Convention is a language. The program has to find a way to understand what people say, and not say. It has to be able to learn the deeper meaning within the human conversation, and systemically get to the heart of every matter. It has to do this in the most-true manner, utilizing evidence-based objectivity where possible. That method exists. It's the one I shared here. That's exactly what it does. What it needs to become automated is the development of its own GUI, as translator. ________________________________ From: Jim Bromer via AGI <[email protected]> Sent: Saturday, 23 June 2018 3:55 PM To: AGI Subject: Re: [agi] Discrete Methods are Not the Same as Logic
I am thinking of a program that would learn by communicating with people using language. It would learn from interacting with people. The problem with that strategy is that it would tend to acquire superficial knowledge. It would however be required to do some true learning. One reason is that a person cannot think of all the relations and implicit categories that an intelligent entity would have to rely on. Secomdly, we cannot, at this time, understand all the sorts of a knowledge items that it would need to gain greater understanding. It would not be given predetermined categories other than some default second level abstract categories. These second level abstractions might be concerned with abstractions of relations found in discrete relationships that would be expected to found in networks of related information. It would have to work around the complexities that might develop. I am not talking about pure logic but discrete learning so the np problem is not a problem. The "discrete networks" would also include weighted reasoning of course. I am just saying that weighted reasoning isn't necessary but that discrete learning, learning by using ideas and developing principles of thought is important. But I have to be able to develop this as an extremely simple programming project that will quickly show some simple results (like feasibility tests) or else I am not going to have anything to start with. Jim Bromer On Sat, Jun 23, 2018 at 8:51 AM, Nanograte Knowledge Technologies via AGI <[email protected]> wrote: > Jim > > I agree with making things simple, but one should not make it more simple > than necessary. Any algorithm relying on deabstraction to provide proof of > true learning would be highly complex. There's no simple solution to that > problem. However, I'm enjoying your sentiment how, within deabstraction, > even complexity should become relative over time. Maybe one day, the machine > would've learned how to invent deabstraction algorithms until it became a > simple matter of instinct. > > Since when do human beings discover all its learning by itself? That's a > fallacy. An AGI platform also does not have to discover all of its learning > by itself. It can be taught until such time it can learn how to organize > resources in order to teach itself and learn via reflection. > > Rob > ________________________________ > From: Jim Bromer via AGI <[email protected]> > Sent: Saturday, 23 June 2018 2:41 AM > > To: AGI > Subject: Re: [agi] Discrete Methods are Not the Same as Logic > > Maybe I should use a name different than judgement. Reflection? > Insightful reflection. The depth of the insight would be relative to > how much knowledge, related to the questions being examined, was > available. So in the primitive model this insight would not be very > good and the program would have to be dependent on what the teacher > could convey to it. But insight would have to be based on putting > different kinds of information together. Novel insight might be > reinforced simply by being in the ballpark, it would not have to be > perfect as long as it was tagging along somewhere within the subject > matter being discussed, described or within the boundaries of > understanding something about a situation that was occurring. I think > different agi's would have to be different if they were thinking for > themselves - to some extent. > Jim Bromer > > > On Fri, Jun 22, 2018 at 3:10 PM, Mike Archbold via AGI > <[email protected]> wrote: >> Judgments are fascinating. It seems like most approaches are some >> variation of reinforcement learning. What have you got in mind? One >> thought from Hegel which always sticks in my mind is that a "judgment >> could be other than what it is." So just think about that last >> sentence. How on earth could anyone automate that? But, more so, two >> distinct AGI's would always be different on that account. >> >> On 6/22/18, Jim Bromer via AGI <[email protected]> wrote: >>> I need to start with something that is extremely simple and which will >>> produce some kind of result pretty quickly. I have had various ideas >>> about it for some time but what I see now is that a necessary >>> advancement for AI would have to exhibit some kind of judgment about >>> what it learns about. I realized the importance of making a program >>> that could learn new ways of thinking. Since I believe that >>> categorical reasoning is important then that means that it would not >>> only have to use abstractions but it would also have to be able to >>> discover abstractions of its own. This does not seem too difficult >>> because I am not being unreasonable about requiring it to be a >>> historical singularity inflexion point. I need to start with >>> something simple that demonstrates an ability for true learning. What >>> I see now is that it also has to exhibit some kind of simple >>> judgement. I need to come up with simple judgement algorithms. I >>> cannot get started unless I can come up with simple feasible models >>> that I can test. >>> I respectfully disagree with you about one thing. The elaboration of >>> an extensive framework and management system is, in my opinion, a >>> waste of time. It is like planning an AI program that will create AGI >>> for you completely on its own. It might be ok to think about such a >>> thing but it is nowhere to start out for an actual programming >>> project. I have to start with something that is very simple and which >>> can show some immediate results. For me, simplification is a necessity >>> but it is also necessary to avoid the wrong kinds of simplification. >>> Jim Bromer >>> >>> >>> On Fri, Jun 22, 2018 at 12:13 AM, Nanograte Knowledge Technologies via >>> AGI <[email protected]> wrote: >>>> Jim, I think for this kind of reasoning to evolve, one would always have >>>> to >>>> return to an ontological platform. For example, for reasoning, one would >>>> require a meta-methodology for reasoning effectively with. For >>>> selectively >>>> forgetting and learning, an evolution-based methodology is required. For >>>> managing Logic, one would need a suitable framework and management >>>> system, >>>> and so on. These are all critical components, or nodes, that would have >>>> to >>>> exist for self-optimized reasoning functionality to become >>>> spontaneous.The >>>> real IP lie not only in the methods, in the sense of AI apps. >>>> >>>> Yuu stated: "...DL story is compelling it is not paying out to stronger >>>> AI >>>> (Near AGI)..." >>>>>>>Is it possible that AGI is an outcome, an act of becoming, and not a >>>>>>> discrete objective at all? >>>> >>>> Rob >>>> ________________________________ >>>> From: Jim Bromer via AGI <[email protected]> >>>> Sent: Thursday, 21 June 2018 5:20 PM >>>> To: AGI >>>> Subject: Re: [agi] Discrete Methods are Not the Same as Logic >>>> >>>> Symbol Based Reasoning is discrete, but a computer can use discrete >>>> data that would not make sense to us so the term symbolic might be >>>> misleading. I am not opposed to weighted reasoning (like neural >>>> networks or Bayesian Networks) and I think reasoning has to use >>>> networks of relations. If weighted networks can be thought of as a >>>> symbolic network then that suggests that symbols may not be discrete >>>> (as different from Neural Networks.) I just think that there is >>>> something missing with DL, and while the Hinton...DL story is >>>> compelling it is not paying out to stronger AI (Near AGI). For >>>> example, I think that symbolic reasoning which is able to change its >>>> categorical bases of reasoning is something that is badly lacking in >>>> Discrete Learning. You don't want your program to forget everything it >>>> has learned just because some doofus tells it to, and you do not want >>>> it to write over the most effective methods it uses to learn just to >>>> deal with some new method of learning. So, that, in my opinion is >>>> where the secret may have been hiding. A program that is capable of >>>> learning something new must be capable of losing its more primitive >>>> learning techniques without wiping out the good stuff that it had >>>> previously acquired. This requires some working wisdom. >>>> I have been thinking about these ideas for a long time but now I feel >>>> that I have a better understanding of how this insight might be used >>>> to point to simple jumping off point. >>>> Jim Bromer >>>> >>>> >>>> On Thu, Jun 21, 2018 at 2:48 AM, Mike Archbold via AGI >>>> <[email protected]> wrote: >>>>> So, by "discrete reasoning" I think you kind of mean more or less "not >>>>> neural networks" or I think some people say, or used to say NOT "soft >>>>> computing" to mean, oh hell!, we aren't really sure how it works, or >>>>> we can't create what looks like a clear, more or less deterministic >>>>> program like in the old days etc.... Really, the challenge a lot of >>>>> people, myself included, have taken up is how to fuse discrete (I >>>>> simply call it "symbolic", although nn have symbols, typically you >>>>> don't see them except as input and output) and DL which is such a good >>>>> way to approach combinatorial explosion. >>>>> >>>>> To me reasoning is mostly conscious, and kind of like the way an >>>>> expert system chains, logically. The understanding is something else >>>>> riding kind of below it and less conscious but it has all the common >>>>> sense rules of reality which constrain the upper level reasoning which >>>>> I think is logical, like "if car won't start battery is dead" would be >>>>> the conscious part but the understanding would include such mundane >>>>> details as "a car has one battery" and "you can see the car but it is >>>>> in space which is not the same thing as you" and "if you turn around >>>>> to look at the battery the car is still there" and all such details >>>>> which lead to an understanding. But understanding is an incredibly >>>>> tough thing to make a science out of, although I see papers lately and >>>>> conference topics on it. >>>>> >>>>> On 6/20/18, Jim Bromer via AGI <[email protected]> wrote: >>>>>> I was just reading something about the strong disconnect between our >>>>>> actions and our thoughts about the principles and reasons we use to >>>>>> describe why we react the way we do. This may be so, but this does not >>>>>> show >>>>>> how we come to understand basic ideas about the world. This attempt to >>>>>> make >>>>>> a nearly total disconnect between reasons and our actual reactions >>>>>> misses >>>>>> something when it comes to explaining how we know anything, including >>>>>> how >>>>>> we learn to make decisions about something. One way to get around this >>>>>> problem is to say that it all takes place in neural networks which are >>>>>> not >>>>>> open to insight about the details. But there is another explanation >>>>>> which >>>>>> credits discrete reasoning with the ability to provide insight and >>>>>> direction and that is we are not able to consciously analyze all the >>>>>> different events that are occurring at a moment and so we probably are >>>>>> reacting to many different events which we could discuss as discrete >>>>>> events >>>>>> if we had the luxury to have them all brought to our conscious >>>>>> attention. >>>>>> So logic and personal principles are ideals which we can use to >>>>>> examine >>>>>> our >>>>>> reactions - and our insights - about the what is going on around us >>>>>> but >>>>>> it >>>>>> is unlikely that we can catalogue all the events that surround us and >>>>>> (partly) cause us to react the way we do. >>>>>> >>>>>> Jim Bromer >>>>>> >>>>>> On Wed, Jun 20, 2018 at 6:06 AM, Nanograte Knowledge Technologies via >>>>>> AGI >>>>>> < >>>>>> [email protected]> wrote: >>>>>> >>>>>>> "As Julian Jaynes put it in his iconic book *The Origin of >>>>>>> Consciousness >>>>>>> in the Breakdown of the Bicameral Mind* >>>>>>> >>>>>>> Reasoning and logic are to each other as health is to medicine, or — >>>>>>> better — as conduct is to morality. Reasoning refers to a gamut of >>>>>>> natural >>>>>>> thought processes in the everyday world. Logic is how we ought to >>>>>>> think >>>>>>> if >>>>>>> objective truth is our goal — and the everyday world is very little >>>>>>> concerned with objective truth. Logic is the science of the >>>>>>> justification >>>>>>> of conclusions we have reached by natural reasoning. My point here is >>>>>>> that, >>>>>>> for such natural reasoning to occur, consciousness is not necessary. >>>>>>> The >>>>>>> very reason we need logic at all is because most reasoning is not >>>>>>> conscious >>>>>>> at all." >>>>>>> >>>>>>> https://cameroncounts.wordpress.com/2010/01/03/mathematics-and-logic/ >>>>>>> >>>>>>> >>>>>>> >>>>>>> <https://cameroncounts.wordpress.com/2010/01/03/mathematics-and-logic/> >>>>>>> Mathematics and logic | Peter Cameron's Blog >>>>>>> >>>>>>> <https://cameroncounts.wordpress.com/2010/01/03/mathematics-and-logic/> >>>>>>> Apologies: this will be a long post, and there will be more to come. >>>>>>> But >>>>>>> it may be useful to you if you are getting to grips with logic: I >>>>>>> have >>>>>>> tried to keep the overall picture in view. >>>>>>> cameroncounts.wordpress.com >>>>>>> >>>>>>> >>>>>>> ------------------------------ >>>>>>> *From:* Jim Bromer via AGI <[email protected]> >>>>>>> *Sent:* Wednesday, 20 June 2018 12:01 PM >>>>>>> *To:* AGI >>>>>>> *Subject:* Re: [agi] Discrete Methods are Not the Same as Logic >>>>>>> >>>>>>> Discrete statements are used in programming languages. So a symbol (a >>>>>>> symbol phrase or sentence) can be used to represent both data and >>>>>>> programming actions. Discrete Reasoning might be compared to >>>>>>> something >>>>>>> that has the potential to be more like an algorithm. (Of course, >>>>>>> operational statements may be retained as data which can be run when >>>>>>> needed) >>>>>>> For an example of the value of Discrete Methods, let's suppose >>>>>>> someone >>>>>>> wanted more control over a neural network. Trying to look for logic >>>>>>> in >>>>>>> a neural network does not really make all that much sense if you want >>>>>>> to find relationships between actions on the net and output. Using >>>>>>> Discrete Methods makes a lot of sense. You might want to try fiddling >>>>>>> with the weights of some of the nodes as the nn is running. If >>>>>>> certain >>>>>>> effects can be described (or sensed by some algorithm) then >>>>>>> describing >>>>>>> what was done and what effects were observed would be the next step >>>>>>> in >>>>>>> the research. Researchers are not usually able to start with detailed >>>>>>> knowledge of exactly what is going on. So they need to start with >>>>>>> descriptions of some actions they took and of what effects were >>>>>>> observed. If these actions and effects can be categorized in some way >>>>>>> then the chance that more effective observations will be obtained >>>>>>> will >>>>>>> increase. >>>>>>> Jim Bromer >>>>>>> >>>>>>> >>>>>>> On Tue, Jun 19, 2018 at 11:12 PM, Mike Archbold via AGI >>>>>>> <[email protected]> wrote: >>>>>>> > It sounds like you need both for AI, certainly there is always a >>>>>>> > place >>>>>>> > for logic. What's "discrete reasoning"? >>>>>>> > >>>>>>> > On 6/18/18, Jim Bromer via AGI <[email protected]> wrote: >>>>>>> >> I am wondering about how Discrete Reasoning is different than >>>>>>> >> Logic. >>>>>>> >> I >>>>>>> >> assume that Discrete Reasoning could be described, modelled or >>>>>>> >> represented by Logic, but as a more practical method, logic would >>>>>>> >> be >>>>>>> >> a >>>>>>> >> tool to use with Discrete Reasoning rather than as a >>>>>>> >> representational >>>>>>> >> substrate. >>>>>>> >> >>>>>>> >> Discrete Reasons and Discrete Reasoning can have meaning over and >>>>>>> >> above the True False values of Logic (and the True False >>>>>>> >> Relationships >>>>>>> >> between combinations of Propositions.) >>>>>>> >> >>>>>>> >> Discrete Reasoning can have combinations that do not have a >>>>>>> >> meaning >>>>>>> >> or >>>>>>> >> which do not have a clear meaning. This is one of the most >>>>>>> >> important >>>>>>> >> distinctions. >>>>>>> >> >>>>>>> >> It can be used in various combinations of hierarchies and/or in >>>>>>> >> non-hierarchies. >>>>>>> >> >>>>>>> >> It can, for the most part, be used more freely with other >>>>>>> >> modelling >>>>>>> >> methods. >>>>>>> >> >>>>>>> >> Discrete Reasoning may be Context Sensitive in ways that produce >>>>>>> >> ambiguities, both useful and confusing. >>>>>>> >> >>>>>>> >> Discrete Reasoning can be Active. So a statement about some >>>>>>> >> subject >>>>>>> >> might, for one example, suggest that you should change your >>>>>>> >> thinking >>>>>>> >> about (or representation of) the subject in a way that goes beyond >>>>>>> >> some explicit propositional description about some object. >>>>>>> >> >>>>>>> >> You may be able to show that Logic can be used in a way to allow >>>>>>> >> for >>>>>>> >> all these effects, but I believe that there is a strong argument >>>>>>> >> for >>>>>>> >> focusing on Discrete Reasoning, as opposed to Logic, when you are >>>>>>> >> working directly on AI. >>>>>>> >> >>>>>>> >> Jim Bromer >>>>>>> *Artificial General Intelligence List >>>>>>> <https://agi.topicbox.com/latest>* >>>>>>> / AGI / see discussions <https://agi.topicbox.com/groups/agi> + >>>>>>> participants <https://agi.topicbox.com/groups/agi/members> + delivery >>>>>>> options <https://agi.topicbox.com/groups> Permalink >>>>>>> >>>>>>> >>>>>>> <https://agi.topicbox.com/groups/agi/Tcc2adcdd20e1add4-M155d4762ea9c7b0f14fefd47> >>>> Artificial General Intelligence List / AGI / see discussions + >>>> participants >>>> + delivery options Permalink > Artificial General Intelligence List / AGI / see discussions + participants > + delivery options Permalink ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tcc2adcdd20e1add4-M717b3fdac055bfb69bb56ef8 Delivery options: https://agi.topicbox.com/groups
