3
manyfollow-upstudiesonmultimodalfoundationmodelssuch AItoanalyzeandextractvaluefromthisdataiscritical.Some
as BriVL [56], BLIP [57], GPT-4 [58], KOSMOS-1 [59], and common AI use cases in IoT include:
KOSMOS-2 [60].
1) Predictive Maintenance: Using machine learning to
analyzesensordatafromindustrialequipmentandpredict
D. Towards AGI Paradigm for IoT failures before they occur [64]. This avoids costly IoT
downtime.
A point broadly shared in the AGI community is called the
2) Anomaly Detection: Identifying anomalies in real-time
core AGI hypothesis [61], which refers to “the creation and
IoTdatastreamstodetectcyberattacks,equipmentfaults,
study of synthetic intelligence with sufficiently broad scope
or other issues requiring intervention [65].
and strong generalization capability, is qualitatively different
3) Personalization: Applying deep learning to user data
from the synthetic intelligence with significantly narrower
from smart devices to provide personalized recommen-
scope and weaker generalization capability”. We characterize
dations and customized experiences [66].
the AGI systems with a list of capabilities motivated by the
4) SmartGrids:AIalgorithmsanalyzedatafromsmartme-
competencies summarized by AI researchers and psycholo-
ters and sensors to optimize energy distribution, manage
gists [62] in Fig. 2.
demand response, and forecast energy needs, enhancing
1) Perception:Thesenseandunderstandingofvisual,hear-
the reliability and efficiency of power grids [67].
ing, touch information from environment, and the abil-
5) Smart Home: AI-driven systems in smart homes [68]
ity of integrating multi-modal information from various
learn user preferences and behavior to automate light-
senses.
ing, heating, and cooling, resulting in an intelligent
2) Actuation: The ability of manipulating physical objects,
environment that enhances comfort and reduces energy
using tools, and navigation in complex environments.
consumption.
3) Memory: The memory regarding facts or beliefs, the
6) Smart Agriculture: AI enhances smart agriculture by
outcome of sequential/parallel combinations of actions,
processing data from sensors and satellite images for
and experience attributed to a particular instance.
precision farming, optimizing irrigation, and predicting
4) Learning: The ability to learn from teachers, other ob-
environmental impacts on crop yields [69], [70], [71].
served agents, media, experimentation, and positive or
7) Smart Robotics:AI-drivenroboticsystemsachieveself-
negative reinforcement signals in the environment.
perception and environmental perception in robots and
5) Reasoning: The ability of deduction, induction, and ab-
dronesbycollectingandanalyzingreal-time,multimodal
duction.Theabilityofreasoningfromobservedpremises,
data from sensors and cameras. This enables precise
physical rules and spatio-temporal associations.
navigation and manipulation [72] across a wide range of
6) Planning:Theabilityofconductingstrategical,physical,
applications,fromlast-miledeliverytoindustrialautoma-
and social planning.
tion and automated farming.
7) Motivation: The ability of creating sub-goals based on
8) Smart Manufacturing: In industrial IoT, AI is used
the pre-programmed goals, or driven by curiosity, emo-
for modeling, monitoring, diagnosis, and control by ana-
tions, empathy, and altruism.
lyzing machine data to forecast breakdowns, optimizing
8) Emotion: The ability of expressing emotion, as well as
production schedules, and improving supply chain effi-
perceiving or interpreting emotion.
ciencies [73].
9) Interaction: The ability to initiate communication and
9) Smart Healthcare: AI applications in smart healthcare
organize group activities, with appropriate behavior. The
[74] include predictive analytics for patient monitoring,
communication could be achieved through verbal, gestu-
natural language processing for electronic health records
ral, pictorial or even cross-modal signals.
[75], [76], [77], and image analysis for diagnostic imag-
10) Quantitative: The ability to comprehend and articulate
ing [75], [78].
mathematicalconcepts,solvemathematicalproblems,and
10) Smart Transportation: AI analyzes data from sensors
apply quantitative reasoning to solve problems that de-
and cameras to optimize traffic flows, improve public
mand mathematical thinking and model-building skills.
transportationsystems,enhancenavigationwithreal-time
11) Creation: The ability to build and modify physical ob-
data, and facilitate autonomous vehicles [79].
jects, assemble and organize social groups, and form
11) Smart Public Safety: AI helps in analyzing data from
novel concepts.
IoT devices for faster emergency response, uses pattern
AGI within the IoT context would imply the development of
recognitiontoanticipateincidents,andimprovescommu-
highlyadvancedandversatileAIsystemsthatpossesshuman-
nication systems in crisis situations [80].
likecognitivecapabilities(i.e.,morethanonecapabilitylisted
12) Smart Environmental: AI is utilized in smart environ-
above) and can seamlessly interact and adapt across a broad
mental systems to analyze data from sensors for pollu-
spectrum of IoT devices and scenarios.
tion control, predict environmental trends, and facilitate
wildlife monitoring for conservation efforts [81].
E. AI in IoT Applications and the Necessity for AGI
While traditional AI applications perform well for narrow
IoT generates massive amounts of heterogeneous data from tasks, developing more versatile systems requires AGI. Con-
interconnected sensors, devices, and systems [63]. Applying ventional AI lacks the flexible reasoning, contextual adapta-