15
[391], [392]. For instance, [389] investigated the feasi- nose the abnormal states of unknown artifacts without
bility of transferring the process knowledge at various having prior knowledge on them [405]. The collection of
manufacturing scales. [392] proposed a framework that manufacturing know-hows and building a reliable AGI
consists of a data layer and a physics layer, to capture for diagnosis and troubleshooting is promising but needs
the statistically-correlated temporal dynamics and im- tremendous effort.
poses regularizations through system working principles 5) Manufacturing assistance and workforce training:
and governing physics, respectively. More endeavors are ManufacturingworkforcetrainingintheIndustry4.0and
needed to deal with the manufacturing heterogeneity and IoT era is of super importance to national security, and
achieve generalizable models. there is a huge lack in the manufacturing talent pool
2) Manufacturing data augmentation and labeling, and [407], [408]. AGI has the potential to revolutionize the
learning with limited data: The manufacturing data way workers learn and interact with machines. For in-
collection and labeling can be expensive. AGI has the stance, AGI can facilitate interactive question answering
potential to deal with the challenge, by performing data andlearningandsmooththelearningcurve.AGIcanalso
augmentation [393], [394], adaptive/interactive data la- help simplify the operation of machines by developing
beling and annotation [395], [396], and learning with smart manufacturing assistants. These assistants could
limiteddata[397],[398].GANhasbeenusedtogenerate optimize the recommendations based on real-time data
useful samples for teeth aligner printing and point cloud from the machines and adapt to individual workers’
characterization [393], and data augmentation for super- preferences.
visedanomalydetectioninadditivemanufacturing[394].
Generative models have been used in CAD software
H. Smart Education
for engineering designs [399]. Reinforcement learning
and active learning can adaptively identify the important The potential of integrating AGI and IoT into education
samples to annotate and achieve process characterization has prompted educators to reconsider traditional teaching
with limited annotation efforts [395], [396]. [397] used a methods and embrace innovative approaches [409], [410].
Siamese CNN based few-shot learning network to mea- When applied to education, AI and IoT can address various
sure distances of input samples based on their optimized challenges and enhance the learning experience. Recent de-
feature representations, which is helpful for achieving velopments in AGI technologies, such as OpenAI’s GPT-4,
good anomaly detection performance with limited sam- have increased awareness about using digital resources in K-
ples. 12 and higher education to equip students for 21st-century
3) Multimodal manufacturing data analysis: Manufac- problem-solving[411]. AGI tools like ChatGPT and GPT-4
turing system conditions can be measured and reflected enhancepersonalizedandinteractivelearningbyprovidingfor-
with various data types and sources, such as IoT sensors, mativeassessmentprompts,feedback,andrelevantreferences.
images, domain knowledge, and texts. How to make the They assist teachers with lesson planning, including content
bestuseofmultimodalinformationandsupportdecision- knowledge and assessment strategies, offering a wider range
making is attracting much attention. While data fusion of content to enrich teachers and help students gain diverse
and multimodal analysis have been studied for decades perspectives.AGIfreesteachersfromtextbooklimitationsand
[400], [401], [402], the solutions are generally tailored expands their content knowledge. For instance, using the 5Es
for an application scenario and can be hard to general- model, ChatGPT designed a learner-centered teaching unit
ize. AGI has the potential to overcome this challenge. on renewable and nonrenewable energy sources, generating a
Recently, contrastive learning was studied to acquire rubric for student self-evaluation, exam questions, and a scor-
effective data representations from multiple modalities ing key for teacher-led evaluation [412]. Key concepts arising
for downstream tasks [403], [404]. For instance, Ai from discussions on AI and education include improving
et al. integrated knowledge distillation to transfer the teachers’ content knowledge, facilitating individualized and
information from handcrafted features to deep learning adaptive learning, differentiated instruction and assessment,
and supervised contrastive learning to enhance feature and enhancing educational outcomes for students. As AGI
discrimination [404]. technologiesprogress,furtherresearchwillexploretheirprac-
4) Manufacturing diagnosis and troubleshooting: Diag- tical applications in the classroom. While the integration of
nosis and troubleshooting is an important step in the AGI to bridge the gap between IoT and AGI within education
manufacturing value chain, but was mainly done man- remains an ongoing endeavor, the following section explores
ually by experts and workers in the past. With the the insights from diverse scholarly works to elucidate the
advent of tools such as LLMs, AGI has been explored transformative potential of AGI and IoT within education.
for the manufacturing process and system diagnosis and Educational Goals. The attainment of educational goals
troubleshooting[405],[406],[354].Thishasthepotential within any learning environment is a critical endeavor, aimed
tobridgehumanknowledgeandAImethodstotrainreli- at fulfilling designated learning objectives and desired student
able decision-making tools. Power et al. studied artifacts outcomes. Despite resource constraints, shortages of skilled
such as cars and circuit designs using Non-Axiomatic educators, and inadequate attention to diverse learning needs,
Reasoning System (NARS), which demonstrated certain strideshavebeenmadewiththeintegrationofIoTtechnology.
features of the generalized diagnostics. NARS can diag- This integration has the potential to reshape the educational