AGI Discussion Forum, May 14: Minecraft for aGI?

72 views
Skip to first unread message

Ben Goertzel

unread,
May 10, 2021, 3:52:11 AM5/10/21
to AGI, opencog
Hi all,

Fri May 14, 2021 we will have the next AGI Discussion Forum session on the topic

"Minecraft for AGI? -- Brainstorming Session on Minecraft as an
Experimentation/Demonstration Platform for Early-Stage AGI systems."

Location: https://singularitynet.zoom.us/my/benbot

Some relevant thoughts in a Hyperon context are here

https://docs.google.com/document/d/1OaocoMBWh72roMkxbBaUYxmUGGM02-taYHNZvARI86I/edit#

The previous and inaugural AGI Discussion Forum was an interesting
discussion but the focused and technical nature of the paper used to
seed discussion did not lend itself so much to wide-open
brainstorming/discussion. I've aimed this time to choose a topic
that's super easy for anyone to enter into... opposite end of the
spectrum perhaps...

thanks
Ben

--
Ben Goertzel, PhD
http://goertzel.org

“He not busy being born is busy dying" -- Bob Dylan

Ben Goertzel

unread,
May 10, 2021, 3:58:49 AM5/10/21
to AGI, opencog
Ah, forgot to re-include the general AGI Discussion Forum link,

https://wiki.opencog.org/w/AGI_Discussion_Forum

Jon P

unread,
May 15, 2021, 3:15:18 PM5/15/21
to opencog
Nice discussion this week. At the end there was a call for topic suggestions so here is one in case it is helpful.

Alexey made an interesting point about not wanting to focus too much on a single problem, or learning environment, while building Hyperon because that might cause the system to become biased towards that way of thinking rather than becoming general.

More learning environments may be good as it helps promote generality in the design of the system. Having only a few might help focus resources and makes it easier to make visible progress. Finding some which are "orthogonal", as in they stress the system in different ways, might be helpful.

It may also matter if there are people on the project who are passionate about a particular area and would enjoy developing it. And also if some environments are premade (like OpenAI Gym or Minecraft, for instance) this might save work compared to making bespoke ones, however bespoke ones might have desirable features.

It is also possible that building a suite of learning environments can be done early in development without needing foundational decisions about Hyperon's Atomese and cognitive algorithms implementation to be made yet.

So maybe that would make a good discussion point? What is a good list of learning environments to prepare for Hyperon for when it is ready to interact with them?

Jon
Reply all
Reply to author
Forward
0 new messages