In a new paper published in the journal Nature Machine Intelligence, top computer scientists worldwide are discussing how recent advances in machine learning are leading towards creating a collective machine-learned intelligence. They suggest that these advancements could bring about new types of scalable, resilient, and sustainable AI systems.
Dr. Andrea Soltoggio from Loughborough University and his team see similarities between Collective AI and concepts from science fiction.
One example they mention is The Borg from Star Trek, which are cybernetic organisms connected through a shared hive-mind.
But unlike many sci-fi stories, the scientists believe that Collective AI could lead to significant positive breakthroughs across various fields.
Dr. Soltoggio explains, “With instant knowledge sharing across a collective network of AI units that continuously learn and adapt, we could respond quickly to new situations, challenges, or threats.” For instance, in cybersecurity, if one AI unit detects a threat, it can quickly share information and prompt a collective response, similar to how the human immune system defends against invaders.
This advancement could also lead to disaster response robots that can adapt quickly to different conditions, or personalized medical agents that merge advanced medical knowledge with specific patient information, improving health outcomes. The possibilities are extensive and thrilling.
However, the researchers acknowledge risks associated with Collective AI, such as the rapid spread of unethical knowledge. They emphasize a crucial safety aspect of their vision: AI units maintain their own goals and independence from the collective.
Dr. Soltoggio adds, “This would create a democracy of AI agents, reducing the risk of domination by a few large systems.”
The scientists reached this conclusion about the future of AI after analyzing recent machine learning advancements. They found that efforts globally are focused on enabling lifelong learning for AI agents and developing universal protocols and languages for AI systems to share knowledge.
This is different from current large AI models like ChatGPT, which have limited lifelong learning and knowledge-sharing abilities. These models acquire most of their knowledge during intensive training sessions and cannot continue learning.
Dr. Soltoggio notes, “Recent trends in research are extending AI models to continuously adapt once deployed and share their knowledge with other models, effectively recycling knowledge to optimize learning speed and energy use.”
The scientists believe that current large, expensive, non-shareable, and non-lifelong AI models may not thrive in a future where sustainable, evolving collectives of AI units are likely to emerge.
“Human knowledge has grown over millennia through communication and sharing,” says Dr. Soltoggio. “We expect similar dynamics in future societies of AI units, which will implement democratic and collaborative collectives.”
Source: Scinews
