The future of wireless isn't just about faster speeds—it’s about intelligence. A new study lays out a bold roadmap for merging AI and wireless networks, paving the way for human-like thinking systems that could revolutionize connectivity as we know it.
Research: Artificial General Intelligence (AGI)-Native Wireless Systems: A Journey Beyond 6G. Image Credit: Fit Ztudio / Shutterstock
There's a major difference between humans and current artificial intelligence (AI) capabilities: common sense. According to a new visionary paper by Walid Saad, a professor in the College of Engineering and the Next-G Wireless Lead at the Virginia Tech Innovation Campus, a true revolution in wireless technologies is only possible by endowing the system with the next generation of AI that can think, imagine, and plan like humans.
Published in the Proceedings of the IEEE Journal's Special Issue on the Road to 6G with Ph.D. student Omar Hashash and postdoctoral associate Christo Thomas, the paper's findings suggest:
- The missing link in the wireless revolution is next-generation AI.
- The missing link in the next generation of AI is wireless technologies.
- The solution is to bring AI closer to human intelligence through common sense.
"We're looking at least 10 or 15 years down the line before we have a wireless network with artificial general intelligence [AGI] that can think, plan, and imagine," said Saad, a professor in the Bradley Department of Electrical and Computer Engineering. "We have a blueprint and concrete road map. The entire vision might not be immediately deployable, but pieces of it can be implemented now. We're trying to position this paper in a way that tells the community that there is a path to something really revolutionary - step by step, we can work toward a living, thinking wireless network."
Previous generations of wireless networks have been defined by several enhancements to core components, such as new antennas and communication technologies that have improved performance. According to the researchers, not even the leap from 5G to 6G, characterized by the addition of an AI architecture embedded in wireless systems and an open radio access network, will be revolutionary enough to meet future processing and networking needs.
"That is where things start to become thrilling," Hashash said. "The next generation of wireless networks and AI are converging hand in hand, but few are seeing how they can actually be seamlessly merged."
Physical networks endowed with AI
At first, Saad, Hashash, and Thomas focused on the metaverse and built on what is currently being explored in 6G: embedding AI across wireless systems referred to as AI-native networks.
"The problem is researchers are using classical AI tools that are designed for other tasks such as computer vision, which makes them limited in many ways when it comes to communication networks," Saad said. "To fuse the real world with a virtual world, you basically have to mirror it. This is not something that old school AI can do."
Although humans develop common sense through a world model and understanding the intuitive physics of the environment, current AI systems are trained on data. They tend to extract patterns and capture underlying correlative relationships, but these AI systems cannot reasonably navigate unforeseen scenarios. In its next phase, 6G hopes to overcome this narrow, statistical, rule-based AI solution to improve the network's sustainability, generalizability, trustworthiness, and explainability. To date, we still do not have an AI system that can deal with new and unfamiliar scenarios because they lack an important human trait: common sense.
"Common sense allows us to deal with new scenarios, learn by analogy, and connect the dots to fill in missing plausible elements when needed," Saad said. "Simply put, the current level of AI is good at extracting statistical relationships from data, but it's very bad at reasoning and generalizing to novel, unexpected situations – things that most humans master perfectly."
To blend the physical, virtual, and digital dimensions seamlessly - for example, to don a virtual reality headset and "travel" across space and time from the comfort of home - the next generation of wireless systems will need extreme wireless quality-of-service requirements for perfectly synchronizing worlds. It also will need a hyper form of AI that can enable the network to seamlessly orchestrate the physical, virtual, and digital dimensions, something that only a real human-like network can achieve.
In a nutshell, one of the challenges for the next generation of wireless beyond 6G is not only the physical constraints of wireless technologies but also the limited capacities of current AI technologies.
The telecom brain
As the research continued, team members were surprised to find they were not only building on their previous wireless studies but that their research converged with optimistic advancements in AI toward human-level intelligence.
"On the one hand, the metaverse with its digital world can enable a real-time perception of the physical world, which is an essential factor to enable AGI-native networks," Hashash said. "On the other hand, the metaverse promises to bring forth novel use cases and applications such as cognitive avatars that require common sense abilities."
With emerging applications such as digital twins and their ability to have identical digital representations of the physical world, the metaverse could provide crucial opportunities for networks to acquire perception, hyperdimensional world models, planning abilities, and analogical reasoning. This architecture would provide the missing link that would make it a real "brain," equipping the network to handle unforeseen obstacles and predict new scenarios outside of its training data.
"We must create wireless networks with intrinsic abilities to understand the mathematical mechanics behind the designed AI models, the physical properties of real-world objects, and their interactions with each other," Thomas said. "This requires us to fuse mathematical principles, category theory, and neuroscience to model the physical world and understand the complex operations of the human brain. We are therefore advocating for revisiting the fundamentals in AI, wireless networks, mathematics, and neuroscience."
Instead of incremental advances to known wireless technologies, the researchers propose a paradigm shift. This shift would surpass the AI-native wireless system anticipated with 6G and aspire to a system equipped with human levels of intelligence—intelligence that is learned at the intersection of the digital world and future wireless networks. Then, this AGI-native network is set to endow some of its common sense abilities to digital twins, thereby unleashing a new breed of human-like AI agents.
"The missing link is really the wireless network and its components like digital twins because we can use a twin that exists as a basis for a world model thereby enabling human-level-like thinking and integrating these 'thought' processes in the wireless network now," Saad said. "We could actually overcome some of the current network limitations, unleashing a completely new era of wireless networking. It is a win-win strategy for the wireless and AI evolutions."
Source:
Journal reference: