In a new paper in the journal Nature Machine Intelligence, leading computer scientists from around the world review recent advances in machine learning that are converging towards creating collective machine-learned intelligence. They propose that this convergence of scientific and technological advances will lead to the emergence of new types of AI systems that are scalable, resilient, and sustainable.
Saltoggio other. In other words, we will see the emergence of collective AI, where many artificial intelligence units, each able to continuously acquire new knowledge and skills, form a network and share information with each other.
Loughborough University Dr. Andrea Sortoggio and colleagues recognize striking similarities between collective AI and many science fiction concepts.
One example they give is Borg – a cybernetic life form that appears in the Star Trek universe that operates and shares knowledge through a linked collective consciousness.
However, unlike many science fiction stories, the authors envision that collective AI will bring major positive breakthroughs across a variety of fields.
“Instantaneous knowledge sharing across a collective network of AI units that can continuously learn and adapt to new data enables rapid response to new situations, challenges, and threats,” said Dr. Sortogeo.
“For example, in a cybersecurity environment, when one AI unit identifies a threat, it can quickly share knowledge and prompt a collective response, which helps the human immune system protect the body from external intruders. It’s the same as protecting it.”
“It could also lead to the development of disaster response robots that can quickly adapt to the situation they are dispatched to, and personalized medical agents that combine cutting-edge medical knowledge with patient-specific information to improve health outcomes. Yes, the potential applications are vast and exciting.”
Researchers acknowledge that there are risks associated with collective AI (such as the rapid spread of potentially unethical or illegal knowledge), but that AI units have their own objectives and independence from the collective. The authors emphasize the important safety aspect of their vision: to maintain
“This will enable democracy for AI agents and greatly reduce the risk of AI domination by a few large systems,” said Dr. Sortoggio.
After analyzing recent advances in machine learning, the authors concluded that the future of AI lies in collective intelligence.
The study focuses global efforts on enabling lifelong learning (where AI agents can extend their knowledge throughout their operational life) and developing universal protocols and languages that allow AI systems to share knowledge with each other. It became clear that it was.
This differs from current large-scale AI models such as ChatGPT, which have limited lifelong learning and knowledge sharing capabilities.
Such models are unable to continue learning because they acquire most of their knowledge during energy-intensive training sessions.
“Recent research trends are extending AI models with the ability to continuously adapt once deployed, allowing their knowledge to be reused in other models, and effectively recycling knowledge to increase learning speed and energy.” It’s about optimizing demand,” said Dr. Sortogeo.
“We believe that the currently dominant large-scale, expensive, non-sharable, non-lifetime AI models will be replaced by sustainable, evolving, and shared collections of AI units in the future. I don’t believe I will survive.”
“Thanks to communication and sharing, human knowledge has increased step by step over thousands of years.”
“We believe that similar movements are likely to occur in future societies of AI units that achieve democratic and cooperative collectives.”
_____
A. Saltoggio other. 2024. Collective AI with lifelong learning and sharing at the edge. nat mach intel 6, 251-264; doi: 10.1038/s42256-024-00800-2
Source: www.sci.news
