Implementing Agent-Human Collaboration Using GPT-4o
Key point
Altera has developed digital humans capable of long-duration autonomous operation by combining GPT-4o with a system that mimics brain structure.
Details
Altera is building 'Digital Humans' that can collaborate and form bonds with humans. Beyond simple assistive tools, they aim to become digital colleagues that solve problems over extended periods, or multi-agent environments that can measure economic policy and advertising responses.
The phenomenon of Data Degradation, which occurs when agents operate autonomously for long periods, was a key challenge. This is a problem where data quality declines over time as AI's output is reused as input, becoming a major cause that hinders long-term autonomy.
To solve this, Altera combined GPT-4o with a parallel multi-module system that mimics the structure of the human brain (such as the frontal lobe). This system enhances the agent's cognitive abilities by mimicking the following brain functions:
- Attention bottleneck
- Working memory
- Social cognition
The agent developed through this can play together like a friend in Minecraft, and as of mid-2024, it demonstrates performance capable of operating autonomously for up to 4 hours.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.