3 ms·

There are two epistemic poles: the atomistic and the probabilistic. The author subscribes to a rule-based atomistic worldview, asserting that any perspective mi

by nialse 2y ago

There are two epistemic poles: the atomistic and the probabilistic. The author subscribes to a rule-based atomistic worldview, asserting that any perspective misaligned with this framework is incorrect. Currently, academia is undergoing a paradigm shift in the field of artificial intelligence. Symbolic AI, which was the initial research focus, is rapidly being replaced by statistical AI methodologies. This transition diminishes the relevance of atomistic or symbolic scientists, making them worry they might become irrelevant.

irunmyownemail 2y ago
Not sure I followed all of that lingo but it sounds like a fancy way of saying, if you're losing the game, try shifting the goal post.
- nialse 2y ago
  Indeed and unfortunately. I've been reading up on "the binding problem" in AI lately and came across a paper that hinged on there being an "object representation" which would magically solve the apparent issues in symbolic AI. In the discussion some 20 pages later, the authors confessed that they, nor anybody else, could define what an object was in the first place. Sometimes the efforts seem focused on "not letting the other team win" rather than actually having something tangible to bring to the table.
dbmikus 2y ago
I never want to claim certainties, but it seems pretty close to certain that symbolic AI loses to statistical AI. I think there is room for statistical AI to operate symbolic systems so we can better control outputs. Actually, that's kind of what is going on when we ask AI to write code.
Shorel 2y ago
I think that transition already happened, and the next big jump in AI will be the combination of these two approaches in an unified package. Kind of the way our right brain hemisphere does probabilistic computation and the left brain hemisphere does atomistic computation. And we use both. So, whoever develops the digital equivalent of the corpus callosum wins.
- nialse 2y ago
  An observation with scientific paradigm shifts is that they tend not to reverse. Also the lingo someone commented on is that the fundamental problem is the different philosophical views of what knowledge is and can be. Either knowledge is base on symbols and rules like in mathematics or they are probabilities like in anything we actually can measure. Both these views can coexist and maybe AI will find the missing link between them some day. Possibly no human will grasp the link.

by nialse 2y ago

irunmyownemail 2y ago

Not sure I followed all of that lingo but it sounds like a fancy way of saying, if you're losing the game, try shifting the goal post.

nialse 2y ago
Indeed and unfortunately. I've been reading up on "the binding problem" in AI lately and came across a paper that hinged on there being an "object representation" which would magically solve the apparent issues in symbolic AI. In the discussion some 20 pages later, the authors confessed that they, nor anybody else, could define what an object was in the first place. Sometimes the efforts seem focused on "not letting the other team win" rather than actually having something tangible to bring to the table.

dbmikus 2y ago

I never want to claim certainties, but it seems pretty close to certain that symbolic AI loses to statistical AI. I think there is room for statistical AI to operate symbolic systems so we can better control outputs. Actually, that's kind of what is going on when we ask AI to write code.

Shorel 2y ago

I think that transition already happened, and the next big jump in AI will be the combination of these two approaches in an unified package. Kind of the way our right brain hemisphere does probabilistic computation and the left brain hemisphere does atomistic computation. And we use both. So, whoever develops the digital equivalent of the corpus callosum wins.

nialse 2y ago
An observation with scientific paradigm shifts is that they tend not to reverse. Also the lingo someone commented on is that the fundamental problem is the different philosophical views of what knowledge is and can be. Either knowledge is base on symbols and rules like in mathematics or they are probabilities like in anything we actually can measure. Both these views can coexist and maybe AI will find the missing link between them some day. Possibly no human will grasp the link.