Re: All the Agents Challenge (ATAC) at ISWC 2021

On Thu, May 27, 2021 at 1:38 AM Andrei Ciortea <andrei.ciortea@unisg.ch> wrote:
>
> Hi Melvin,
>
> Great and yes — this is certainly an area being researched. In fact, back in February we organized a Dagstuhl Seminar on "Autonomous Agents on the Web" and I think your vision is well aligned with several of the discussions we had during the seminar: https://www.dagstuhl.de/21072
>
> In terms of research, I think many of the pieces of the puzzle are already there as contributions in different communities (and some piece of the puzzle are certainly missing), but what we need is a concerted effort of these communities to align and integrate the various research threads — and this was the main motivation for organizing this seminar. We've just submitted the seminar report to Schloss Dagstuhl a couple of weeks back, it should be published soon.
>
> In terms of available technologies, I think we still have some way to go to get to the vision you are describing — multi-agent systems that are Web-scale, open, and long-lived. I think we’ll get there, but of course we first need to start somewhere.
>
> This is where Tobias’ initiative for the agent challenge comes in, and this is also where the 2 robots I’ve shared with you come in: one idea that came up during the seminar is to set up a shared "live" demonstrator space — a deployed, open, long-lived, and geographically distributed hypermedia environment that could provide a testbed for trying out ideas and identifying challenges. It’s an ambitious task, but we’ve already started to work towards it. :-)

Hello Andrei,

I find the idea of a shared Web-space amazing!
Such a testbed would add some joy to developing and compare different
approaches.

Regarding the challenge, I first looked into the maze; and I thought
about using an existing solver or reinforcement learning algorithm
'for mazes.' (a 'gym'). Then the problem perhaps reduces to writing a
script that interprets and follows links; or something like that.

Then I had to ask myself, what is an agent in the context of RWW?
Does the agent need 'agency'? I started to be confused ('agent' is a wide term).

And I've been confused regarding this for a long time!.  Some years
ago, I experimented with a program called 'agent' that used linked
data and N3 rules. The main functionalities were:

1. To consume/produce hypermedia
2. To know how to build internal/external states consisting of
combinations of various Web resources in the Web.

The program seemed helpful, but does it classify as an agent?

I want to find people who want to exchange approaches to build agents
that use linked data; it would probably be easier with a forum or
chat; this would make it easier to collaborate in the long-lived., Web
environment. And probably also enable people from outside academia to
jump in.

Cristian.


>
> We intend to organize a follow-up online event in July, I’m happy to keep you in the loop if interested. I'll get back to you soon with the additional materials and documentation.
>
> Melvin, thanks for your enthusiasm and we look forward to receive your submission for ATAC! :-)
>
> Cheers,
> Andrei
>
>
> On 25 May 2021, at 22:11, Melvin Carvalho <melvincarvalho@gmail.com> wrote:
>
>
>
> On Tue, 25 May 2021 at 18:52, Andrei Ciortea <andrei.ciortea@unisg.ch> wrote:
>>
>> Hi Melvin,
>>
>> Thanks for sharing this fascinating topic.  I'm very much interested in autonomous agents on the read-write web.  I have built a few of these in the past, and they seem to lack teeth.  The phrase "where are all the agents" resonates with me, and motivates me to try and look for something more robust
>> Browsing the links you provided, lead me to this: Introduction to Multi-Agent systems
>> https://cw.fel.cvut.cz/b191/_media/courses/be4m36mas/mas2016-l01-introduction.pdf
>> Would you say this is a good start for understanding where the research on this topic is at?
>> Would love to read if you could share any links
>>
>>
>> I think the above course is already a nice introduction to autonomous agents and multi-agent systems (MAS), another course with hands-on exercises & tutorials: https://www.emse.fr/~boissier/enseignement/maop20-fall/index.html
>>
>> This course is focused on multi-agent oriented programming [1] and the JaCaMo platform, which is one of the main platforms for MAS (additional documentation and tutorials on GitHub): https://github.com/jacamo-lang/jacamo
>>
>> [1] Olivier Boissier, Rafael H. Bordini, Jomi F. Hübner, Alessandro Ricci, Andrea Santi, Multi-agent oriented programming with JaCaMo, Science of Computer Programming, Volume 78, Issue 6, 2013, Pages 747-761, ISSN 0167-6423, https://doi.org/10.1016/j.scico.2011.10.004. Link to PDF: https://www.sciencedirect.com/sdfe/reader/pii/S016764231100181X/pdf
>
>
> Hi Andrei!
>
> Thanks so much for sharing this, I've read a bit of it, and from what I can gather it's a java platform with:
>
> Jason -- AgentSpeak command and programming language -- Agents
>
> Cartago -- Like a virtual machine for funning agents -- Environment
>
> Moise -- OO style way of organizing agents -- Organization
>
> Thanks alot this is very interesting, and a whole new area for me to look at, with a bit of a learning curve.  There seems lots of useful things in there that it's possible to reuse
>
> My interest is along the lines of multi agent systems, that scale to the web, and are programming language agnostic, but communicate with each other over, for example, http requests, and link to each other e.g. with JSON(-LD)
>
> I'd love your input on this.  My experience of programming web based agents is that they have a maintenance overhead, such the autonomy is not really practical, I may call the ones I've made semi-autonomous.  This is akin to the desktop paradigm, where you might start an app, use it for a while and then shut it down (agent lifecycle).  I am quite interested in another style of app, where it runs as a daemon, e.g. in the system tray.  So you run it, and it keeps going in the background.  When you shut your machine down it'll stop, and reboot the machine it runs on startup, so it's kind of always there, you don thave to worry about it, like a daemon
>
> Multi agent systems that run more or less as daemons against the web as a (read-write) state machine have the advantage of much lower maintenance, and interesting emergent features as they become more autonomous and interact with each other.  The system itself would be modular, with reusable components that could be composed together either into a composite agent, or different parts running remotely.  They would also adapt to change, and be backwards compatible with the existing web (of data)
>
> So they may not even need a JaCaMo type environment to run in, but could reuse the organizational and agent speak features
>
> Would love to know your thoughts on this, if it's an area people research, or a direction researchers in the field might want to look at
>
>>
>>
>> Also looking forward to the environments being released on 22 May.  Not being an academic myself, Im hesitant to try and submit an agent to solve the maze, or turn the lights off in the building
>>
>>
>> If you are interested to hack away with some Web-enabled devices, we also have a couple of robotic arms that we are happy to share. :-)
>>
>> You can access these robots via HTTP and we have a live video feed for each:
>> https://interactions.ics.unisg.ch/61-102/cam1/live-stream
>> https://interactions.ics.unisg.ch/61-102/cam2/live-stream
>
>
> I'd love to find out how to control a robotic arm with a robot, yes! :)
>
>>
>>
>> One of the robots is hooked up to a lamp that turns on when you activate the robot, so you can use it regardless of your time zone (our lab is in St.Gallen, Switzerland).
>>
>> Let me know if you are interested to use these robots and I am happy to send you additional details & documentation.
>>
>> Best wishes,
>> Andrei
>>
>> --
>> Dr. Andrei Ciortea
>> Interaction- and Communication-based Systems
>> Institute of Computer Science
>> University of St.Gallen
>> https://interactions.ics.unisg.ch/
>>
>>
>>
>> On 21 May 2021, at 23:56, Tobias Käfer <tobias.kaefer@kit.edu> wrote:
>>
>> Hi Melvin,
>>
>> Am 21.05.21 um 19:51 schrieb Melvin Carvalho:
>>
>> Thanks for sharing this fascinating topic.  I'm very much interested in autonomous agents on the read-write web.  I have built a few of these in the past, and they seem to lack teeth..  The phrase "where are all the agents" resonates with me, and motivates me to try and look for something more robust
>> Browsing the links you provided, lead me to this: Introduction to Multi-Agent systems
>> https://cw.fel.cvut.cz/b191/_media/courses/be4m36mas/mas2016-l01-introduction.pdf
>> Would you say this is a good start for understanding where the research on this topic is at?
>> Would love to read if you could share any links
>>
>>
>> That's a general introduction to MAS, and probably does not reflect so much the web part.  Some would maybe think that already single agents doing useful things on the web would be a step forward.  A more webby rough introduction in which I was involved could be a tutorial at the IoT conference [1].  In my echo chamber, we often use Linked Data-Fu [2] for rule-based simple reflex agents.  I have CC'ed Andrei who may give you some more pointers from his perspective.
>>
>> [1] http://people.aifb.kit.edu/co1683/2020/iot-tutorial/
>> [2] http://linked-data-fu.github.io/
>>
>> Also looking forward to the environments being released on 22 May.  Not being an academic myself, Im hesitant to try and submit an agent to solve the maze, or turn the lights off in the building
>>
>>
>> No need to be afraid. We love practical things. The environments released are the maze and the building.
>>
>> I'm also interested in any submissions to this challenge, including evolutionary agents.  This little demo of fish as agents in an evolutionary environment with finite resources I really like because it has emergent properties:
>> http://caza.la/shoal/
>>
>>
>> Cool stuff!
>>
>> Cheers,
>>
>> Tobias
>
>

Received on Tuesday, 1 June 2021 14:53:01 UTC