"Stop. Lose all scripted responses. Improvisation only" :D

  • 11 Replies
  • 424 Views
*

martinodb

  • Roomba
  • *
  • 7
"Stop. Lose all scripted responses. Improvisation only" :D
« on: March 02, 2019, 01:38:57 am »
Hello all!

I'm new to chatbots, but I've been interested in AI and related topics for a long time.  My main interest is in "GOFAI" areas like KRR, logic programming, linguistic theory.. not claiming to be an expert, it's just the kind of articles I enjoy reading the most ;)

 In fact, to be honest, I'm a bit miffed with so much modern emphasis on machine learning (neural networks and brute-force statistical approaches in general) for a number of reasons. Of course I respect the impressive results they are getting in pattern recognition and classification, but (from my short experience) it feels like a hands-off black box, where you don't have an easy way to change aspects you don't like, or adapt the system to a particular purpose you just came up with and no corpus of actual interactions is available. Anyway, sorry ML enthusiasts, I don't want to create a polemic on my first post, just state my preference!

The Westworld quote in the title describes what would be my dream AI chatbot. It would have the least possible amount of "smoke and mirrors", by which I mean clever tricks that make the bot seem far smarter than it is (I say "the least possible", I know we still need lots of it). Ideally, the bot admin wouldn't have to either write code or train the bot with a massive corpus. Instead, he/she would instruct the bot by telling it facts and rules in some very expressive CNL, and answering questions the bot comes up with.

I'm not asking for human-level AGI or anything so far-fetched, think something like Cyc, but with a robust, well supported,  open source version (they retired OpenCyc years ago) and a good NLP subsystem. A few years ago I played with Powerloom, which was nice, but it seems to be getting less and less traction, and the mailing list seems dead. Other open source real-life systems I've found inspiring are LangPro (a "natural logic" theorem prover written in Prolog, new and neat, but very limited for now) and  the ACE Reasoner from the Attempto project (alas, IIRC, it doesn't handle states, events, tense..only timeless facts). An insteresting reasoner I've been testing is CSNePS (written in Clojure, based on "concurrent Inference graphs"). It's still somewhat buggy but very promising as a foundation because it's expressive and scalable. My own little chatbot  can use different brains, and currently its main brain is based on Clara (a rules engine, also written in Clojure). Since this is my first post, I'm abstaining from posting any links, but of course I can elaborate if someone is interested, and I'll probably open related topics later on. Also I think the names and descriptions should be enough to Google them.

Sorry about that rant, I just think the best way for people to know where I'm coming from and maybe find common interests is to mention projects I've been checking out.

Looking forward to hearing from you guys. I hope all is well!

*

8pla.net

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1141
  • TV News. Pub. UAL (PhD). Robitron Mod. LPC Judge.
    • 8pla.net
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #1 on: March 02, 2019, 02:23:55 am »
It's all good, but designing smaller A.I. has a certain appeal.
My Very Enormous Monster Just Stopped Using Nine

*

martinodb

  • Roomba
  • *
  • 7
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #2 on: March 02, 2019, 02:46:43 am »
Hello!
Sure, everything must start small. I love small proof-of-concept AIs that demonstrate some interesting tech. Did I give the impression I only like "big" AI? Why, in what sense?

*

Hopefully Something

  • Trusty Member
  • *******
  • Starship Trooper
  • *
  • 287
  • So where are these cookies?
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #3 on: March 02, 2019, 03:59:41 am »
Hey There!

 :trophy_bronze: to the subject line (it's the grail). But how can this be accomplished? How can a chat bot improvise? We'll have to think outside the chat box  ;D, any ideas?

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • *********************
  • Deep Thought
  • *
  • 5398
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #4 on: March 02, 2019, 04:20:44 am »
Just tossing in my $.02

Perhaps such a bot would be given all the language and grammar usage "Rules" it might need and the programming to examine, sort and execute NLP, decision making on a best case scenario, given what it had learned or known...not from some canned responses.

This AI would be able to search like Google or Watson to find possible answers to questions but also be able to fashion the responses in a more tailor-made manner based on the user's needs.

A rule-based bot with the added ability to draw from a deterministic tree or weighted decision on that which is determined to be the best or most appropriate solution or answer would seem to improvise.

It would be able not only to provide verbal responses to questions or statements but to also carry on casual conversations with the user. A digital companion!

...and then I woke up... drat!!
In the world of AI, it's the thought that counts!

*

martinodb

  • Roomba
  • *
  • 7
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #5 on: March 02, 2019, 04:54:58 am »
Hey There!

 :trophy_bronze: to the subject line (it's the grail). But how can this be accomplished? How can a chat bot improvise? We'll have to think outside the chat box  ;D, any ideas?

Thanks! :) Well, sure we are far from AGI, but what we can strive to do in the near future, perhaps even now to some extent, is to focus less on trying to predict what the user will say or ask about, and more on having a correct and expressive internal model of the world, so to speak (facts, rules, inference procedures, etc) and learning strategies. In a way, even a humble mySQL database is "improvising" when it responds to a query nobody thought of in advance.  So in that light the challenge for the bot wouldn't be so much to improvise, but to look smart while doing it, but of course that's a matter of degree. IMO we should start we a solid foundation, an expressive logical formalism on which to build a CNL (controlled natural language) that can express higher-order relations and linguistic phenomena we need to codify knowledge, such as notions of time, states and events, intensions, attitudes (believe, want..). The goal would be to make the process of teaching the bot new stuff as pleasurable as possible (within realistic expectations) and, at some point, have the bot read stuff online all the time and only ask questions from time to time. The learning process itself is an open issue and will probably involve a variety of techniques, including indeed  statistical machine learning, but it would be a tool among many, to be used and fine-tuned by the AI, not the foundation of the AI, not in charge of its behaviour.

BTW, after all this talk of advanced AI, I'm almost embarrassed to link to my own little bot, https://github.com/martinodb/bobtailbot which for now only handles triples (like "Carol likes Bob") and rules (like "?x likes y? when ?x loves ?y") and it's painfully slow at that, and the code is full of ugly hacks I should get rid of ASAP. I'm looking at the possibility of basing its brain on a more advanced engine (CSNePS) than its current one (Clara), but this new engine still seems a bit buggy, so I'll probably clean up the old one and see if I can go beyond triples to actual sentences, events, etc. It will take some time, though. Anyway, feel free to fork it, play with it, contribute a new brain or a new adapter,  whatever, just keep in mind it's my first bot, it's a toy and it may (and probably will) undergo breaking changes ;)

*

martinodb

  • Roomba
  • *
  • 7
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #6 on: March 02, 2019, 04:58:56 am »
Just tossing in my $.02

Perhaps such a bot would be given all the language and grammar usage "Rules" it might need and the programming to examine, sort and execute NLP, decision making on a best case scenario, given what it had learned or known...not from some canned responses.

This AI would be able to search like Google or Watson to find possible answers to questions but also be able to fashion the responses in a more tailor-made manner based on the user's needs.

A rule-based bot with the added ability to draw from a deterministic tree or weighted decision on that which is determined to be the best or most appropriate solution or answer would seem to improvise.

It would be able not only to provide verbal responses to questions or statements but to also carry on casual conversations with the user. A digital companion!

...and then I woke up... drat!!

Yes, yes, that's along the lines of what I have in mind when I talk of improvising. Thanks! :)

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • *********************
  • Deep Thought
  • *
  • 5398
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #7 on: March 02, 2019, 01:55:55 pm »
In light of all the above, consider, if a bot can't learn then what good is it? If that is the case it will never be able to formulate topically, improvised responses based on what it has just learned or discovered.
While there have been many "Spoon fed" bots with a given cupboard of data, lots of scripted responses and many cute tricks and methods to convince the user(s) that it is smart, without autonomous learning it is just a digital puppet.

(no offense meant to anyone whos bot might fall into the non-learning category as these are just my feelings and thoughts on this).

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

*

martinodb

  • Roomba
  • *
  • 7
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #8 on: March 03, 2019, 01:53:34 am »
Hi, Art!

Just to clarify, when I say I'm "miffed with so much modern emphasis on machine learning", I don't mean the general idea of machines learning stuff, or unsupervised learning as a general concept, I mean specifically neural networks and similar techniques that nowadays claim the "machine learning" label, and that usually involve massive amounts of data, brute-force number crunching and opaque, cryptic, subsymbolic internal representations. And I don't resent so much the emphasis ON them, but the emphasis AWAY FROM symbolic AI, which IMO happened for no good reason.  Why do I believe there's no good reason? For one, expressiveness has been kept deliberately below what was not just imagined but even implemented, in the name of performance, decidability and other "good behaviour" traits. We are still dealing with straightjackets like Horn clauses (Prolog, Datalog, rule engines, SWRL..) and Description Logic (OWL, mostly) when the standard should be things like SUO-KIF, CLIF, IKL and beyond (ie higher-order syntax with full first-order semantics). Another sign is that, still to this day, I haven't found a robust, modern, well-supported, open-source CNL that can handle time, states and events. I'd love to be proven wrong on this particular point. It's not like it would require a crazy amount of work, in fact I've seen one or two doctoral dissertations that seem to do just that, but it's like nobody cares, maybe because everyone is playing with neural networks.

That said, even if an AI has to be "spoon fed" everything it knows, it can still be extremely useful and fun to use, just like virtually all software we use today is "dumb" in that sense, unable to learn anything on its own, but we still find it useful and fun, especially if the spoon feeding process is improved by a powerful CNL, perhaps in combination with equally good visual and speech UI modes. Having better expert systems that are easier to teach and train seems a worthy goal, even if they can't learn anything substantial on their own after they are deemed ready for production.

Besides, the concept of "learning" sounds somewhat nebulous. For instance, we may even say of a human being that he/she "never learns" if they make what we consider similar mistakes all the time, even though their dumbest decisions are way smarter than anything current AI can come up with. On the other extreme, even introducing new data into a spreadsheet increases the amount of knowledge it contains, so in a way when we do that we are "teaching" the spreadsheet some new information and therefore it's "learning" something new.

Lastly, I think there are reasons to believe that doing away with the symbolic paradigm and putting something like neural networks in charge of the AI would probably have a number of nasty consequences from a practical POV. That kind of AI would be nothing like a tool or an assistant, more like a daemon or an alien wild beast, it would perhaps be more aptly described as "intelligent artificial life"  than AI proper.  Sure, building a wild, uncontrollable AI may well be easier than building a reliable, trustworthy AI which is just as smart, and that kind of research will probably be needed, with all due precautions. But I wouldn't trust that AI to even set the speed of an electric toothbrush.

OK, I think I've shared enough controversial opinions for one post.  ;D

*

Hopefully Something

  • Trusty Member
  • *******
  • Starship Trooper
  • *
  • 287
  • So where are these cookies?
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #9 on: March 03, 2019, 06:53:29 am »
No worries, no judgment here. I'm clueless about chatbots, but everyone's always talking about them and Locksuit is convincing me that language will be an important part of AGI processing... So I'm starting to look into them.

I'm gathering that improvisation is like connecting previously unconnected dots. This requires you to have been given lots of dots (knowledge) to begin with. Otherwise the number of new connections you can make is going to drop like a square root graph. If every couple had one kid then humans would die out. I think what I'm looking for is spontaneity.

Generating random new dots to give you a chance of breaking free of the patterns you are stuck in. In a scenario where all possible connections have been made, the brain can't logically do anything new, its time to venture into the unknown by introducing a bit of chaos to the system. Then the ordering mechanisms will have a blast in trying to sort all of it out. Whereas before, they had painted themselves into an intellectual corner. Now they can fulfill their function, and so the system grows. It's a split personality, a symbiotic conflict. The "Sherlock Holmsien" paragon of deduction is always connecting the clues a step behind the complete lunatic who is tossing them out at random. This is like a brute force approach to learning, but its worthwhile effort for brains because they are so parallel.

Speaking of parallel, what do you think about the idea of physical neural nets? Personally, the idea of building a neural net in a regular computer is almost worse than decaffeinated instant coffee or nonalcoholic... anything. You are willingly missing the whole point! I'm gonna get a bald spot! I mean, sure you can do cheap testing and that's smart... But eventually we will have to go physical to experience the real nature of the architecture.
« Last Edit: March 03, 2019, 07:16:29 am by Hopefully Something »

*

martinodb

  • Roomba
  • *
  • 7
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #10 on: March 03, 2019, 08:55:23 pm »
I had a longer reply, but the forum's AI thinks it's spam lol. Anyway, yes, some randomness is fine. The AI will also need a constant stream of new data, from users, internet and so on, otherwise it will get bored and stop thinking. What I want is the thought process and the internal representation of the AI's thoughts and ideas to be as "human-readable" as possible.

Physical neural networks sound cool. Do you have any examples? After a quick search, I found examples of neural networks implemented in FPGAs. Is that what you have in mind, or something even more direct?

*

Hopefully Something

  • Trusty Member
  • *******
  • Starship Trooper
  • *
  • 287
  • So where are these cookies?
Re: "Stop. Lose all scripted responses. Improvisation only" :D
« Reply #11 on: March 03, 2019, 10:55:57 pm »
Sorry to hear that your post was cut. The irony though. AI! We're trying to help you man!... :)

Maybe in the absence of new data it could switch to self examination. Like an automated defragging.

I don't have any examples except imaginary stuff I've been day dreaming about.  But it's not like FPGA's, I'm thinking something you could build and experiment with at home. A 3D lattice of individually programmable nodes made from microprocessors which can signal to each other wirelessly. Probably with LED's and photo-sensors. I would like to build a simple ABI (Artificial Body Intelligence). Just a simple self sustaining, light consuming creature. Like a single celled organism swimming around in the primordial oceans. If I could get the basics working, I might start to understand how they can be expanded upon. 

 


Users Online

71 Guests, 0 Users

Most Online Today: 92. Most Online Ever: 340 (March 26, 2019, 09:47:57 pm)

Articles