Friendliness vs. Compassion, revisited (plus a bunch of babbling about what I've been up to this year)
Wow, it's been a long time since I've blogged on here -- apparently I haven't been in a bloggy mood.
It's been a busy year ... I've sent my oldest son Zarathustra off to college at age 16 (to Simon's Rock College, www.simons-rock.edu, the same place I, my sister and my ex-wife went way back in the day), which is a very odd feeling ... I finished a pretty decent draft of a novel, Echoes of the Great Farewell, which is a completely lunatic prose-poetic novel-thing told from the stream-of-consciousness point of view of a madman who believes that hallucinogenic mushrooms have told him how to create a superhuman AI (perhaps I'll actually try to get this one published, though it's not a terribly publisher-friendly beast) ... I came up with a substantial simplification of the Novamente AI design, which I'm pretty happy with due to its deep foundations in systems philosophy ... worked with my Novamente colleagues to take a few more incremental steps toward implementation of the Novamente AGI design (especial progress in the area of probabilistic reasoning, thanks to the excellent efforts of Ari Heljakka) ... did some really nice data mining work in the context of some commercial projects ... make some freaky instrumental music recordings that my wife at least enjoyed ... hiked the Na Pali Trail on Kaui and a whole bunch of trails near the Matterhorn in the Alps with my mountain-maniacal young wife Izabela ... co-organized a conference (the AGIRI workshop) ... published a philosophy book, The Hidden Pattern, which tied together a whole bunch of recent essays into a pretty coherent statement of the "world as pattern" perspective that has motivated much of my thinking ... developed a new approach to AGI developmental psychology (together with Stephan Vladimir Bugaj) ... starred in a few animations created by my son Zebulon (zebradillo.com), including one about rogue AI and another in which I mercilessly murder a lot of dogs ... helped discover what seems to be the first plausible genetic underpinnings for Chronic Fatigue Syndrome (together with colleagues at the CDC and Biomind LLC) ... and jeez, well this list is dragging on, but it's really not the half of it...
A pretty full year -- fun to live; too much going on to permit much blogging ... but frustrating in the big picture, given that it's been yet another year in which only modest incremental progress has been made toward my most important goal of creating AGI. My understanding of AGI and the universe has increased significantly this year so far, which is important. And the Novamente codebase has advanced too. Again, though, balancing the goal of achieving AGI with the goal of earning cash to support a family (send the kids to college, pay the alimony (which runs out in another 9 months -- yay!!), etc.) proves a tough nut to crack, and is just a dilemma I keep living with, without solving it satisfactorily so far.... I'll be spending much of the next 6 weeks trying to solve it again, by doing a bunch of meetings and social-networking events partially aimed at eventually putting me in touch with investors or other partners who may be interested in funding my AGI work more fully than is currently the case. (Don't get me wrong, we are moving toward AGI in the Novamente project right now, but we could be moving 10 times faster with some fairly modest investment ... the small amount of investment we've gotten so far, combined with the small surplus value my colleauges and I have managed to extract from our commercial narrow-AI contracts, is far from enough to move us along at maximum rate.)
BUT ANYWAY ... all this was not the point of this blog entry. Actually, the point was to give a link to an essay I wrote on a train from Genova to Zermatt, following a very interesting chat with Shane Legg and Izabela. Shane wrote a blog entry after our conversation, which can be found by going to his site
http://www.vetta.org/
and searching for the entry titled "Friendly AI is Bunk." I wrote an essay with a similar theme but a slightly different set of arguments. It is found at
http://www.goertzel.org/papers/LimitationsOnFriendliness.pdf
The essay is informal in the sense of a blog entry, but is too long to be a blog entry. My argument is a bit more positive than Shane's in that, although I agree with him that guaranteeing "AI Friendliness" in a Yudkowskian sense is very unlikely, I think there may be more general and abstract properties ("compassion" (properly defined, and I'm not sure how), anyone?) that can be more successfully built into a self-modifying AI.... (Shane by the way is a deep AI thinker who is now a PhD student working with Marcus Hutter on the theory of infinitely powerful AI's, and who prior to that did a bunch of things including working with me on the Webmind AI system in the late 1990's, and working with Peter Voss on the A2I2 AGI architecture.)
While you're paying attention, you may be interested in another idea I've been working on lately, which is a variant of the Lojban language (tentatively called Lojban++) that I think may be very useful for communication between humans and early-stage AGI's. If you're curious you can read about it at
http://www.goertzel.org/papers/lojbanplusplus.pdf
With a view toward making Lojban++ into something really usable, I've been spending a bit of time studying Lojban lately, which is a slow but fascinating and rewarding process that I encourage others to undertake as well (see www.lojban.org).
Well, OK ... that's enough for now ... time for bed. (I often like late-night as a time for work due to the quiet and lack of interruptions' but tonight my daughter is having a friend sleep over and they're having an extremely raucous post-midnight mop-the-dirty-kitchen-floor/mock-ice-skating party which is more conducive to blogging than serious work ;-). I hope to blog a bit more often in the next months; for whatever obscure human-psychology reason it seems to gratify some aspects of my psyche. Hopefully the rest of 2006 will be just as fun and diverse as the part so far -- and even more productive for Novamente AGI...
It's been a busy year ... I've sent my oldest son Zarathustra off to college at age 16 (to Simon's Rock College, www.simons-rock.edu, the same place I, my sister and my ex-wife went way back in the day), which is a very odd feeling ... I finished a pretty decent draft of a novel, Echoes of the Great Farewell, which is a completely lunatic prose-poetic novel-thing told from the stream-of-consciousness point of view of a madman who believes that hallucinogenic mushrooms have told him how to create a superhuman AI (perhaps I'll actually try to get this one published, though it's not a terribly publisher-friendly beast) ... I came up with a substantial simplification of the Novamente AI design, which I'm pretty happy with due to its deep foundations in systems philosophy ... worked with my Novamente colleagues to take a few more incremental steps toward implementation of the Novamente AGI design (especial progress in the area of probabilistic reasoning, thanks to the excellent efforts of Ari Heljakka) ... did some really nice data mining work in the context of some commercial projects ... make some freaky instrumental music recordings that my wife at least enjoyed ... hiked the Na Pali Trail on Kaui and a whole bunch of trails near the Matterhorn in the Alps with my mountain-maniacal young wife Izabela ... co-organized a conference (the AGIRI workshop) ... published a philosophy book, The Hidden Pattern, which tied together a whole bunch of recent essays into a pretty coherent statement of the "world as pattern" perspective that has motivated much of my thinking ... developed a new approach to AGI developmental psychology (together with Stephan Vladimir Bugaj) ... starred in a few animations created by my son Zebulon (zebradillo.com), including one about rogue AI and another in which I mercilessly murder a lot of dogs ... helped discover what seems to be the first plausible genetic underpinnings for Chronic Fatigue Syndrome (together with colleagues at the CDC and Biomind LLC) ... and jeez, well this list is dragging on, but it's really not the half of it...
A pretty full year -- fun to live; too much going on to permit much blogging ... but frustrating in the big picture, given that it's been yet another year in which only modest incremental progress has been made toward my most important goal of creating AGI. My understanding of AGI and the universe has increased significantly this year so far, which is important. And the Novamente codebase has advanced too. Again, though, balancing the goal of achieving AGI with the goal of earning cash to support a family (send the kids to college, pay the alimony (which runs out in another 9 months -- yay!!), etc.) proves a tough nut to crack, and is just a dilemma I keep living with, without solving it satisfactorily so far.... I'll be spending much of the next 6 weeks trying to solve it again, by doing a bunch of meetings and social-networking events partially aimed at eventually putting me in touch with investors or other partners who may be interested in funding my AGI work more fully than is currently the case. (Don't get me wrong, we are moving toward AGI in the Novamente project right now, but we could be moving 10 times faster with some fairly modest investment ... the small amount of investment we've gotten so far, combined with the small surplus value my colleauges and I have managed to extract from our commercial narrow-AI contracts, is far from enough to move us along at maximum rate.)
BUT ANYWAY ... all this was not the point of this blog entry. Actually, the point was to give a link to an essay I wrote on a train from Genova to Zermatt, following a very interesting chat with Shane Legg and Izabela. Shane wrote a blog entry after our conversation, which can be found by going to his site
http://www.vetta.org/
and searching for the entry titled "Friendly AI is Bunk." I wrote an essay with a similar theme but a slightly different set of arguments. It is found at
http://www.goertzel.org/papers/LimitationsOnFriendliness.pdf
The essay is informal in the sense of a blog entry, but is too long to be a blog entry. My argument is a bit more positive than Shane's in that, although I agree with him that guaranteeing "AI Friendliness" in a Yudkowskian sense is very unlikely, I think there may be more general and abstract properties ("compassion" (properly defined, and I'm not sure how), anyone?) that can be more successfully built into a self-modifying AI.... (Shane by the way is a deep AI thinker who is now a PhD student working with Marcus Hutter on the theory of infinitely powerful AI's, and who prior to that did a bunch of things including working with me on the Webmind AI system in the late 1990's, and working with Peter Voss on the A2I2 AGI architecture.)
While you're paying attention, you may be interested in another idea I've been working on lately, which is a variant of the Lojban language (tentatively called Lojban++) that I think may be very useful for communication between humans and early-stage AGI's. If you're curious you can read about it at
http://www.goertzel.org/papers/lojbanplusplus.pdf
With a view toward making Lojban++ into something really usable, I've been spending a bit of time studying Lojban lately, which is a slow but fascinating and rewarding process that I encourage others to undertake as well (see www.lojban.org).
Well, OK ... that's enough for now ... time for bed. (I often like late-night as a time for work due to the quiet and lack of interruptions' but tonight my daughter is having a friend sleep over and they're having an extremely raucous post-midnight mop-the-dirty-kitchen-floor/mock-ice-skating party which is more conducive to blogging than serious work ;-). I hope to blog a bit more often in the next months; for whatever obscure human-psychology reason it seems to gratify some aspects of my psyche. Hopefully the rest of 2006 will be just as fun and diverse as the part so far -- and even more productive for Novamente AGI...
posted by Ben at 12:55 AM
6 comments
