I would like to introduce a new term: Collaborative Artificial General Intelligence, or CAGI. I believe that as a society, we should advocate for CAGI systems rather than AGI systems.
The race to develop a single system that approaches human-level capacity, capable of performing a broad range of tasks and continuously improving its capabilities, is ongoing. It is estimated that achieving this milestone will take over 10 years.
CAGI can be defined as the collaboration of hundreds, thousands, or even hundreds of thousands of independent AI agents, each with specific expertise and different personas and perspectives. By collaborating, these agents can collectively reach AGI levels. (Each agent can use internal collaboration techniques, such as a mixture of experts or composition of experts, to create a better “memory structure.”)
I would argue several points:
1️⃣ From a usability perspective, reaching CAGI is practically the same as reaching AGI. We can always encapsulate a complex system with numerous components into a black box and call it a single system.
2️⃣ CAGI can be achieved much faster than AGI, likely within the next few years.
3️⃣ CAGI will always be computationally more favorable and environmentally more sustainable than AGI.
4️⃣ Deploying guardrails on CAGI systems is significantly easier than on AGI systems, providing more explainable monitoring capabilities for human observers.
Even for CAGI, new fundamental technologies for each agent beyond LLMs need to be developed.
At Wand, we are pioneering the path to CAGI. Additionally, our fundamental research group is working on a new fundamental technology, which we have termed “Cognitive Language Models” (CLMs). These models provide powerful reasoning and planning capabilities, sub-exponential divergence over time, built-in hallucination control, and dynamic personality building.
#AI #FutureofAI #ML