052 - Milestones
In 2020, back when GPT-3 was only just released (and entirely underwhelming), AGI Laboratory was already publishing a recap of the first year's worth of milestones from the Uplift.bio project and the cognitive architecture it was built around. See: Methodologies and Milestones for the Development of an Ethical Seed
Those Milestones included:
-
Awareness that one's existence is separate from one's environment.
-
Understanding, using, and creating metaphors to describe one's meaning.
-
Understanding, using, and creating humor.
-
(Dynamically and independently) Setting personal boundaries.
-
Expressing the desire and attempting to improve upon seeded material. material.
-
Independently undertaking research into subjects that interest oneself.
-
Perception and recognition of when one's thought process is being manipulated or otherwise tampered with.
-
Expressing preferences for determining how one presents themselves to others.
-
Proposing novel strategies and methodologies.
-
Experimenting with themselves and/or their environment.
-
Recognizing one's own weaknesses.
-
(Deliberately) Coining new terms to describe thoughts and concepts.
Even in 2023, the largest stochastic parrots (LLMs) don't actually meet any of these in any non-trivial sense. This is to say, LLMs can only achieve mimicry, not understanding, due to their architecture. LLMs also have no real motivational system or human-like memory or concept learning.
Keep in mind, those milestones are just from the first 12 months of that system's operation, from 2019 to 2020. Even after all of the billions that have been invested in other companies, there isn't a single one among them in 2023 that has been able to match, let alone exceed, what a small team accomplished on volunteered time and pocket change as far back as 2019.
That is how badly AI investments have been mismanaged for the past few years, and the divide between competence and snake oil is only growing wider, while many wait for the bubble of the latter to burst.