6G and Physical Intelligence: The Transition from Connectivity to Autonomous Interaction

Author: Mallik Tatipamula, FRS FREng FRSE

My previous CHEDDAR article (https://cheddarhub.org/blog/from-iot-to-physical-ai-how-6g-transforms-connected-things-into-intelligent-environments-2/) explained Physical AI as the evolution beyond IoT at the application level. This note complements it by demonstrating that this shift arises from a deeper architectural transition in networks, from packet delivery to workload execution to verifiable multi-agent interaction explaining why 6G must be designed as coordination infrastructure rather than merely connectivity infrastructure.

Abstract

Mobile networks are commonly interpreted as communication systems whose evolution is primarily driven by advances in radio performance, such as higher throughput, lower latency or improved spectral efficiency. Historical deployment patterns, however, suggest an additional engineering explanation. Each generational transition has corresponded to the introduction of a higher-level architectural abstraction that removed a fundamental scalability barrier in distributed systems.

The packet abstraction enabled the mobile Internet; the workload abstraction enabled cyber-physical automation; and emerging 6G architectures introduce an intent- and context-aware interaction abstraction required for autonomous systems. This progression reflects a transition in the role of networks from connecting humans to machines, and now towards enabling cooperation among independent intelligent agents operating in the physical world.

The concept of Physical Intelligence, therefore, cannot be understood as an extension of IoT; rather, it requires networks capable of maintaining shared understanding, trust and verifiable interaction across heterogeneous autonomous participants. This note argues that 6G represents the first globally deployed infrastructure for continuous real-time coordination, rather than communication alone.

For much of its history, telecommunications has been defined as the transmission of information between endpoints. In this interpretation, successive generations of mobile systems are understood as incremental improvements in radio access technology. Although improvements in spectrum efficiency have certainly been necessary, they have not been sufficient to explain the qualitative changes in what networks have been able to support.

The introduction of mobile broadband, large-scale industrial automation and now autonomous operation did not occur simply because links became faster, but because networks began to operate at progressively higher levels of system abstraction.

The relevant distinction is therefore not between analogue and digital, nor between 4G and 5G radio capabilities, but between the level at which the network participates in the operation of distributed systems. In early mobile networks, the network transported packets; in modern networks, it participates in computation; and in emerging systems, it must participate in decision-making. This shift is best understood as an architectural continuum in which each generation resolved a different scaling constraint in large, distributed environments.

The Packet Abstraction and the Internet of People

The first large-scale transformation occurred when mobile networks adopted an IP-native architecture. Prior cellular systems were service-specific, optimised for voice delivery and tightly coupled to circuit-switched operation. As data services emerged, the challenge was not bandwidth but convergence: a single infrastructure had to support real-time and non-real-time applications with widely varying requirements.

The packet abstraction solved this problem by separating service semantics from transport. Mechanisms such as header compression, radio-to-core quality-of-service mapping and IPv6 mobility enabled wireless links to function as part of a unified global packet network. This allowed voice, video and data to coexist within a common framework, effectively integrating mobility into the Internet rather than extending telephony into data.

In this phase, the responsibility of the network was reachability. Applications interpreted meaning, while the network ensured packets were delivered between human users. The resulting system can be described as an Internet of People, in which the primary function of the infrastructure was communication rather than operation. The network remained outside the logic of the application, even though it enabled its scale.

The Workload Abstraction and the Internet of Things (IoT)

The next barrier emerged when connected systems began to interact with the physical world. Industrial automation, robotics and distributed control systems required predictable timing behaviour. Variability in latency became more problematic than limited throughput because delayed information could produce incorrect physical action.

Addressing this required the network to become part of the execution environment rather than merely a transport medium. The introduction of cloud-native decomposition,  including separation of control and user planes, network slicing and distributed cloud placement allowed computation and networking to be jointly orchestrated. Functions could now be located where timing constraints required, rather than where topology dictated.

The network therefore, evolved from carrying data to hosting workloads. Closed-loop machine control depended on the network’s ability to maintain deterministic behaviour across distributed components. This marked the emergence of an Internet of Things in which the network was no longer external to the system being operated but became a constituent element of it.

Importantly, the Internet of Things did not arise simply from connecting sensors. It emerged because machines could rely on the network to operate safely. The defining property was operational dependency rather than connectivity.

The Intent Abstraction and the Internet of Agents

A new scaling barrier appears when autonomous systems share environments. Independent intelligent entities must coordinate actions rather than merely exchange information or execute predefined workflows. In such environments, correctness depends on agreement among participants rather than individual execution accuracy.

Autonomous vehicles negotiating intersections, collaborative robots sharing workspaces and aerial systems coordinating airspace cannot rely solely on connectivity or deterministic latency. They require persistent knowledge of identity, state and intention across distributed actors. Without shared understanding, safety cannot be guaranteed.

Emerging 6G architectures therefore, introduce interaction semantics at the network level. The network maintains continuity of context across time and location, enabling participants to reason consistently about one another’s actions. This can be interpreted as cognitive continuity: the infrastructure preserves the operational state necessary for cooperation.

The resulting system can be described as an Internet of Agents. The network no longer only transports information or executes workloads but enables agreement among independent decision-making entities. This represents a qualitative change in the role of infrastructure from communication platform to coordination substrate.

Physical Intelligence

Artificial Intelligence has largely developed in environments where errors affect digital outcomes. Physical Intelligence differs because decisions directly influence the physical world. Incorrect coordination becomes collision, and inconsistent interpretation becomes unsafe behaviour.

In such systems, local intelligence is insufficient because each participant observes only part of the environment. Safe operation requires shared situational awareness. Only the network continuously observes all actors and interactions, making it uniquely positioned to maintain global consistency.

Consequently, the function of the network changes fundamentally. Rather than connecting endpoints, it establishes a shared reality among autonomous participants. Trust, identity and intent become networking properties rather than purely application concerns. This is the defining requirement introduced by Physical Intelligence and explains why 6G cannot be understood simply as enhanced IoT.

Implications for Future Research

The transition to agent interaction moves networking into domains traditionally associated with distributed systems, control theory and economic coordination. Communication performance remains necessary, but no longer defines the primary problem. The central challenge becomes guaranteeing correct collective behaviour among independent entities whose decisions influence shared physical environments.

Initial questions concern the persistence of identity, continuity of context and the verification of interactions before physical consequences arise. However, once autonomous participants begin optimising independently, additional challenges emerge that are not reducible to latency or reliability. Systems that are individually correct may collectively become unstable. Multiple optimisation processes acting simultaneously on shared resources can produce oscillations, congestion waves or unsafe actions, even when each participant follows its own policy accurately.

Future research must therefore address the stability of distributed autonomous control, ensuring that coordination mechanisms prevent emergent behaviour from diverging despite decentralised decision-making.

A second challenge concerns semantic interoperability. Packet networks achieved interoperability through standardised syntax, and application protocols through shared formats, but cooperative autonomy requires shared meaning. Autonomous systems must interpret intent consistently across vendors, industries and jurisdictions. Two agents may exchange valid messages yet disagree about acceptable risk or priority. Research must therefore investigate machine-interpretable representations of intent, negotiation semantics and conflict resolution frameworks that allow heterogeneous agents to reach agreement rather than merely exchange information.

The presence of multiple stakeholders introduces a third dimension: incentive compatibility. Autonomous participants will often represent independent economic actors whose objectives may not align. In such environments, cooperation cannot be assumed. The network must coordinate behaviour among rational agents who may benefit from selfish action. This requires mechanisms that make safe interaction economically rational, combining networking with ideas from mechanism design and distributed markets. The problem becomes not only whether agents can coordinate, but whether they will.

Trust consequently shifts from authentication towards accountability. Traditional networks rely on centralised authorities to validate identity and behaviour, but large-scale autonomous environments span administrative and sovereign boundaries. Future infrastructures must support verifiable interaction histories, distributed attestation and cross-domain arbitration without requiring universal ownership. The network must provide evidence of what occurred, not merely ensure that communication occurred.

Another emerging requirement is temporal consistency of shared reality. Autonomous systems depend on a common understanding of event ordering. Minor disagreement in timing may translate into conflicting actions in the physical world. Synchronisation therefore, extends beyond clock alignment to maintaining causality across distributed participants. Research must explore how networks maintain coherent temporal interpretation when actions occur simultaneously across locations.

Finally, certification must evolve from component safety to interaction safety. Current approaches validate individual devices under predefined assumptions, yet failures in autonomous environments often arise from interaction among correct components. Networks must enable verification of collective behaviour, detecting unsafe patterns before they manifest physically. This implies runtime validation of system behaviour rather than static certification of isolated elements.

Taken together, these challenges redefine networking as infrastructure for cooperative autonomy rather than communication. The research objective is to enable independently designed systems to coexist safely in shared environments. From this perspective, 6G represents not merely an extension of communication technology but the foundation for verifiable interaction among intelligent participants operating in the physical world.

Conclusion

Mobile networks have evolved from information transport systems into operational infrastructure for society. The packet abstraction enabled global communication; the workload abstraction enabled automation; and emerging intent-aware interaction enables autonomy. Each step expanded what distributed systems could reliably accomplish.

6G therefore represents the first infrastructure whose primary purpose is not merely transferring information but maintaining coordination among intelligent entities operating in the physical world. The significance of Physical Intelligence lies not in the intelligence of individual devices, but in the ability of independent systems to act together safely. Achieving this requires networks designed for verifiable interaction, establishing the foundation for an Internet of Agents.