On Mon, Sep 12, 2022 at 5:11 PM Mike Archbold <[email protected]> wrote:
> ...
> There is a growing awareness in industry, academia, and even ML/DL
> aficionados that something more may be needed. Creating ever bigger DL
> models may not be the path
>
The real question is why the entire field of natural sciences has ignored
the proof that smaller models are better provided by Algorithmic
Information Theory for over 50 years. Only once you have answered that
question can you address the question of why machine learning is swimming
around in the same stagnant pools of ideas that conflate what "is" (natural
sciences) with what "ought" (engineering/technology/applications) to be the
case. Of course, in order for the machine learning community to address
the more fundamental question, they have to face the fact that any "AGI"
has to try to make sense of its sense-data in an unbiased manner so that it
has an accurate model of the world ("is") and therefore must embody, as
_part_ of its capabilities, operating as a natural scientist. Such mental
discipline is far beyond the capacity of "data scientists", "machine
learning experts", "artificial intelligence researchers", etc. otherwise
the Hutter Prize would have prize purse in the billions of dollars.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/Tbc5ab14e17293c6f-Meea764cf1cd033a175097b5e
Delivery options: https://agi.topicbox.com/groups/agi/subscription
