Hey, folks! It's been a long while since I was active on this group, but
I've been getting back into the swing of things lately. I hope all is
well with my old friends here!
I've started an open source project, Engineering Intelligence
<https://github.com/hosford42/EngineeringIntelligence>, that I thought
might be of interest on this email list. The goal is to define
intelligence clearly and concretely, to offer a clear path for not only
identifying but /engineering/ systems that can be considered intelligent
in the traditional sense of the word: Being able to think and to
understand things. (This is in contrast to how 'AI' is typically used to
refer to programs that are /designed/ cleverly, rather than being clever
themselves, which leads to the ever moving goalposts of what counts as
intelligence in software.) The project consists of a growing body of
articles meant for human consumption; once it's sufficiently mature, a
secondary project will be created to actually implement a working proof
of concept in code.
I've already got the ball rolling with an informal set of requirements
that center around the idea that there should be an internal
representative model of the environment (a broad term including whatever
embodiment the system may manifest) which is continually maintained in
correspondence and synchronization with the environment through observed
sensory data. If you look at the senses as information bottlenecks, the
idea is to reconstruct the environment across a range of abstraction
levels, and then use this reconstructed model as a best-guess proxy for
the environment itself in order to make reasoned predictions and
decisions and, ideally, communicate with human beings and other
intelligent systems.
Again, this project is open source, and I welcome meaningful,
no-nonsense contributions (which will receive due credit documented
directly in the project). Please read the Contributing
<https://github.com/hosford42/EngineeringIntelligence#contributing>
section first, though. The most important rule is: *Zero woo.* Second
is: Be polite and respectful. *No flame wars!* Also, we can discuss
phenomenal consciousness, ethics, societal risks, spirituality, and
other meta-level topics here on this list; the project is strictly
focused on the /how/ of creating intelligence in software and ensuring
that it behaves as intended. And finally, I know many of you here have
your own pet theory of how to implement (general) intelligence. (Kudos!)
The broad definition of intelligence I've already started working on is
meant to be accommodating to many of these. It is agnostic as to how the
environment is represented, how multiple levels of abstraction are
attained, how the model is updated from new information, etc. I hope
that each of your conceptualizations can find a home within this broader
framework, but if not, we should discuss why not, and in what way the
framework and/or the approach can be adjusted to make them compatible. I
look forward to seeing issues
<https://github.com/hosford42/EngineeringIntelligence/issues> opened for
these discussions, so that all can participate and benefit.
Let's put our minds together and think like engineers to build something
indisputably intelligent!
--Aaron
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T30e487c54062d930-Ma1175e244b0bd222c0a47ea0
Delivery options: https://agi.topicbox.com/groups/agi/subscription
