On using the term AGIA for Artificial General Intelligence Agent

In yesterday's weblog post, we used the term AGIA for "Artificial General Intelligence Agent," assuming that it was already commonly used, but a later search revealed that it's an entirely new acronym. After checking with a few different search engines, it turns out there are only a handful of uses of the phrase "artificial general intelligence agent" on the Internet today. Surprised us, but hey it's a pretty new concept, so okay.

Looking in detail on how the phrase is used1, we find only a single instance of the capitalized-letter version, Artificial General Intelligence Agent -- with no acronym -- from 2020 and a few instances of "AGI agent." The remaining instances are all lowercase "artificial general intelligence agent" and usually appear in discussions where different types of agents are being discussed. Others do what we did, which is assume the use of these four words together is common, although it isn't. After reviewing the results of this search, it seems reasonable to assume that we coined the acronym AGIA yesterday without realizing it. Maybe it's no big deal, but we think the underlying concept is a big deal. It's new, and it will tend to be acronymized eventually. As is the way with such things, there may be more people using this term going forward, or not.

So today, we want to encourage others to use the term AGIA for a specific kind of artificial general intelligence agent. Because of our focus on decentralization, and keying on the word "agent" and its link to "free agency," we feel that AGIA are different from a more centralized artificial general intelligence agent in respect to how "independent" they are. In other words, we foresee many different flavors of artificial general intelligence agents, from the highly-decentralized very-independent model that we're developing to the strongly-centralized model that others are developing (like Siri, Alexa, Bing, and so forth), and all the range between.

We offer the acronym AGIA to refer to agents which exist at the furthest edges of a network, while others existing at the center can have another acronym or none at all. To simplify what we mean: when you see or use the acronym AGIA, imagine there is a silent "D" at the beginning, for "Decentralized." More centralized implementations could be CAGIA, or something like that.

By this we mean there is a kind of plurality embedded in the name: AGIA are agents which can exist alone, but usually exist in multiple implementations. In comparison, a more centralized model is the opposite: it can exist in multiple copies, but tends to exist as a centralized singularity, with agents that can be quite dumb because they are strongly tied to a central system where the intelligence happens.

There is a place for such centralized singularities -- a well-known reference point has its merits, when done wisely regulation is a good thing, &tc. -- but we think such implementations should always be thoughtfully limited in scope because of the risk of absolute power corrupting absolutely. To put it concisely, until all people are wise a republic is a wiser design choice than a democracy. But once everyone is wise -- an achievable goal -- a democracy would be even better2.

That bit about wisdom means we understand there is an opposite risk to centralized power which accompanies the AGIA model. Remember the Tribbles from Star Trek and how they multiplied until they became too much to handle? Funny story, but it made a good point: This kind of thing can happen. However, we feel this kind of risk is ultimately easier to manage than the centralized-power problem which humans have historically faced and have had to fight hard to overcome.

Footnotes:

1. The more interesting articles online which use variations of the term "artificial general intelligence agent": 

2. For that matter, once everyone is wise, anarchy would be even better than democracy. And would be sustainable. Who needs government when everyone is wise?

 

Page top