AI-Kindergarten approach to upbringing AIs through teaching

Here is the post itself.

Personally, I found the idea of an impossible singularity quite refreshing.
It does make sense that no matter how smart the system is, it still needs time to learn and master the environment.
So even an above-human artificial intelligence will not immediately cause singularity.

The core idea, however, revolves around replacing (normally big to huge) static training-based rule sets (e.g. neural network structures with trained coefficients)
with much leaner and more flexible many compact rules representation.
And somehow achieving this through training, instead of hard-coding.
And also achieving AI self-improvement (self-modification) along the way.

Unfortunately, there are no technical details in the post, and also not in the linked videos.
I haven't read linked papers, though; maybe they do have more details.

I believe that the goals are absolutely reasonable, and also not exactly new.
Interaction/training-based generation of intelligent systems is a concept with long history,
and there are other projects which try to generate intelligent behavior through interactions with a (simulated) environment.
Dr. Danko Nikolic is also quite clearly highly-engaged in his project.

However, a combination of non-tech content delivery and the lack of any kind of demos or proofs of principle implementations seriously undermines this effort.
Regarding demos/proofs, the GAiuS system (General Artificial Intelligence using Software) looks much, much more convincing.

Syndicate content