EDITOR'S PERSPECTIVE

Decentralized AI: Edge Intelligence

and Smart Blockchain, Metaverse, Web3,
and DeSci
Longbing Cao , University of Technology Sydney, Sydney, NSW, 2007, Australia

Centralization has dominated classic scientific, social, and economic

developments. Decentralization has also received increasing attention in
management, decision, governance, and economics, despite its incomparability in
AI. Going beyond centralized and distributed AI, this article reviews and delineates
the conceptual map, research issues, and technical opportunities of decentralized
AI and edge intelligence. The complementarity and metasynthesis between
centralized and decentralized AI are also elaborated. We further assess where
decentralized AI and edge intelligence can enable and promote smart blockchain,
Web3, metaverse and decentralized science disciplinarily, technically, practically,
and more broadly.

D
ecentralization1 complements and enhances decentralization-oriented initiatives, such as Web3
centralization for systematic, all-round, and and the metaverse. Decentralization is also not a
multifold objectives, functionality, and conse- new or mysterious concept.1 It has a 200-year his-
quences. The paradigm shift from centralization to dis- tory and is grounded in political science (such as
tribution has substantially mitigated methodology, subsidiarity, democracy, liberty, equality, and the
knowledge, and capability gaps in social science, man- decentralist movement), management and decision
agement science, decision science, economics, comput- science (e.g., systems theory, self-organization, and
ing, complex systems, and AI. On the other hand, open- self-determination), and economics (e.g., decent-
source and science have made paramount complemen- ralized free markets and fiscal decentralization).
tary achievements and contributions, and blockchain is Decentralization (with forms, such as devolution,
transforming and inspiring finance to decentralized deconcentration, delegation, etc.) is complementary
finance (DeFi), and the World Wide Web to Web3. In con- or an alternative to centralization (and concentra-
trast, decentralizing AI is still an open area. By reviewing tion), such as decentralized organizations, infrastr-
decentralized movements, systems, and technologies, uctures, administration, operations, and services,
including the recent movements on the blockchain, corresponding to their centralized counterparts. In
Web3, and decentralized science (DeSci), we paint a contrast to centralizing resourcing, manufacturing,
research picture of decentralized AI (DeAI), the research and supply, globalization partitions, distributes, and
issues in edge intelligence, and the tasks of synthesizing decentralizes industrial, manufacturing, and supply
centralized AI (CeAI) and DeAI. These further envisage chains to usually isolated and local economies and
and inspire the opportunities of DeAI-enabled smart productivity. Open society, open government, and
blockchain, Web3, metaverse, and DeSci. open science further promote the decentralization
of administration, governance, governmentality, and
scientific activities.
DECENTRALIZATION
Decentralization is not a buzzword, hype or myth,
although such concerns have been widely raised on
Distributed to Decentralized Systems
In technology, the thinking and conceptualization of
technological decentralization have evolved for
1541-1672 ß 2022 IEEE over half a century. A flagship decentralized IT
Digital Object Identifier 10.1109/MIS.2022.3181504 movement is the open and shareable initiatives and
Date of current version 22 July 2022. programs, represented by open source, open data,

6 IEEE Intelligent Systems Published by the IEEE Computer Society May/June 2022
 EDITOR'S PERSPECTIVE

open access, and open science.a A representative virtually, mentally, or jointly, resulting in evolving block-
early stage distributed and decentralized computing chain, Web3, the metaverse, and DeSci movements.
flagship is the formation of Linux operating systems Blockchain: A blockchainb,4,5 is a decentralized net-
and their ecosystems. This movement has been further work where cryptographically hashed blocks (digital
intensified by public and private cloud services for records) are chained one to another, and each node rep-
decentralized infrastructures, storage, sharing, com- licates its previous blockchain. A blockchain is a P2P
puting, applications (DApps), and services. In IT prac- network for DeFi, where each node replicates, stores,
tices, a recent trend is that an increasing number and updates an identical ledger (called a distributed led-
of organizations have removed or are decentralizing ger). The distributed ledger database thus records, vali-
their existing centralized IT divisions, substituted by dates, manages and shares the consensus of the
subscribing to vendor-centralized cloud-based infra- chained digital information across the blockchain eco-
structures, computing, applications, and services. Fur- system. Proof-of-work allows new blocks to join, secures
thermore, edge computing2,3 emerges to connect a blockchain, and serializes changes in the chain. Proof-
isolated end-level devices and the Internet of Things of-stake builds consensus over a blockchain. Transac-
(IoT) through edge nodes (networks) rather than cen- tions of digital assets (called cryptocurrencies) are
tralized cloud infrastructures and to support decentral- recorded by blockchains, and smart contracts allow
ized computation, communication, storage, sharing, asset contracting between entities to be executed on
and management at end devices or edge nodes. blockchains without human interaction. In blockchain
However, the early version of decentralized comput- ecosystems, blockchains transfer assets between chains
ing, systems, and services is also not new. They have through blockchain interoperation. Consequently, block-
been built on the conceptual systems and success of chain is decentralized without a central administrator,
distributed systems, computing, algorithms, and serv- server, or trusted authority for authentication, gover-
ices and more recent successes, such as in distributed nance, communication, management, and mediation,
AI (DAI) and peer-to-peer (P2P) and device-to-device etc. Accordingly, the blockchain architecture, identity,
(D2D) systems. P2P, D2D, and team-driven activities, communication, contracting, interoperability, manage-
such as open science, crowdsensing, crowdsourcing, ment, governance, security, and privacy differ signifi-
and crowdfunding, further relax distributed couplings cantly from centralized networks and present unique
between nodes and with centralized or shared facilities decentralized characteristics, properties, challenges,
(such as shared memory or databases). Here, peers and and opportunities.
end devices replace shared central controllers, servers, Web3: The Web3 movementc aims to transform the
mediators, or matchmakers. P2P and D2D authentica- static, consumer-oriented Web 1.0 and the dynamic, pro-
tion, collaboration, and communication at running time ducer and platform-oriented Web 2.0 into a decentral-
replace the design-time protocols and specifications for ized web ecosystem. Some initial Web3 trials are
authentication, collaboration, cooperation, communica- typically implemented in terms of decentralized autono-
tion, and governance. This shift from distributed loose mous organizations (DAOs), blockchain technologies,
couplings to ad hoc and running-time P2P interactions cryptocurrencies, and other decentralized technologies
enhances the effect of decentralization. Blockchain, for exchanging digital assets over the web. It is envis-
Web3, and DeSci thus emerged, and the metaverse aged that Web3 may evolve to be a decentralized web
upgrades the initiatives of digital twins and virtual reality. ecosystem, which could 1) mitigate issues in the existing
Web 2.0, such as overcentralization and domination, by
big tech companies, network vulnerability, misinforma-
Blockchain, Metaverse, Web3, and tion and information disorder; 2) provide more secure,
DeSci private, scalable, and free information creation,
Blockchain and Web3 further decouple the dependence exchange, sharing, and transfer; 3) enable the creation,
on and interactions with central or intermediary authen- identification, contracting, exchange, trade and man-
tication, organization, computation, communication, and agement of public or private content, products (such as
mediation. Decentralized operations, authentication, digital assets), and services (e.g., financial services); and
communication, and collaborations engage scalable 4) supporting immersive web development and experi-
decoupled and loosely connected individuals and local ence by the decentralized end, and edge devices.
nodes to form decentralized ecosystems physically,

b
[Online]. Available: htt_ps://en.wikipedia.org/wiki/Blockchain
a c
[Online]. Available: htt_ps://en.wikipedia.org/wiki/Open_science [Online]. Available: htt_ps://en.wikipedia.org/wiki/Web3

May/June 2022 IEEE Intelligent Systems 7

EDITOR'S PERSPECTIVE

Metaverse: The metaverse6 coined in Snow Crash or myth, typically driven by commercial narratives and
has been substantially enriched in terms of its vision, purposes. The existing initiatives and activities are
mission, conceptualization, implementation, and appli- inclined to motivation, vision, conceptualization, story-
cations over the past 30 years. The metaverse journey telling, and specific settings and applications building
has evolved from digital twins represented by the on the blockchain, cloud techniques, and sometimes
online virtual world Second Lifed to creative games simple applications of AI. One critical reason lies in their
empowered by virtual and augmented reality and ignorance of or weakened attention to identifying, defin-
more recently online virtual and multiplayer games ing, and exploring decentralization-centric fundamental
enabled by social interactions, training and trading challenges, research questions, and foundational
creatures, and virtual economy with nonfungible enabling technologies. This motivates this article to dis-
tokens. The renaming of Facebook to Meta has further cuss the concepts, aims, research issues, and opportu-
substantially accelerated the attention, debate and nities of DeAI and its roles in enabling smart blockchain,
exploration as to what it is, how it works, and where Web3, the metaverse, and DeSci.
it applies, etc. Technically, the metaverse pursues
1) a comprehensive virtualized or virtually-physically
fused immersive experience; 2) an online or cloud-to-
DEAI: DECENTRALIZED AI
device 3D interactive world with human and device inter- Why DeAI?
facing and teaming; 3) a vivid application for demonstrat- From Centralized to DAI and DeAI: Conventional AI
ing the metaverse design, applications and services; 4) a aims, technologies, systems, and services are central-
rich ecosystem with trading markets, financial services, ized, static, predefined at design time, and user-neu-
economic activities, and social services; and 5) a suite of tral. CeAI presents their outputs as either a
enabling technologies for infrastructure, interfacing, standalone system or a multiagent system. When a
interaction, design and development tools. multiagent system carries the AI task, CeAI usually
DAOs and DeSci: The early progress of decentral- has a central controller, mediator, matchmaker, coor-
ized blockchain and Web3 has promoted a new age of dinator, or communicator (for message passing) to
decentralized movements, initiatives, and concepts. manage and govern agents and task execution. CeAI
DAOse and DeSci7 represent their new trend toward met with fundamental issues, such as inflexibility, high
broad-reaching decentralized organizations, opera- vulnerability, poor reliability, and low scalability and
tions, services, and implementations. DAOs and DeSci adaptability, to complex real-world demand and large-
have witnessed various initiatives, such as decentral- scale problem solving. The CeAI thinking does not fit
ized cryptocurrency, finance (DeFi), biomedicine, fund- the individualized nature and personalized demand/
ing, review, professional services, and the metaverse. supply in the physical and social worlds. DAI then
Technically and operationally, blockchain, Web3, and emerged, followed by DeAI to further flatten DAI.
other decentralized technologies, platforms and services Distributed AI involves multiagent systems for par-
provide essential and prerequisite infrastructures, pri- allel or distributed problem solving. A DAI system
vacy and security assurance, communication, token- decomposes a problem or task to subproblems or sub-
based assets and valuation,8 and individual accessibility tasks and then assigns them into specialized agents
and scalability to enable DAOs and DeSci at large and at for handling; the agent-based partial solutions are
scale. Other important enablers and supporting techni- then synergized to form the solution to the problem.
ques include decentralized cloud computing, edge com- While taking a bottom-up approach, DAI usually
puting, smart contracting, P2P systems, autonomous involves mediation, cooperation, coordination, and
agents, distributed and cloud-based analytics, such communication between agents for task decomposi-
as federated and edge learning, and technologies for tion, assignment, execution, and consensus-building.
the metaverse. Orchestration and consensus-building mechanisms
Despite the intensive foci on blockchain and its further consolidate distributed agents, logic, tasks,
applications, the decentralized movements, including and results for solution aggregation or formation.
Web3 and DeSci, have not attracted wide scientific Therefore, DAI is not truly decentralized, nor the clas-
attention in academia. Some even believe they are hype sic parallel and distributed computing.
The emergence of DeAI addresses the thinking
and technological gaps in CeAI and DAI for genuine
d
AI decentralization, including decentralizing multi-
[Online]. Available: htt_ps://en.wikipedia.org/wiki/Second_Life
e
[Online]. Available: htt_ps://en.wikipedia.org/wiki/Decentralized_ agents, AI tasks, task execution, resource allocation,
autonomous_organization storage and management, transaction management,

8 IEEE Intelligent Systems May/June 2022

 EDITOR'S PERSPECTIVE

intelligence management, and consensus-building coordinate, cooperate and communicate with other
between decentralized agents and AI tasks, etc. On nodes in proximity to share, exchange, and assist in
one hand, DeAI promotes the evolution and devel- tasks, resources, and intelligent capabilities. Nodes
opment of AI toward individualized, personalized, act on their own, and there are no controlling,
distributed, secure, privacy-preserving intelligence authoritative, or coordinating nodes in the node
and intelligent problem solving. On the other hand, network.
DeAI also aims to incorporate intelligence and intel- Decentralized AI: DeAI refers to the AI thinking,
ligent problem solving into the existing decentral- methodologies, technologies, systems, and services for
ized systems and their enabling platforms, such as developing, managing, and deploying decentralized
blockchain and Web3, where AI has not played a intelligence in decentralized settings; storing, updating,
major driving role and they are not smart enough. sharing, and exchanging decentralized intelligence
DeAI thus will fundamentally promote the develop- between decentralized agents, nodes, or devices; and
ment of intelligent decentralized technologies and integrating decentralized intelligence from local agents
systems and escalate decentralized movements, and across decentralized ecosystems (with their serv-
including Web3, DAOs, and DeSci. ices, and environments) for higher level intelligence and
DeAI, blockchain, DAOs, DeSci, and Web3 require intelligent problem solving.
and take advantage of local capabilities, resources, high DeAI aims: Accordingly, the research aims and
computation, high bandwidth, high privacy, high secu- objectives of DeAI include but are not limited to:
rity, low energy, and low latency in decentralized devices
and contexts. Without sharing resources, capabilities › studying fundamental and unique characteris-
and computation with centralized servers and media- tics and properties in making decentralized sys-
tion by controlling the authority, DeAI maximizes the tems intelligent, for example, enabling smart
power of end and edge devices and resources. blockchain, Web3, DAOs, and DeSci systems,
and services; their characteristics and properties
What is DeAI? may influence the decentralized architectures,
Decentralization in AI: Decentralization is presented in functions, operations, production, services, qual-
various aspects and on different levels of AI systems. ity assurance, and risk mitigation, etc.;
› quantifying decentralized interactions, complexi-
1) Decentralized intelligent devices and agents on ties, and intelligence in DeAI systems, which form
the edge level: referring to the lowest individual- some of the most fundamental system character-
ized intelligent agents. Each device alone could istics and properties in DeAI systems, and drive
be an intelligent, autonomous agent with stand- and enable smart blockchain, Web3, DAOs, and
alone universal or specific AI functions, forming DeSci. Interaction, complexity, and intelligence
edge intelligence in terms of the capabilities of cause or influence their fundamental frameworks,
thinking, emotion, perception, vision, conversa- working mechanisms, evolution, performance,
tion, interaction, action, and decision-making. and risk; quantifying them thus makes it possible
2) Decentralized local intelligent devices and to characterize, compute and manage decentral-
agents on the node level: referring to edge devi- ized systems in a quantitative manner;
ces in a local network area (such as the smart › developing key enabling techniques for designing
home) where they form a node (a local intelligent and producing intelligent decentralized systems
network). Nodes serve as an intelligent agent and services, and making blockchain, Web3, DAOs,
cluster to fulfill node-level AI functions. DeSci, and their ecosystems intelligent with smar-
3) Decentralized cooperation, coordination, and ter autonomy, self-organization, and resilience;
communication on the end level: end agents › enabling the integration and integrity of plugging
may interact, cooperate, coordinate, and com- in third-party AI systems, tools, and services to
municate with each other in terms of on-device decentralized platforms, services and DApps;
cooperation and coordination mechanisms, › identifying and addressing systematic issues,
communication protocols, and resource man- risks and exceptions of decentralized platforms
agement without coordinating, controlling, or and services in an adaptive and autonomous
authoritative management between devices. manner to enhance their reliability, resilience,
integrity, and adaptability;
On the edge level, nodes perceive, reason, and › enabling, driving, and delivering new value prop-
act on their own goals, tasks, and expectations and ositions (e.g., economic and social benefits) of

May/June 2022 IEEE Intelligent Systems 9

EDITOR'S PERSPECTIVE

FIGURE 1. DeAI: Synthesizing on-device AI, edge AI, and cloud AI. A P2P on-device AI. B Data-sharing edge AI. C Model-sharing
f
edge AI. D Device-edge-cloud hybrid edge AI.

decentralized technologies, platforms and serv- DeAI research issues: DeAI systems may comprise
ices by advancing their intelligence, personalized many AI agents who take different roles to achieve
services, autonomous decisions, and optimal the abovementioned aims. Examples are local AI
performance, etc. agents for executing local or specific AI tasks or tailor-
ing AI algorithms or models for local AI tasks. A large
In summary, DeAI aims to develop decentralized DeAI ecosystem may consist of multiple DeAI edge
intelligent systems and services. It lightens and advan- networks, where each DeAI edge network is com-
ces smart blockchain, Web3, the metaverse, DAOs, posed of many AI devices and edges. Each AI node
and DeSci, and their ecosystems. may have one to multiple devices affiliated, forming a
DeAI local edge network. Nodes are identified (e.g., by
hashing) and connected in decentralized chains, rings,
f
Note: some icons are from the Internet. or other topological architectures.

10 IEEE Intelligent Systems May/June 2022

 EDITOR'S PERSPECTIVE

› DeAI architecture: supporting decentralized agent Figure 1 illustrates a conceptual map, the main func-
networking, communication, collaboration, and tions, and research issues of DeAI systems. DeAI may be
task undertaking; connecting and communicating operated on 1) P2P and D2D networks, 2) an edge net-
P2P agents, nodes and their affiliated devices in a work, or 3) a cloud network. Accordingly, there are five
decentralized manner, for example, by a P2P AI DeAI architectures and patterns as follows.
network;
› Edge intelligence: supporting edge-level AI task A) P2P on-device AI systems: In D2D or P2P2 net-
creation, execution, discovery, and verification; works, end devices connect and chain with each
enabling edge networking, communication, and col- other to form D2D and P2P2 AI systems and net-
laboration; supporting undertaking, coordinating works, and fulfill AI functions on devices.
and communicating AI tasks across devices in an B) Data-sharing edge AI systems: In an edge net-
edge network; managing and discovering existing work, data from devices are shared with edge
and new edge networks; managing the registration nodes (edge server, mediator, gateway, etc.),
or deregistration of third-party AI edge networks; where AI tasks are executed on the device data
and maintaining the integrity of edge networks; and results are returned to respective devices.
› On-device intelligence: enabling device-level AI C) Model-sharing edge AI systems: In an edge net-
task creation, execution, discovery, and verifica- work, pretrained AI models are dispatched to
tion; managing and discovering existing and new end devices, where the AI tasks are executed on
devices; managing the registration or deregistra- devices with their data and the reparameterized
tion of third-party AI devices and tasks; supporting and retrained models are transferred to edge
D2D cooperation, computation, and communica- nodes.
tion; and maintaining the integrity between devi- D) Device-edge-cloud hybrid model-sharing edge AI
ces in an edge network; systems: Device-edge-cloud networks are
› Decentralized contracting: forming AI task con- formed with end devices, edge nodes, and cloud
tracts to allow intelligence (e.g., AI models, algo- servers, where pretrained cloud AI models are
rithms, resources, services, or tasks, and knowledge dispatched to edge nodes, which are customized
and findings) to be shared, executed, outsourced, or for edge-specific functions and further dis-
crowdsourced between edges or between devices; patched to end devices for execution, and
› Decentralized knowledge and intelligence feder- retrained models are returned to edge nodes
ation: storing, sharing, and federating knowledge with registration, and are then transmitted to
and intelligence, such as AI models, findings, the cloud for AI task execution with consoli-
insights, and prior knowledge, to and from AI dated AI models.
networks, edges, and devices; E) Device-edge-cloud hybrid data-sharing edge AI
› Decentralized consensus-building: recording, systems: Device-edge-cloud networks are for-
broadcasting, validating, orchestrating, securing, med with end devices, edge nodes, and the cloud
and managing AI agents, edges, and devices in a servers, where data in the end devices are trans-
DeAI network; building consensus over DeAI net- ported to edge nodes for transformation and
works in an ecosystem; and mitigating conflict; registration, which are then further transmitted
› Decentralized communication: enabling secure, to the cloud to execute AI tasks. Edge-specific
transparent, scalable message passing between AI task executive results are returned to edge
AI agents, edges, devices, and networks, for nodes, which are further transformed and
example, in a P2P or D2D manner; decomposed with end device-specific results
› Decentralized governance: enforcing the agree- passed to each end device.
ment, accountability, reliability, integrity, coordi-
nation, and interoperation between AI devices, Here, it is not possible to discuss all the functions
edges, or between DeAI networks; and research issues due to the space limitation, so we
› Decentralized security and privacy: encrypting AI refer interested readers to Cao’s work9–11 on X-interac-
resources; maintaining the security of data, knowl- tions, X-complexities, and X-intelligences in complex
edge, and intelligence across edges and devices; systems and their data science and AI-driven solu-
protecting the privacy of agents, edges, and devi- tions. X-interactions, X-complexities, and X-intelligen-
ces, such as keeping them at local when AI tasks ces determine the future and challenges of the new
are executed and findings, are shared to edges. age of AI, the X-AI age,12 which are also intrinsic and

May/June 2022 IEEE Intelligent Systems 11

EDITOR'S PERSPECTIVE

intricate to DeAI and DeAI-enabled smart blockchain, problem solving, typically, perception, data acquisi-
Web3, the metaverse, and DeSci. tion, analytics and learning, knowledge discovery,
reaction, reasoning, and intervention on end devices,
in edge networks, or with cloud servers; 3) supporting
Edge Intelligence decentralized sharing, storage, coordination, coopera-
Edge intelligence emerges in synergizing edge com- tion, and communication between end devices, edge
puting with AI.13,14 Edge intelligence refers to the intel- networks, and within the cloud; 4) supporting privacy-
ligent paradigms, techniques, capabilities, activities, preserving, secure, distributed and collaborative AI
behaviors, and problem solving of end devices by activities across decentralized devices and edge net-
those in proximity, or jointly by the end, edge and works without moving and sharing private data and
cloud devices to achieve end, edge (cluster) or collab- specialized AI capabilities around and without central-
orative AI. AI goals, tasks, and capabilities are directly izing them to a mediator, matchmaker, or gateway;
undertaken on decentralized end and edge devices for 5) supporting privacy-preserving, secure transfer and
every device, every local system or organization, and exchange of intelligent capabilities, AI models, and
devices in proximity or distributed to end/edge devi- problem-solving results between end devices, edge
ces. Decentralized devices, end users, and their data, networks, or within clouds; 6) supporting privacy-pre-
behaviors, and capabilities carry forward AI tasks and serving, secure consolidation, federation, orchestra-
fulfill AI problem solvers locally and in proximity. tion, and consensus-building of collaborative AI tasks
Similar to CeAI, edge intelligence also fosters and between end and edge devices; 7) supporting the effi-
fulfills general AI capabilities, such as thinking, per- cient decomposition, splitting, parallelism, concur-
ceiving, communicating, reasoning, planning, analyz- rency, and pipelining of resources (data), AI tasks,
ing, learning, behaving, acting, and decision-making. models, computation, optimization (e.g., learning per-
Differing from CeAI, these edge intelligent capabilities formance), communication, etc.; 8) developing energy,
are fulfilled by individual end devices (e.g., task-spe- bandwidth, latency, security, and efficiency-sensitive
cific drones) or edge devices in proximity (e.g., a clus- communication protocols, message-passing mecha-
ter of nearby drones). Accordingly, edge AI studies the nisms, transmission mechanisms between end devi-
challenges, techniques, and opportunities of edge ces, edge networks, or within clouds; 9) developing AI
vision for sensing, perception, identification and rec- system architectures to support edge AI tasks, data
ognition in the physical, virtual, or cyber-physical management, information flow, task and workflow
world; edge learning, including analytics, reasoning, management, optimization, computation, communica-
inference, and optimization; edge interfacing between tion, and aggregation.
humans/brains and devices and virtual worlds; edge Accordingly, enabling techniques are required to
conversation, including understanding and communi- empower edge intelligence. Typical enabling techni-
cating in characters, voice, text, or multimedia; edge ques comprise privacy-preserving data sharing and
cooperation for ad hoc, on-demand or scheduled task- communication; smart DApp development; encrypted,
ing physically, virtually, or dually; edge action and P2P, and asynchronous communication; data, task
intervention, including response, reaction, interven- and model splitting, pipelining, dispatching, coordina-
tion, and next-best action; and edge consensus-build- tion, orchestration, and aggregation; computation bal-
ing, such as task orchestration, modeling, and result ancing, efficiency and sharing; prior and knowledge
aggregation, etc. sharing and transfer; objective balancing and optimi-
In particular, edge learning emerges as a frontier of zation for end devices, end-to-edge networks, edge
applying AI and machine learning on devices, edge networks, or edge-to-cloud devices.
nodes, and over the air. New areas of cloud edge learn- Typical applications of edge intelligence are real-
ing, federated edge learning, mobile edge learning, and time video analytics by smartphones or drones, autono-
blockchain edge learning will become paramount in mous decision-making by driverless vehicles, crop-spe-
6G, smart cities, industrial IoT, unattended task under- cific pest control and fertilization in digital agriculture,
taking, and the metaverse, etc. unattended mining machines, autonomous industrial
Critical research challenges in acquiring, discover- robotics, automating home facilities and services in
ing, and managing edge intelligence include: 1) charac- smart homes, utility and transport planning in smart cit-
terizing decentralization in end devices, end-edge ies, and unattended space exploration robots. A typical
networks, and end-edge-cloud networks, in particular, use case is to enable on-device deep learning (such as
the interaction, complexity and intelligence of end and real-time videoing and analysis) on smartphones, driver-
edge devices; 2) enabling decentralized intelligent less cars, or drones while also supporting collaborative

12 IEEE Intelligent Systems May/June 2022

 EDITOR'S PERSPECTIVE

TABLE 1. Differences between CeAI and DeAI.

Aspect CeAI DeAI

Methodology Top–down, holism, authority, and autocracy P2P, bottom–up, reductionism, autonomy, and
democracy
Objective Global objectives Local objectives
Intelligence General and strong intelligence Local, edge and weak intelligence
Task Global and central task Local and distributed task
Data/repository Central and single resourcing and storage Local, distributed, multiple, end resourcing and
storage
Model Central and global model End, and local model
Architecture Central, vertical and hierarchical control, mediation, Horizontal, P2P, D2D, distributed, chain, and flat
matchmaking, gateway, and server/client structure structure
Process Consensus-building, aggregation, and orchestration Partition, decomposition and splitting
Mechanism Predefined, vertical, design-time alliance, Ad hoc, horizontal, run-time self-motivation and
coordination, cooperation, normalization, and organization, and negotiation
standardization
Computation Central, global computing infrastructure Distributed, local computing systems
Communication Broadcasting and hierarchical P2P, D2D
Decision Global goal-driven authority, and strategic minority Local goal-driven personalization and majority
voting
Output Global, aggregated, and integrative Local, individual, and personalized
Privacy Weak protection Strong protection
Security Central authorization, monitoring, risk, and Local or distributed authorization, monitoring, and
governance risk
Pros Unification, capacity, order, efficiency, stability, Personalization, flexibility, adaptivity, resilience,
strategic, concentrated resources, computing, and transparency, autonomy, expandability, high fault
data tolerance, scalability, throughout, low cost, and
risk
Cons Low flexibility, adaptivity, autonomy, robustness, Low capacity, energy, resource, computation, and
reliability, accountability, fault tolerance, high stability, high latency, and chaos
vulnerability, risk, cost, and catastrophic mistake

tasks (e.g., transport planning and crop pest-control) in systems seem to be contradictory. It is believed our
edge networks. Here, enabling techniques for edge brain issues the intention and control of the behaviors
learning require in-device energy- and storage-efficient of our body parts. Consequently, for the same prob-
perception, recognition, and diagnosis by deep mod- lem, different people may have opposing opinions and
els,15 energy-efficient16 compressed communication responses, reflecting not only the strong decentralized
between devices and edge nodes, knowledge transfer intelligence of individuals but also the strong central-
between devices and edges, and coordination between ized intelligence of the individual mental, mind, and
end devices and between end and edge devices for emotion. In contrast, artificial systems, including
intervention. human-made systems, social systems, political sys-
tems, economic systems, cultural systems, and reli-
gious systems, are more or less mixed with both
SYNTHESIZING CeAI, DAI, AND centralized, top–down organization, governance, and
DeAI management by layered authoritative operations, gov-
Balancing CeAI, DAI, and DeAI erning bodies, and regulatory bodies and decentralized,
In reality, natural systems, including botanical sys- individualized and lower end autonomy, self-organiza-
tems, environmental systems, earth systems, oceanic tion, cooperation, coordination, and communication by
systems, and astronomical systems, seem to be more autonomous and self-organizing individuals or individual
decentralized and self-organized, whereas human clusters. The balance between centralization and

May/June 2022 IEEE Intelligent Systems 13

EDITOR'S PERSPECTIVE

decentralization forms the spectrum from full openness, aims to consolidate the respective thinking, design
free of control, and democracy to full close, control, and powers, working mechanisms, and functional modules
autocracy, corresponding to DeAI to CeAI in terms of of CeAI and DeAI, and to build consensus between maxi-
their intelligence and intelligent problem solving. mizing global and local objectives simultaneously and
AI systems are part of human-made systems, despite minimizing conflicts in understanding complex systems
different AI paradigms inspired by human intelligence, and pursuing effective problem-solving solutions. Har-
natural intelligence, and social intelligence. This has moti- monic, sustainable, and balanced integrative AI involves
vated different AI design thinking and systems. CeAI prin- and integrates top–down and bottom–up human intelli-
ciples and tools support global, top–down, and holistic gence in mind, emotion, vision, conversation, action, and
design and development. DAI and DeAI lean toward low- natural and social intelligence inspired by nature and
end requirements and preferences with a focus on society. This thinking has been widely applied in hybrid
addressing their local goals, priorities, exercising their intelligent systems by integrating two-to-multiple AI
individual intelligence, and utilizing their local resources paradigms, techniques, objectives, tasks, or models,
and facilities to the greatest extent. Real-world complex, from research areas, such as neural translation and ethi-
intelligent systems and intelligent problem-solving cal AI, to specific aims and tasks, such as deep Bayesian
heavily depend on synthesizing both CeAI and DeAI learning and explainable AI.
thinking, systems, and operations in both unattended Second, synthesizing CeAI and DeAI can inspire
and autonomous systems, such as space stations, moon new AI thinking, methodologies, research questions,
rollers, drones, and autonomous vehicles, and hybrid sys- and developments. New perspectives and opportuni-
tems, such as intelligent transportation systems, smart ties include systematic AI thinking, inclusive AI, inte-
cities, and macroeconomic and epidemic monitoring and grative AI systems, intelligence metasynthesis, and
control. CeAI and DeAI are not a binary and exclusive conflict mitigation. Each is affiliated with new research
choice; rather, they are inclusive and complementary, challenges and issues.
coexist and jointly contribute to complex systems and AI
problem solving. The thinking, mechanisms, and techni- › Systematic AI thinking: New AI thinking, para-
ques of CeAI and DeAI are also complementary. digms and methodologies of integrative AI, inclu-
Table 1 compares the properties, characteristics, sive AI, and sustainable AI; quantifying significant
pros, and cons of CeAI and DeAI in terms of the major interactions and complexities; metasynthesizing
AI aspects of methodology, objectives, intelligence, major AI paradigms, such as human cognitive,
tasks, data and resources, models, architectures, pro- emotional and imaginary intelligence, behavioral
cesses, mechanisms, computation, communication, intelligence, algorithmic intelligence, natural intel-
decision-making, output, privacy, and security. This ligence, social intelligence, etc.
comparison also shows the need to integrate and bal- › Inclusive AI: Developing AI thinking, paradigms,
ance CeAI and DeAI in complex AI systems and intelli- methodologies, architectures, processes, mech-
gent problem solving. anisms, and tools to complement, consolidate,
harmonize complementary, conflicting, or com-
peting CeAI and DeAI thinking, capabilities, and
Hybridizing CeAI, DAI, and DeAI advantages; developing inclusive AI optimiza-
What would an AI system integrating CeAI and DeAI tion, decision-making theories, and methods to
look like? How would synthesizing CeAI and DeAI pro- complement and consolidate global and local
mote AI developments? What are their major research objectives.
challenges and issues? This section briefly discusses › Integrative AI: Enabling both centralized and
these questions. decentralized working mechanisms, integrating
First, the integration of CeAI and DeAI is problem- different AI techniques and tools driven by
and requirement-driven. It does not aim to enable all- diverse objectives, problems, and requirements;
round and omniscient intelligence and functionality and hybridizing specific AI techniques and tools for
incorporating and integrating all individual intelligent integrative problem solving.
paradigms and objectives, such as human-level AI, social › Metasynthetic AI: Developing the representa-
AI, and natural AI into one system. On a high level, syner- tion, modeling, and metasythesis of qualitative
gizing CeAI and DeAI seeks to enhance the inclusivity, to quantitative human intelligence, domain intel-
complementarity, harmony, sustainability, and balance ligence, data intelligence, network intelligence,
between CeAI and DeAI thinking, methodologies, func- and machine intelligence in complex intelligent
tionalities, and strengths. On the low level, integrative AI problem-solving.

14 IEEE Intelligent Systems May/June 2022

 EDITOR'S PERSPECTIVE

› Conflict mitigation: Developing methodologies devices, models, and tasks; and 3) intelligent comput-
and tools to mitigate the respective advantages, ing platforms and tools for programmable edge com-
disadvantages, and conflicts of coexisting intelli- putation, communication, and AI modeling on devices
gence, interactions, and complexities; developing and by average users.
resolution mechanisms for tolerant, inclusive, and First, on-device intelligence may be incorporated
resilient open AI systems. into blockchain transactions, currency, blocks, tokens,
and ledgers on demand and to make them more
active, personalized, secure, adaptive, autonomous,
SMART BLOCKCHAIN, METAVERSE,
fault tolerant, risk tolerant, and resilient. For example,
WEB3, AND DESCI
blocks could be empowered with task-specific intelli-
How would DeAI and synthesizing CeAI and DeAI
gence, such as intrusion and alternation detection.
make blockchain, Web3, the metaverse, and DeSci
Hashing codes may be embedded with functions to
smart or smarter? In decentralized settings, DeAI is a
monitor and detect the order, intrusion, and integrity
critical enabling technique for smart blockchain,
of chains. In specific applications, blocks and chains
Web3, the metaverse, and DeSci.
may be customized with intelligent functions, e.g.,
video recording and analysis for encrypted drone
Smart Blockchain chains to detect environmental, climate or agricultural
Existing blockchain technology and systems concen- change. Cryptocurrencies may be encoded with price
trate on the smartness of digital contracts; hence, smart monitoring and manipulation detection. Distributed
contracts that can securely, fault and risk-tolerantly and ledgers could be incorporated with various functions
automatically codify, transfer, execute, audit, and man- to enforce the integrity and consensus of blocks, pre-
age cryptographically signed transactions or an agreed vent illegal or unexpected permission, access, replica-
value on a blockchain network or distributed ledger. tion, sharing, and the approval of ledgers over chains.
However, the smartness of contracts is typically prede- Second, edge intelligence may be incorporated
signed and precoded without autonomous adaptation into P2P and D2D blockchain nodes, networking, com-
and intelligence. Other important components, such as munication, computation, and infrastructure to
a block, ledger, token, cryptocurrency, and chain and the enhance their intelligence.18,19 Opportunities include
P2P network infrastructure, networking, computing, con- 1) incorporating intelligent behaviors and functions,
sensus-building, and management, are not necessarily such as perception, acquisition, analysis, learning, rea-
smart. In addition, although it is claimed that blockchain soning, reaction, and intervention into block nodes
systems can make other systems, such as a city or and chains; 2) enhancing secure, risk/fault-tolerant
Industry 4.0 smart, enabling smart blockchain, is an registration, access, replication, sharing, discovery,
open issue, and smart blockchain represents a new gen- storage, validation, and transfer of currencies and
eration of blockchain 2.0. ledgers on the chain; 3) automatically monitoring and
Cao et al.,17 introduced the concept of smart cur- regulating currency pricing, price stability, trading
rencies, smart cryptos, and smart blockchain in the strategies, and portfolios; 4) improving the efficiency,
context of smart FinTech enabled by data science and energy consumption, and latency of bitcoin mining,
AI. Various opportunities are illustrated to enable blockchain infrastructure, computation, and commu-
smart blockchain, such as in pricing, trading strategy, nication; 5) improving the automatic detection and
portfolio management, bitcoin mining, risk analytics, prevention of security and privacy violation in smart
fragility detection, governance, and regulation. contracting processing and contracts; 6) analyzing
Here, I discuss the fundamental roles of DeAI in exceptional behaviors and activities of and between
enabling smart blockchains through making smart blocks, nodes, ledger, and contracts over time; and
infrastructure, smart networking, smart computing, 7) promoting, innovating and regulating blockchain
smart consensus-building, smart governance, and businesses and services.
smart transaction, currency, block, and ledger. This
relies on 1) incorporating and strengthening on-device
intelligence into blockchain components and edge Smart Metaverse
intelligence into chain nodes, ledgers, and the net- The metaverse forms a virtual society and a virtuality-
work; 2) intelligent infrastructure to partition, pair, and reality continuum. The smart metaverse is thus essen-
virtualize computing and communication for end devi- tial to make the digital or dual world smarter and
ces and edge nodes and facilities to decentralize incorporate AI success in the physical world with the
resource management and the orchestration of AI metauniverse. This includes enabling digital agents

May/June 2022 IEEE Intelligent Systems 15

EDITOR'S PERSPECTIVE

and society with intelligent capabilities of understand- › Reasoning and planning: for agents to reason
ing, perceiving, reasoning, communicating, analyzing, about and plan their behaviors, actions, activi-
learning, optimizing, planning, and reflecting on their ties, and responses.
self and societal behaviors, activities and consequen- › Teaming: enabling agents to select and team up
ces in the digital world and intelligent interactions, with other agents for team tasks.
cooperation, communication and problem solving › Vision: for agents to perceive, track, detect, iden-
between agents, between agents and humans, and tify, recognize, analyze and predict the behaviors
through teaming and alliance in the cyber-physical- of teaming agents, competitors or defenders.
social world over time, space, and episode. › Communication: enabling agents to communi-
Perspectives and opportunities of producing the cate with team players about their goals, inten-
smart metaverse include enabling the following. tions, beliefs, and plans in private chatbots,
queries, or instant messaging.
› General intelligence in virtual societies: such as › Search: enabling agents to search historical
empowering agents with abilities of perception, transactions, results, and resources.
recognition, detection, identification, conversa- › Learning: enabling agents to analyze, predict, and
tion, behavior, and action-taking, reasoning, plan- optimize their behaviors, states, and objectives.
ning, search, retrieval, and optimization in their › Reinforcement: enabling agents to reinforce their
virtual communities. objectives and optimize their goals and actions by
› Human-like intelligence in digital worlds: mimick- optimizing policies, strategies, and actions.
ing and replicating human intelligence in digital › Environment: enabling agents to sense, perceive
avatars and agents for them to better interact, and predict their environment.
interface, communicate, and team with humans in › Personalization: enabling agents to personalize
the dual worlds and agent-human cooperation. their preferences, objectives, states, strategies,
› Embodied and immersive agent-human interac- behaviors, and actions.
tion, communication, teaming and cooperation in › Reflection: enabling agents to receive feedback and
the dual world: supporting mind, haptic, vision, to evaluate and reflect on their prior states, behav-
action and device-based cyber-physical interfac- iors, actions, and strategies and to improve them
ing, communication, and feedback for telepre- during planning, action-taking, and interactions.
sence, and agent-human teaming and cooperation. › Automation: enabling agents to automatize their
› Rational and sustainable virtual economy: behaviors, actions, and decisions.
ensuring the rationality and sustainability of › Risk: enabling agents to proactively detect, ana-
business models and economic activities in cre- lyze and predict harm, risk, intrusion, attack and
ating a virtual economy, currency, payment, automate reactive actions and strategies.
cost, and utility models in the digital worlds. › Assistance: enabling agents to make enquiries
› Virtual safety and security: ensuring information and ask for instant help from peers.
order, user safety and mental health, and infor-
mation security in interfacing and communicat-
ing with agents in virtual communities. Smart Web3
› Ethical virtual society: ensuring social intelli- Web3 is deemed to be a new edition of the World Wide
gence in virtual interactions and agent-human Web enabled by decentralized blockchain technology,
teaming; enforcing rationality, norms, privacy, tokens, and DeFi for trustworthy, secure, and immersive
trust, and accountability in social interactions, decentralized web platforms, content, and services. It
virtual marketing, trading, and competition. contrasts with the centralized content, services, and
technology in Web 2.0 for user interaction and social
We further illustrate some specific general intelli- media and Web 1.0 for static consumer-oriented content
gence, human-like intelligence, and agent–human- listing. However, the Web3 movement has not made
teaming intelligence in creating intelligent avatars and much progress and mainly stays at the conceptual level.
agents in the metauniverse. Grounded on decentralized thinking and technol-
ogy, Web3 has the potential to 1) support and favor
› Duality: supporting predictive, active and proactive the preference, friendliness and concentration of scal-
interaction, cooperation, and response between able, heterogeneous, individualized and personalized
digital avatars in the cyberspace and humans (or end and edge devices, users, and services; 2) offer and
devices and vehicles) in the physical world. enhance individual engagement, security, privacy, and

16 IEEE Intelligent Systems May/June 2022

 EDITOR'S PERSPECTIVE

scalability; 3) connect to and interact with large-scale enabling active and proactive detection and pre-
and heterogeneous end devices, and diversified low- vention of intrusion, manipulation, and misuse, etc.
level edge nodes and networks; 4) operate in secure, › Web3 privacy and ethics: preserving and protect-
privacy-preserved, and P2P networking, communica- ing individual privacy of their accountable, valid,
tion, token and cryptocurrency-based financial serv- and trustworthy content, communication, and
ices, and ecosystems; and 5) provide immersive, activities in the Web3 network and ecosystem, etc.
personalized, cyber-physical-social-human interactive
experience, services, and ecosystems. In addition to blockchain technology, the abovemen-
DeAI can play a foundational role in enabling and tioned areas, functions, and opportunities require advan-
shaping open, secure, trustworthy, privacy-preserving, ces in DeAI, smart AI chips, edge learning, on-device AI
and end user/device-oriented Web3 and make it smart. programming, interactive and visual techniques, and
metaverse techniques for individuals and individual clus-
› Web3 intelligence: enabling Web3 end users, ters to code, program, interact, and communicate with
devices, and edge networks to develop and their smart devices, apps, digital avatars, and create their
deploy on-device and edge intelligence. personalized apps, web content, pages, and services.
› Web3 metaverse: developing immersive capabili-
ties, experience, and services in Web3 and incor- Smart DeSci
porating virtual-physical interaction, teaming and DeSci extends the concept of decentralization, DAO and
cooperation. decentralized thinking to scientific funding, publishing,
› Web3 protocols: supporting encrypted, trace- operations, organizations, and inter-, multi-, and cross-
able, P2P, D2D, edge networking, and edge-cloud disciplinary studies.20 Complementing centralized sci-
networking protocols for connection, content ence and developments, DeAI could play an essential
exchange, message passing, data sharing, coor- role in DeSci and open science to enable and enhance
dination, etc. bottom–up, self-organizing, self-motivated, free, open,
› Web3 infrastructure: supporting efficient, energy- secure, privacy-preserving, personalized, adaptive, and
sensitive, trustful, latency-sensitive P2P, D2D, trustworthy scientific thinking, fields, organizations,
edge, edge-cloud networking, computing, resourc- operations, activities, and developments.
ing, programming, and servicing, etc. In the following, we discuss a few aspects to illus-
› Web3 operation: supporting decentralized operat- trate where and how DeAI could substantially pro-
ing systems, device registration and identification, mote the DeSci movement and make it smart.
access control, resource management, task deco-
mposition, computation offloading, function virtual- › Enabling secure, trustworthy, privacy-preserving,
ization, search and discovery, resource and activity and autonomous open science.
orchestration, regulation, consensus-building, etc. › Enabling decentralized, secure, open, and collab-
› Web3 communication: enabling secure, privacy-pre- orative scientific infrastructures, platforms, and
serving, P2P and D2D communications between services.
heterogeneous end devices, between personalized › Enabling decentralized, P2P, autonomous, and
end users, with heterogeneous devices in proximity, incentive-driven scientific teaming, organization,
and in the ecosystem with heterogeneous commu- association, and collaboration.
nication protocols, networking connectivity, etc. › Enabling transparent, and open trans-, multi-,
› Web3 services: enabling end/edge-oriented real and cross-disciplinary science and scientific pro-
time, active, adaptive, personalized and trustworthy grams and collaborations.
services, personalized content, P2P and D2D inter- › Enabling original scientific breakthrough and
action, communication, instant messaging, and innovation driven by free and open self-motiva-
encrypted economic activities, which may be pro- tion, incentive, interest, and collaboration.
vided by end users, from end devices, and through › Enabling interoperable, open, federated and col-
edge networks, etc. laborative scientific crowdsourcing.
› Web3 development toolkit: supporting open › Enabling secure, privacy-preserving, and person-
source, end user, end device, edge network-ori- alized scientific sharing, storage, publishing, and
ented DApp programming, and interactive and transfer.
workflow-oriented interfacing. › Enabling decentralized, privacy-preserving, free
› Web3 security: protecting, tracing and monitoring or priced, responsible, and accountable peer
information security, order and authenticity, and review and evaluation.

May/June 2022 IEEE Intelligent Systems 17

EDITOR'S PERSPECTIVE

FIGURE 2. Smart virtuality-reality metaverse ecosystem: Metasynthesizing DeAI, metaverse, blockchain, Web3 and other tech-
nologies and applications.

› Enabling secure, market-driven, crowd-based e-health, digital innovation, digital economy, e-com-
valuation, protection, and the transfer of intel- merce, games, digital life and society, cyberspace, and
lectual property. social media. The virtuality-reality continuum includes
large screens, smartphones, smart cities, smart homes,
and smart IoT (or industrial IoT) devices.
CONCLUDING REMARKS DeAI develops decentralized intelligence and
Decentralization has been widely promoted in man- intelligent systems by taking advantage of individu-
agement science, economy, and governance. Dis- alized and personalized on-device intelligence
tributed and decentralized computing and systems and edge intelligence. DeAI can also substantially
have also demonstrated their significance. Decen- promote smart blockchain and the metaverse and
tralizing AI and intelligent science and technology promote Web3 and DeSci movements and develop-
are still in its infancy. DeAI forms a critical enabler ments. This will inspire DeAI thinking, techniques,
and enhances various recent research highlights platforms, systems, and services. DeAI-enabled
and areas, including DAI, the cloud and edge com- Web3 and DeSci will be substantially expanded to
puting, blockchain, and the metaverse. This article broad technical developments, business and eco-
reviews the concepts and perspectives of DeAI and nomic opportunities, and intelligent decentralized
their opportunities in enabling smart blockchain, ecosystems and society.
Web3, the metaverse, and DeSci. Figure 2 illustrates
a smart virtuality-reality metaverse ecosystem,
which synthesizes DeAI, the metaverse, blockchain, ACKNOWLEDGMENTS
Web3 and other relevant technologies for smart This work was supported in part by ARC Discovery under
immersive digital and virtual-physical applications Grant DP190101079 and Future Fellow FT190100734
and services. The applications include FinTech, Grants.

18 IEEE Intelligent Systems May/June 2022

 EDITOR'S PERSPECTIVE

REFERENCES 13. D. Xu et al., “Edge intelligence: Empowering

intelligence to the edge of network,” Proc. IEEE,
1. United Nations Development Programme,
vol. 109, no. 11, pp. 1778–1837, Nov. 2021.
“Decentralization: A sampling of definitions,” UNDP,
14. Z. Zhou, X. Chen, E. Li, L. Zeng, K. Luo, and J. Zhang,
New York, NY, USA, Working Paper, Oct. 1999.
“Edge intelligence: Paving the last mile of artificial
2. Y. Mao, C. You, J. Zhang, K. Huang, and K. B. Letaief, “A
intelligence with edge computing,” Proc. IEEE, vol. 107,
survey on mobile edge computing: The
no. 8, pp. 1738–1762, Aug. 2019.
communication perspective,” IEEE Commun. Surv.
15. I. Varlamis et al., “Using Big Data and federated learning
Tut., vol. 19, no. 4, pp. 2322–2358, Oct.–Dec. 2017.
for generating energy efficiency recommendations,”
3. Y. Zhang, Mobile Edge Computing. Berlin, Germany:
Int. J. Data Sci. Anal., pp. 1–17, 2022.
Springer, 2022, pp. 1–105.
16. Z. Yang, M. Chen, W. Saad, C. S. Hong, and M. Shikh-
4. M. N. M. Bhutta et al., “A survey on blockchain
Bahaei, “Energy efficient federated learning over
technology: Evolution, architecture and security,”
wireless communication networks,” IEEE Trans. Wireless
IEEE Access, vol. 9, pp. 61048–61073, 2021.
Commun., vol. 20, no. 3, pp. 1935–1949, Mar. 2021.
5. M. Nasir et al., “Scalable blockchains—A systematic
17. L. Cao, Q. Yang, and P. Yu, “Data science and AI in
review,” Future Gener. Comput. Syst., vol. 126,
FinTech: An overview,” Int. J. Data Sci. Anal., vol. 12,
pp. 136–162, 2022.
no. 2, pp. 81–99, 2021.
6. L. Lee et al., “All one needs to know about metaverse:
18. Y. Du et al., “Blockchain-enabled edge intelligence for
A complete survey on technological singularity, virtual
IoT: Background, emerging trends and open issues,”
ecosystem, and research agenda,” 2021,
Future Internet, vol. 13, no. 2, 2021, Art. no. 48.
arXiv:2110.05352.
19. T. Gadekallu et al., “Blockchain for edge of things:
7. S. Hamburg, “Call to join the decentralized science
Applications, opportunities, and challenges,” IEEE
movement,” Nature, vol. 600, p. 221, 2021.
Internet Things J., vol. 9, no. 2, pp. 964–988, Jan. 2022.
8. D. Angelo and M. Salzer, “Identification of token
20. F. Wang et al., “The DAO to DeSci: AI for free, fair, and
contracts on Ethereum: Standard compliance and
responsibility sensitive sciences,” IEEE Intell. Syst.,
beyond,” Int. J. Data Sci. Anal., pp. 1–20, 2021.
vol. 37, no. 2, pp. 16–22, Mar./Apr. 2022.
9. L. Cao, Data Science Thinking: The Next Scientific,
Technological and Economic Revolution. Berlin,
Germany: Springer, 2018.
LONGBING CAO is a professor and an ARC future fellow
10. L. Cao, Metasynthetic Computing and Engineering of
(professorial level) with the University of Technology Sydney,
Complex Systems. London, U.K.: Springer, 2015.
Sydney, NSW, 2007, Australia. His research interests include
11. L. Cao, “Coupling learning of complex interactions,” J.
Inf. Process. Manage., vol. 51, no. 2, pp. 167–186, 2015. AI, data science, machine learning, behavior informatics, and
12. L. Cao, “A new age of AI: Features and futures,” IEEE their enterprise applications. He is a Senior Member of the
Intell. Syst., vol. 37, no. 1, pp. 25–37, Jan./Feb. 2022. IEEE. Contact him at [email protected].

May/June 2022 IEEE Intelligent Systems 19