Researchers have developed a groundbreaking AI inspired by the navigational skills of a sea slug and the episodic memory of an octopus. This new approach utilizes basic associative learning rules augmented with better episodic memory.
The Marine Biological Laboratory is raising octopuses to be used during scientific experiments. According to scientists, while the sea creatures are alien-looking, they have genes that are similar to humans and other animals. (Photo : Pixabay)
New AI With Enhanced Episodic Memory, Octopus-Inspired Tech
Researchers at the University of Illinois created this AI that can navigate new environments, map landmarks, seek rewards, and overcome obstacles. They built it based on their prior work simulating neural circuits similar to those in sea slugs.
Researchers developed this virtual creature named ASIMOV, inspired by Isaac Asimov, to replicate decision-making processes. Interesting Engineering reported that ASIMOV was designed to monitor its current state, seek validation, and pursue rewards in real time.
However, it struggled to learn from past experiences due to a lack of integrated information retention. To address this limitation, the researchers introduced the Feature Association Matrix, a computational module to enhance episodic memory.
This module allows ASIMOV to encode and recall both spatial and temporal contexts of past events and experiences. According to the authors, episodic memory is a crucial element of natural intelligence, a component currently absent in most AI models.
The research team studied brain network methodologies inspired by octopus behavior, dubbing their agent CyberOctopus under the ASIMOV-FAM framework.
The ASIMOV agent utilizes cognitive maps generated by the FAM to comprehend its environment, enabling it to create new pathways and efficient shortcuts for navigating its surroundings to maximize rewards. This capability represents advanced spatial reasoning.
Moreover, they said this approach can extend beyond spatial navigation to enhance efficiency and tackle more abstract, non-spatial tasks. They anticipated ASIMOV-FAM's potential to be more efficient in computation and problem-solving with minimal training.
The team envisioned a future where AI can learn autonomously and adaptively, akin to how children learn, by focusing on foundational development that reduces reliance on extensive data and fosters creativity and adaptive behavior.
The AI is designed to assist in navigating unfamiliar surroundings, identifying rewards, and creating maps of landmarks while adeptly overcoming obstacles. The team achieved this by integrating simple learning rules similar to those used by sea slugs for foraging, enhanced further with advanced episodic memory reminiscent of octopuses.
Researchers are optimistic that this innovative approach will enable AI to effectively explore and gather comprehensive spatial and temporal data, enhancing its knowledge base through practical experience.
They noted that this method makes AI more similar to animals than current models, aiming to bridge the gap between basic memory functions seen in sea slugs and more complex human-like capabilities.
The team believes this approach is significantly more effective and generates extensive data compared to existing methods. By incorporating a memory module, AI can retain past information and potentially progress from simple spatial learning to handling more complex cognitive tasks.
They also highlighted the versatility of such associative learning techniques, which could extend to tasks like understanding motor behavior sequences, mapping social networks, and solving linguistic problems.
