[Zero]

Hi guys,

Yeah, "real time" isn't the right way to say it.

What I mean is... Chatbot engines I know about, Rivescript, AIML, give you an answer instantly, as soon as you keypress Enter. But when you're in a chatroom, people don't answer intantly: you wait a few seconds, then someone says something, then a few seconds later, someone else says something, ...etc.

I know we can easily simulate a delay before the answer of the chatbot, so it feels like someone is typing on a keyboard, but that's not my question.

Chatbot engines I know work like a REPL. But is there a chatbot engine that would work like some sort of TCP server.

I imagine an engine that's permanently looping, thinking. Sometimes it receives a message from its botmaster, sometimes it sends a message. The messages it receives modify its thinking process. It's "real time". Is there such an engine somewhere?

Am I being understandable at least?

[LOCKSUIT]

So you want a real-time chatbot that answers with a delay but decides/computes instantly but also uses time to loop and think.

I understand your question too :).

Btw real-time is an opinion. Optical AIs would be waiting billions of years for our chatbots to finish. And even they HAVE to wait for some things.

[Zero]

Yes, if it's very fast, we look very slow, like bullet-time.

But the program could adjust the "depth" to balance the time effect. On a fast computer, the think process is very deep (so it takes time). On a slower computer, the think process is more superficial (it takes less time). The depth adjust real-time to balance the subjective speed of the world (the speed of world in AI's eye).

But, yeah that's the idea, it uses time to loop and think, do heurisitics, calculate probabilities about future, explore paths, try to understand better the botmaster's inputs...

[LOCKSUIT]

But, yeah that's the idea, it uses time to loop and think, do heuristics, calculate probabilities about future, explore paths, try to understand better the botmaster's inputs...

At the end of your reply you mean this is what you're looking for on top what I said you want.

So you want a real-time chatbot that answers with a delay but decides/computes instantly but also uses time to loop and think.

[Zero]

I was just confirming that you indeed understood my question.

Interesting subject, isn't it?

BTW, that's another point about programming languages: they are perfectly non-ambigous, so they're a perfect way to communicate. See how we misunderstand each other sometimes? Well, this can't happen when you read source code. Maybe we lack a non-ambigous language to describe AGI.

[ivan.moony]

I heard that Siri (Apple's assistant) is taking some time to process information before it answers. As I recall, people complained about its speed.

About ambiguity, you might be surprised, but the language I'm working on relies on ambiguities to form sets of data. So, instead of unambiguous natural language, I get ambiguous programming language.

[Zero]

Does Siri sometimes speaks autonomously, without being asked something, and for other things than notifications? (never tried Siri, Cortana, ...etc.)

Ambiguity: Mmmh... Interesting. But if you know what ambiguity leads to, is it still ambiguity?

[ivan.moony]

Does Siri sometimes speaks autonomously, without being asked something, and for other things than notifications? (never tried Siri, Cortana, ...etc.)

I didn't try Siri and company either. Maybe someone else did...

Ambiguity: Mmmh... Interesting. But if you know what ambiguity leads to, is it still ambiguity?

I think it is, but in order to exploit it, you have to know exactly what you're doing. But it is not so complicated and magical thing, I'll try to write about it soon.

[Zero]

Cool.

[Freddy]

I only used Cortana a bit - she did used to do stuff at one time but she seemed to stop bugging me - I must have switched something off 

Bots saying things after a period of inactivity is not new really - you could get UltraHal to do this, but it was a bit iffy I recall. That in itself is not tricky, you just need to monitor user activity if an illusion is all you need.

[Art]

Ya beat me Freddy...I was going to mention UltraHal. Sometimes it would take a few moments before responding and once in a great while, it would say nothing at all, as if thinking about what the user said to it. (or typed, as the case might be). Hal, as it is called, can be spoken to verbally or typed to and it can respond via any of one's assigned TTS voices.
There are parts of this chatbot that are hard-coded into its core but there are many "user modified" areas within the program.

As I understand, it is being re-written with a vastly updated / upgraded brain as well as a new avatar / character system. You can certain find out more at; www.zabaware.com

For the record, Siri, Google Now, Alexa, Cortana are NOT chatbots, they are Search Assistants and will not carry a conversation any more than maybe a couple of exchanges. I would not and do not even put them in the same category as Chatbots, who's main goal / purpose is...to Chat.

[Zero]

Yes, they're not chatbots, I agree. I'd try Cortana if she wasn't always asking for my GPS coordinates... too bad.

Thanks, I'll see what's UltraHAL capable of.

But my question really wasn't about simulating delay (an illusion), but rather about the way the engines work internally, which seems to be "always like a REPL". 

[Art]

Is what you're referring to something like a programming structure within LISP which repeatedly reads input from the user, evaluates it, finds the best suited answer then displays the result?

Well most chatbots do that without the looping being done as that is wasteful of both time and CPU. Their choice is done on a case by case basis, usually as soon as the user's statement / question is completed and received by either end of vocal input or by the Enter key being pressed.

The search / Scanning process is usually pretty fast and while there might be several possible answers, the bot usually pick what it believes is the most relevant or "matching" answer. Sometimes the bot might have several from which to choose and it could simply pick the first from a list or choose one from random to simulate a more human behavior.

[Zero]

Is what you're referring to something like a programming structure within LISP which repeatedly reads input from the user, evaluates it, finds the best suited answer then displays the result?

Well most chatbots do that without the looping being done as that is wasteful of both time and CPU.

Yes, this is what I'm referring to. When you say "most chatbots", do you mean "all well-known chatbot engines"? I know Rivescript and AIML, but there are others...

The 3 following assertions are true:
- one of the purposes of chatbots is to pretend to be human,
- pretending to be human is a very complex task
- we have a lot of cpu cycle to waste

Then, how come we don't use the unused cpu cycles (between responses) to enhance the bot's ability to pretend to be human?

How could we use this unused energy?

Maybe we don't use it because we don't know what to do with it. Perhaps, we lack a large dataset to work on, like a huge list of interrelated sentences...

[8pla.net]

Thank you for starting this super interesting discussion thread.  I don't see anyone mentioned it here yet... So let me add, you are discussing what is known as "message-mode" and "character-mode". 

Is there a "character-mode" chatbot competition engine?  Yes.  And, the chatbot contestants must speak the character-mode (and/or message-mode) protocol in order to enter the contest. 

In the annual Loebner Prize Competition, the traditional LPP (Loebner Prize Protocol)  used to be in character-mode (real time), but this year the new LPP is in message-mode.   So the answer is "Yes." to the question, "Is there a 'real time' chatbot engine?"

Reference: https://www.chatbots.org/ai_zone/viewthread/180/

Disclaimer:
This is limited to discussion purposes only, and is not to be used or relied upon otherwise in any way whatsoever.  Chatbot contest rules may have a high level of difficulty.  Do not use this as advice on any contest rules.