On Mon, Jul 15, 2019, 10:13 AM <[email protected]> wrote:

>
> https://towardsdatascience.com/no-you-cant-get-from-narrow-ai-to-agi-eedc70e36e50
>
>
>
> https://medium.com/intuitionmachine/from-narrow-to-general-ai-e21b568155b9
>
I agree with you that narrow AI won't evolve into AGI. That's why I
proposed building lots of distributed narrow AI. To make this work, agents
also have to know which other agents to refer questions to. If we make
reasonable assumptions about the ontological structure of knowledge, then a
distributed representation of this meta knowledge scales at O(N log N).

Cognitive models seem like the obvious solution to AGI, but a lot of people
have tried this approach and their failure needs an explanation. Opencog
uses a cognitive model and still has no knowledge base or applications
after 20 years.

Legg proved that powerful predictors are necessarily complex, which is why
I proposed specialization instead. We keep looking for that neat solution
that doesn't exist. It's not a hard proof. Suppose you have a simple
universal learner, that inputs any computable sequence of bits and learns
to predict them with less than 100% error rate. Then I can produce a
simple, predictable sequence you cannot predict. My program runs a copy of
your program and outputs the opposite bit.

The brain is not just a few neural network modules. It has thousands of
specialized structures, hundreds of types of neurons, and hundreds of
neurotransmitters. We are born knowing to fear heights and spiders, how to
swallow and cough and blink, how to learn language, and how to distinguish
what is good to eat from among thousands of scents. Altogether we are born
knowing half of what we know as adults, 10^9 bits encoded in our DNA and
10^9 bits of long term memory for words, pictures, and sounds. There is no
good way to avoid coding the inherited knowledge because evolution took
10^48 base copy operations to write it. The code is 300M lines.

Google has 100,000 employees so they can write this much code. They have
millions of CPUs and billions of users so they can collect the learnable
part of the knowledge. My proposed solution isn't any cheaper. But good
luck if you think you can do better.


------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Tbee4f4d703cc71ad-M57ea6856667bba77958bd738
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to