jason-users Mailing List for jason

-------- Forwarded Message --------
Subject: Re: [Jason-users] Metacognitive 
(self-modifying/improving/learning) agents in Jason AgentSpeak?
Date: Fri, 8 Jun 2018 10:34:26 -0300
From: Rafael H Bordini <raf...@gm...>
To: jas...@li..., jon...@ya...

Hi Jonatan,

Apologies for the long delay in replying. I noticed Francisco had 
already sent a suggestion, and then failed to add a comment I wanted to 
make. Just about updating AgentSpeak programs... Note you do NOT need to 
stop and run an agent again to change its behaviour. In Jason, agents 
(and therefore planners, humans, etc.) can send new plans to agents via 
message passing and the receiving agents can, if they agree, add those 
plans to their plan library during runtime, hence changing their 
behaviour (their know-how, as well as changing other mental attitudes). 
Personally I don't know about any of the metacognitive features you 
mentioned (interesting the Gödel machines, I didn't know about them) 
having been incorporated into Jason.

Best,

Rafael




On 31/05/2018 05:25, Francisco Grimaldo Moreno wrote:
> Hi Jonatan,
> 
> You can have a look at "Modules & Namespaces in Jason" and, maybe, to this
> library we developed some years ago:
> 
> http://jildt.sourceforge.net/
> 
> 
> Best,
> Fran
> 
> 
> 
> El dimecres, 30 de maig de 2018, a les 22:41:23 CEST, Jonatan Knud Lauritsen
> via Jason-users va escriure:
>> Hi!
>> I am aware of the following efforts e.g. Cox that tries to make agents with
>> metacognitive features - e.g. agents, that can self-reflect, that can infer
>> new goals and subgoals, that can adapt, that can self-modify/improve/learn
>> (there is also research about Goedel machines - GOEDEL MACHINE HOME PAGE ).
>> My question is - are there Jason infrastructure (maybe dual agent
>> preception/action cycles or some hierarchy) that allows such metacognitive
>> features or maybe there are clear way of self-modifying AgentSpeak
>> programs? Of course, one cal always stop execution of the AgentSpeak
>> program, update the code and run AgentSpeak program again - with improved
>> source code. Are there efforts in science (using Jason and AgentSpeak) to
>> implement such metacognitive agents? There is this exciting field of
>> human-level AI (see Prague conference this summer) and agents with
>> metacognitive features are essential to this field.
>>
>>
>>
>>
>> GOEDEL MACHINE HOME PAGE
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> Cox
>>
>>
>>
>> regards, Jonatan
> 
>