust a bit:
First, the appropriate
memory is automatically recalled by the sight of the ball. Second, the memory
actually recalls a temporal sequence of muscle commands. And third, the retrieved
memory is adjusted as it is recalled to accommodate the particulars of the
moment, such as the ball's actual path and the position of your body. The memory
of how to catch a ball was not programmed into your brain; it was learned over
years of repetitive practice, and it is stored, not calculated, in your neurons.
-
Ranch will really like this part from that INCREDIBLELY good book from Jeff Hawkins:
I like how it says a adjustment is needed and the commader above knows what to send back down. Read it if you'rre keen on ignoring the book. It's mostly cus ranch said he liked it I'm posting this.
---The way you memorize sequences and represent them by name as information goes up and down your cortical hierarchy may remind you of the hierarchy of military command. The top army general says, "Move the troops to Florida for the winter." This simple high-level command gets unfolded into ever more detailed sequences of commands as it percolates down the hierarchy. The general's underlings recognize that the command requires a sequence of steps such as preparations to leave, transportation to Florida, and preparations for arrival. Each of these steps breaks down into further, more specific steps, to be carried out by subordinates. At the bottom there are thousands of privates taking tens of thousands of actions that result in the troops moving. Reports of what happened are generated at each level. As they percolate back up the hierarchy, they are summarized again and again, until at the very top of the hierarchy, the general
receives a daily briefing saying, "Move to Florida going okay." The general does not get all the details.
---There is an exception to this rule. If something goes wrong that cannot be handled by subordinates down the chain of command, then the issue rises up the hierarchy until someone knows what to do next. The officer who does know how to handle the situation does not see it as an exception. What was an unanticipated problem to subordinates is just the expected next task on his list. The officer then issues new commands to subordinates. The neocortex behaves similarly. As we will see in a bit, when events (in other words, patterns) occur that aren't anticipated, information about them progresses up the cortical hierarchy until some region can handle it. If lower regions of cortex fail to predict what patterns they are seeing, they consider this an error and pass the error up the hierarchy. This is repeated until some region does anticipate the pattern.
-
Interestingly, this is also the description of a typical async program execution.