AI in STEM: Physics Forums Advisors Share Their Views

Physics Forums advisors expect AI to arrive unevenly across STEM: faster in narrow, well-defined tasks like long-haul trucking routes or lab data analysis, and slower in classrooms and jobs requiring human judgment. Most contributors caution against buzzword-driven adoption and stress that AI’s usefulness depends on whether the underlying problem is genuinely suited to automation.

Key Takeaways

• Physics Forums advisor Ranger Mike compares today’s AI enthusiasm to 1980s predictions of “lights-out factories” that never materialized.
• Advisor jack action describes AI as an extension of statistics that always produces an answer, even when the underlying assumptions are flawed.
• Advisor bhobba notes that early 8K television upscaling relied on machine learning, while newer models use deep learning for improved image quality.
• Advisor neilparker62 cites AlphaZero’s defeat of Stockfish in chess as his most striking real-world encounter with advanced AI.
• Advisor STEMucator lists five concrete near-term applications: automatic cashiers, automated lab experimentation, AI instructors, automated assembly line workers, and self-driving vehicles.

What do Physics Forums advisors think about AI’s impact on STEM?

Physics Forums asked its community of science and engineering advisors how they expect artificial intelligence (AI) to affect laboratories, classrooms, industry, and everyday society. The responses below are drawn directly from that discussion, presented by contributor.

Ranger Mike: AI adoption may repeat past technology hype cycles

Ranger Mike expects heavy investment in AI but relatively little practical product for some time, drawing a parallel to 1980s predictions of “lights-out” factories and a service-only U.S. economy that never came true. He also points to Deming-style quality management and statistical process control (SPC) as business-school ideas that were popular in theory but never became universal practice.

In his view, AI risks becoming the next in a line of buzzwords, following “robotics,” “going green,” and “sustainability.” He references the film The Graduate, where “plastics” was the era’s buzzword, to illustrate how quickly such terms shift meaning.

Ranger Mike sees AI working well in industry when the market need is correctly matched to the technology. He gives long-distance trucking as an example where AI is likely to be effective, versus short-haul trucking, where variability makes human decision-making more important. He also cites manufacturing examples he has encountered directly: CAD models that automatically generate tool code, and machine-driven inspection against CAD nominal specifications, noting this capability took decades to develop.

jack action: AI is a powerful but fallible extension of statistics

Jack action views AI largely as a powerful extension of statistics, particularly promising for research tasks like discovering new molecules or medical treatments by identifying patterns and prioritizing experiments.

His central concern is that AI, like statistics, always produces an answer even when its underlying premises are flawed. He warns that people may analyze complex systems built on subjective assumptions and then accept an AI-generated result uncritically, and that treating findings as certainties rather than possibilities becomes dangerous once they are codified into rules or laws.

He considers it reasonable to trust AI for well-defined, simple systems, such as a production line, but argues that removing human oversight entirely requires reevaluating who holds responsibility and liability.

Dr Transport: AI helps personalize teaching but is often used as a funding buzzword

Dr Transport sees potential for AI to tailor STEM instruction to individual students, similar to how companies like Amazon and Google already use AI-driven targeted advertising.

He is more skeptical of AI in research funding contexts, describing a former supervisor who required GPU-equipped computers and a machine-learning paragraph in every proposal regardless of relevance. Dr Transport says he has read AI research papers but often found them unclear or difficult to reproduce, with statistics applied only “half-heartedly” to support questionable claims.

anorlunda: An AI personal tutor could track individual student mastery

Anorlunda expects AI to enter conventional education slowly, with one promising application being an AI personal tutor that remembers every student answer and analyzes classroom video to gauge attention and confusion.

Such a tutor could estimate which concepts a student has mastered, missed, or misunderstood, then select targeted remedial lessons in written, video, or practice-problem form. Anorlunda argues the goal should be near-complete comprehension of one lesson before a student advances to the next, with progress reported to teachers and parents, and that the tool should be affordable or open-source to avoid exploitative commercialization.

STEMucator: Five concrete ways AI will reshape daily STEM tasks

STEMucator outlines specific near-term applications rather than general predictions.

• Automatic cashiers: self-service registers will keep improving toward minimal human intervention.
• Automated lab experimentation: machines will execute procedures like chemical measuring and mixing with high repeatability.
• AI instructors: intelligent virtual assistants could provide individualized pacing, increasing remote learning.
• Automated assembly line workers: robots and AI systems will take over more factory tasks.
• Self-driving vehicles: autonomous systems already exist in limited forms, such as autopilot and experimental self-driving cars, and could eventually remove the need for drivers in many contexts.

BillTre: AI’s growth will follow stages across labs, classrooms, and industry

BillTre frames AI as a combination of programming and hardware that will expand in stages, first filling useful functional gaps, then moving into areas that compete with human jobs and create social friction.

Lab

BillTre expects AI to assist with experiment execution, data collection, and publishing, discover relationships in large databases, and eventually take over some tasks currently performed by graduate students, potentially affecting how many PhDs are trained or employed.

Classroom

He expects AI teaching assistants to develop deeper domain knowledge over time, but notes that replacing human instructors will require fast, adaptive responses along with credible personality and interaction quality.

Industry

BillTre anticipates increased efficiency and more flexible, just-in-time production, alongside job displacement that could strain the political relationship between industry and government, since business incentives are often justified by job creation.

Everyday society

He expects job displacement to cause social tension unless managed through attrition, retraining, or policy, and predicts that new jobs created by AI may not match the scale or distribution of the jobs lost.

bhobba: AI-driven television upscaling shows rapid consumer-facing progress

Bhobba points to driverless cars as a clear example of AI disruption, potentially eliminating driving-related jobs such as truck, taxi, and ride-share driving, while also reducing parking revenue and the stress of commuting.

He also describes AI-based upscaling in consumer televisions as a notable application he has observed directly. The first generation of 8K upscaling used machine learning, while more recent models use deep learning for improved results. He notes that AI-enabled downscaling of 8K content to 4K can produce superior-looking 4K images, even though it remains unclear whether most viewers can perceive a difference between 4K and 8K at normal viewing distances.

Andy Resnick: Expert systems have not yet replaced instructors or experiment design

Andy Resnick notes that knowledge-based systems and big-data analysis have already changed industry and daily life, and that image analysis, including clinical imaging and pattern discovery in genomic and proteomic datasets, is a prominent AI application in the lab. Aside from a few headline papers, he has not seen expert systems regularly conceiving entire experiments on their own.

In classrooms, he has observed “AI-lite” mastery learning systems for pre-calculus and calculus, and expects similar approaches for vocational training, but has not seen evidence that AI can wholly replace an instructor. His broader concern is that humans may adapt to AI interfaces as presented, rather than AI being designed to fit human needs, a shift he compares to how smartphones have reshaped human interaction patterns.

neilparker62: AlphaZero’s chess victory over Stockfish stands out as a real-world AI milestone

Neilparker62 identifies AlphaZero’s defeat of Stockfish in chess as his main real-world encounter with advanced AI, calling it revealing about how AI can discover strong strategies. In the classroom, he finds low-level AI tools like online graphing utilities and Wolfram Alpha helpful as supplements, not as replacements for teachers.

jfizzix: AI-generated deepfakes pose a growing societal risk

Jfizzix’s primary concern is AI-assisted deception, particularly deepfakes. He argues that as these tools become more widely available, misinformation campaigns will grow more sophisticated, and stresses the importance of developing defenses, potentially using AI itself, to prevent widespread cynicism, paranoia, and societal breakdown.

Frequently Asked Questions

Will AI replace teachers in STEM classrooms?

Most contributors, including Andy Resnick and neilparker62, say current AI tools function as helpful supplements, such as mastery-learning systems or graphing utilities, rather than full replacements for instructors. Anorlunda and BillTre describe more advanced AI tutor concepts as future possibilities, not present reality.

What jobs are most at risk from AI automation according to these advisors?

Bhobba specifically names truck, taxi, and ride-share driving as jobs threatened by driverless car technology. STEMucator adds automatic cashiers and automated assembly line workers as near-term examples of AI-driven job displacement.

Is AI overhyped as a buzzword in research and industry?

Ranger Mike and Dr Transport both express concern that AI is sometimes invoked to secure funding or attention rather than because it solves a genuine problem. Dr Transport recounts a supervisor requiring machine-learning language in every proposal regardless of relevance.

What real-world AI application impressed these advisors most?

Neilparker62 cites AlphaZero’s chess victory over Stockfish. Bhobba highlights AI-based 8K television upscaling, noting the shift from machine learning in early models to deep learning in more recent ones.

What is the biggest risk AI poses to society, according to this discussion?

Jfizzix identifies AI-assisted deepfakes and misinformation as a major risk. Jack action separately warns that AI, like statistics, always produces an answer even from flawed premises, which becomes dangerous if such outputs are treated as certain and codified into rules or laws.

Join the discussion thread on Physics Forums

Read Part 2 of this Physics Forums series