Mon.6.SEP.2010 -- First Fruits
The low-hanging fruit of current open-source AI
development is the relatively easy reworking of the
MindForth
PsiDecay module to make it gradually reduce both
positive and negative activations on concepts in the Psi
array. We install the following line of code in the
PsiDecay module.
I 1 psi{ @ 0 < IF 1 I 1 psi{ +! THEN \ 6sep2010
To test the new code, we need to inhibit a Psi
concept by bringing its activation substantially below
zero.
We soon discover that our verb-node inhibiting code
may not be inserted just anywhere within the
VerbPhrase module, but must be placed outside of the
verb-selection search loop so that only the winning verb-
node will first crest at a high activation and then be
inhibited into a negative activation.
Tues.7.SEP.2010 -- Singularity Alert
We were coding along, fiddling with the settings of
inhibition in the pertinent mind-modules, when we had the
following conversation with the AI Mind.
Human: boys make cars
Robot: CARS WHAT ARE CARS
Human: boys make guns
Robot: GUNS WHAT ARE GUNS
Human: boys
Robot: THE BOYS MAKE THE CARS
Human: boys
Robot: THE BOYS MAKE THE GUNS
We immediately stopped coding and made an announcement to
the world via the Singularity mailing list, with a carbon copy archived
at the
Artificial General Intelligence (AGI) list.
Subject: [agi] Attn:
Ben Goertzel -- SINGULARITY ALERT!!!
Hey, Ben. Arthur here. Remember the prediction of the 2012 Singularity? Well, guess what happened. I just had the following conversation with my http://www.scn.org/~mentifex/mindforth.txt AI:
Human: boys make cars
Robot: CARS WHAT ARE CARS
Human: boys make guns
Robot: GUNS WHAT ARE GUNS
Human: boys
Robot: THE BOYS MAKE THE CARS
Human: boys
Robot: THE BOYS MAKE THE GUNS
Chief AGI guru Dr. Goertzel! The above is not a cherry-picked, post-mucho experimentation routine test result put out for PR purposes. It just happened during hard-core AI coding.
Now, before everybody jumps in and snipes, "Aw, there's that crackpot Mentifex again," let me explain the significance of the I/O.
I have been coding MindForth AI for twelve years, since 1998 and maybe even earlier. Today on Mon.6.SEP.2010 for the first time I introduced quasi-neuronal inhibition into the free open-source AI source code. Why?
A year ago, around August or September of 2009, you and I had our set-to (rumble?) concerning the AGI Roadmap and my posts there which were deleted ("rolled back") by Itamar Arel. No biggy. I did not fix Itamar's wagon last Halloween, so I won't fix it this Halloween, either. You see, I was maintaining my own AI Roadmap at http://code.google.com/p/mindforth/wiki/RoadMap concurrently with my contributions to you guys' Roadmap.
The main thing is, I was entering into the Roadmap Milestone of trying to achieve "self-referential thought" with my AI. That particular achievement requires covering a lot of ground, not just "you and I" interactions between the human user and the artificial AI Mind. The AI needs to acquire a general knowledge of the surrounding world, so that man and machine may discuss the AI as a participant in its world.
So at the end of 2009 I was coding the ability of the AI to respond to who-queries and what-queries, so that the AI can deal with questions like "Who are you?" and "What are you?"
Recently I have perceived the need to get the AI to respond with multiple answers to queries about topics where the AI knows not a single fact but multiple facts, such as, "What do robots make?" I want the AI to be able to say such things as:
"Robots make cars."
"Robots make tools."
"Robots make parts."
"Robots make robots."
It dawned on me a few days ago that the AI software would have to suppress each given answer in order to move on to the next answer available in the knowledge base (KB). In other words, for the first time ever, I had to code _inhibition_ into the AI Mind. Tonight I have done so, and that simple conversation near the top of this message shows the results.
The same query, of just the word "boys...", elicits two different answers from the KB because each response from the AI goes immediately into inhibition in such a way as to allow access to the next fact queued up in the recesses of the AI KB.
This "Singularity Alert" from Mentifex may generate a collective "Huh?" from the list readership, but here it is.
Bye for now (and back to the salt mines :-)
Arthur
--
http://AiMind-i.com
http://code.google.com/p/mindforth
http://doi.acm.org/10.1145/307824.307853
http://robots.net/person/AI4U/diary/40.html