As a (somewhat) active participant in the global AGI R&D effort, I am of course looking for code and services that can be exposed for public consumption via APIs and such, not only because it could be a revenue source and a real-world test, but also because it is a kind of "global embodiment" that will allow one's "system" to collect data from a multitude of users, use-cases, contexts etc. I've previously denounced APIs as too banal to be the key to general intelligence, and here's my simple "why": the hard work in the eyes of the machine would be establishing the proper context and disambiguation, and in the API economy it is the consumer who has done most of the hard work, finding the right calls and parameters.
Then I view "Question Answering" as a bit of an intermediate domain, we can pose rather free-form and all-enveloping questions to Watson, but only after we fed it with a few thousand or million documents of this domain or the other. The emphasis here is on "million", it would be nuts to API-fy this by sending gigabytes of data together with our 5 word question. You could of course send smaller "references" to the corpus such as URI, URL, DOI etc, but one can't escape the feeling that your mind is where your data is. On the next level, what if an AGI-pretender "watched its every step", maintained ontologies that were annotated with both offline and realtime data, tried to figure out what your problem is (rather than what your API call is) and worked out which combination of resources could solve your problem? More than your cognitive assistant that could be your conscious peer. I have previously suggested that a banal Command and Control Center which kept an eye on resources and is in constant reallocation and discovery of resources (I would say "is conscious of its resources"), could very well be the centerpiece of general intelligence. Of course you cannot avoid something like Enterprise Resource Planning with real-world resources and problems, but the first order Computational Resources with their ping times, network outages, slowdowns, L1 and L2 caches and the pains thereof, new SSSE4 instructions and, worst of all, the halting problem where it might be OK if your webservice times out once in a while but your autonomous car going 150mph should never time out, all of this requires its own holistic intelligence and I doubt it would be sub-par to general intelligence. Which brings me to the very sticky problem of "discovering computation", as opposed to using it. What about having the AGI-brain play scientist with assembly opcodes, stringing together C programs and figuring out what they do if anything, and even perform such program generation or metaprogramming without access to the formal specifications of compilers, and/or figuring out workarounds for compiler bugs, the Pentium Bug etc. Just like hackers once in a while discover undocumented features and backdoors in processors, a living being discovers very much the same in an undocumented world. If we take as an example utilitarian traveling as opposed to riding the Harley in order to feel free again, let's say "I need to go to Singapore to talk to Ben to feel good about myself", there is an infinite number of ways to make the Ben Pilgrimage, and a greater infinity of ways to fail. The amount of possible obstacles also approaches infinity (biometric passport? wheelchair access? 2-day aircontroller strike? only domestic credit cards?language barrier?) and it is fair to say that any such non-routine operation is a journey of discovery and creativity. I am afraid that as computer scientists we are lying to ourselves, "computational discovery and self-programming is a bit too tough, but real-world reasoning I can handle". I'd dare say the reason we created computers is that the world is programmable, you get stacks, automatic flow, interrupts, triggers, outages and failures. Obviously that goes together with the world being engineerable, one flick of a switch and you move mountains, and for the same reasons one off-by-one-error and your wonderfully engineered spaceship blows up in mid air. Just to mention our Founding Father Ben again, they are predictably making design decisions for OpenCog on an almost daily basis, and since performance is always a factor they need to reconsider and fine-tune these decisions regularly. That's all Business As Usual, but imagine what a ride it would be if your proto-AGI said "here I am, I have an ethernet connection, 8 cores and I want to make it to Singapore"! AT ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
