Re: [agi] Architecture 101 - Units of Intelligence

[Stanley Nilsen]

Peter, thanks for the links to several thought provoking texts. I'm responding to the comment "Bots aren’t the way to intelligent systems. "  I'm not defending bots, but explaining why I would consider building one.  ... I am not impressed with the way bots are implemented, or the results. There seem to be lots of shortcuts taken for the sake of trying to be first to pass the Turing test. I think bots would be an interesting mechanism for AI experimentation for the following reasons: First, it is easy to provide a bot with many different possible phrases. And, choosing the phrase to use for the given moment is an exercise in making choices. Making better choices is what intelligence is about. Second, there is a “bootstrap” aspect to coming up with a bot. What will the bot perceive as the current situation? What are the situational factors that a bot might detect, and using what method? How can situational factors be used and what type of situations do they describe? I'm not interested in bots “acting” like a person, but rather with how the bot will bring benefit. Not the benefit of being an assistant like Alexa, but more like a companion that might respond to what it thinks my current frame of mind is. If it knows my current state of mind, what should it do with it? If the bot meets (speaks with) a new person, how might it extract out significant information about the person. What are the better phrases to choose and what will they yield? Could a bot develop a bond with the one who converses with it? What is a bond? Is it a loyalty? Is it common humor… Clearly I'm not reaching toward a commercial application of bot technology. I resent speaking with bots on the telephone and am easily frustrated with “bot” help found on websites… My main interest is in the architecture of a machine that can handle thousands of possible scenarios and provide fast response. Hey, self driving cars would be fun, but I'm afraid that's a bit too expensive and complex for my budget of both money and time. Bots might be better suit me for a test implementation of an “opportunity” centered system. Stan On 08/31/2018 12:47 PM, pe...@optimal.org wrote: Bots aren’t the way to intelligent systems. Here are some pointers:   https://medium.com/@petervoss/cant-the-big-players-just-do-what-you-re-doing-2d2eb6fcd85c   https://becominghuman.ai/the-third-wave-of-ai-1579ea97210b   and  https://medium.com/@petervoss/my-ai-articles-f154c5adfd37   From: Stanley Nilsen <senil...@ghvalley.net> Sent: Friday, August 31, 2018 11:10 AM To: Stefan Reich via AGI <agi@agi.topicbox.com> Subject: Re: [agi] Architecture 101 - Units of Intelligence   Rules and facts are the working materials of whatever we construct.  To think about and talk about AI concepts the rules and facts are low level.  As one implements, rules and facts take center stage.  If I were to ask "what do the rules do?" One would likely answer "everything." By the say Stefan, have you checked out the Chatbots.org forums?  They have an AI Zone forum that I look at occasionally.  ( https://www.chatbots.org/ai_zone/ )  I like the idea of using a conversational bot as a platform to try out AI ideas, but I haven't produced a bot yet. I've used "Clips" as a programming environment when I want to experiment with behavior that involves lots of rules.  It's pretty well "old school" but not that hard to learn. Stan On 08/31/2018 09:56 AM, Stefan Reich via AGI wrote: Well... my AI system for one contains *rules* and *facts*.   Facts are just English sentences.   Rules are basically natural language plus certain operators plus embedded code fragments. Rules can process as well as generate user input, bot output & facts.   There are also meta rules and facts (those which talk about other rules and facts).   Stanley Nilsen <senil...@ghvalley.net> schrieb am Fr., 31. Aug. 2018 04:59: A few thoughts...   If an intelligent device is to be planned and built according to a modular architecture, it would be appropriate to determine what the “storage” of the intelligence looks like. What will the device contain that makes it intelligent? Are these “blocks” of intelligence measurable? Can this intelligence be wired into the unit easily?   Part of the reason for looking at this essence of intelligence is to better see what needs to surround the intelligence to have a complete system. In my thinking (over the past few years,) I've come to favor a way of looking at intelligence that can easily lead to a fairly complete “generally” intelligent device.   Intelligence Stuff The stuff that goes into intelligence is knowledge, but knowledge alone isn't enough. For knowledge to contribute to intelligence, it has to be in a “package” consisting of several pieces of knowledge, and physical accessories, which all relate to one another in a specific way. To refer to this package, I call it “opportunity.”   People have a common understanding of opportunity, and it fits nicely into the conversation about intelligence. Consider that an opportunity is related to a benefit, and one's situation. This implies that we can “choose” an opportunity when it is likely to bring benefit. The idea being that we make a good “choice” given the situation – doing the intelligent thing.   You may object to giving opportunity a “high status” because awareness of an opportunity doesn't give us that opportunity. And I would be the first to agree. But, I would say that one only has real opportunity when he can execute a recipe that will bring benefit. In other words, you only have opportunity if you have the whole package, and the opportunity package (the unit of intelligence) includes being able to take action that causes changes (benefits) in the real world. This taking action is what makes intelligence much more than intellectual exercise.   In the view from 50,000 feet, the mature intelligent device has hundreds of opportunity to choose from in a given moment. It's overall “score” on the intelligence scale is determined by both it's quantity of opportunity and the quality of each opportunity. Cloning intelligence would mean giving the less mature unit opportunities that it can add to it's store of opportunity.   There is much more that can be said about architecture centered around opportunity. I would be happy to engage in a discussion if anyone is interested.   Stan     Artificial General Intelligence List / AGI / see discussions + participants + delivery options Permalink