Not Really Problems with AI

13 things blamed on AI, struck through

Not everything blamed on AI holds up. Some concerns are debunked, some misunderstood, and some are real problems that aren't actually about AI. Each claim below is struck through, with what the evidence actually says.

13 concerns blamed on AI that don't survive scrutiny — some debunked, some misunderstood, and some real problems that aren't actually about AI.

6 misunderstood 5 debunked 1 not-AI · broader 1 not-AI · critical

Claim → reality

Misunderstood
The claimAI erodes critical thinking

Reality: With reflective scaffolding, AI can increase metacognition.
Contrary to concerns that AI tools diminish critical thinking abilities, research shows that with proper reflective scaffolding, AI can actually enhance metacognition and critical thinking skills.

Sources: Wang & Fan 2025
Misunderstood
The claimAI makes you less creative

Reality: Co-creative tools raise idea volume/diversity; issue is over-reliance, not the tech per se.
Research shows that co-creative AI tools can actually increase the volume and diversity of ideas. The real issue is over-reliance on AI, not the technology itself.

Sources: Wang & Fan 2025
Debunked
The claimIf AI helps, you didn't learn

Reality: Demonstrations + critique accelerate skill-building (worked-example effect).
Using AI for learning doesn't prevent genuine skill acquisition. In fact, demonstrations followed by critique can accelerate skill-building through the worked-example effect.

Sources: Wang & Fan 2025
Misunderstood
The claimKids just use AI to cheat

Reality: Same panic as calculators/Wikipedia; pedagogy must evolve.
The concern about students using AI to cheat echoes similar panics about calculators and Wikipedia. The solution lies in evolving pedagogical approaches rather than banning the technology.

Sources: Wang & Fan 2025
Misunderstood
The claimLLMs hallucinate so they're useless

Reality: RAG + hallucination-aware tuning sharply cut error rates.
While language models do sometimes generate incorrect information ('hallucinations'), techniques like Retrieval-Augmented Generation (RAG) and hallucination-aware tuning have significantly reduced error rates.

Sources: Song et al. 2024
Misunderstood
The claimLLMs don't understand anything

Reality: On many reasoning benchmarks they beat humans; 'understanding' is philosophically contested.
The claim that language models don't 'understand' anything is complicated by the fact that they outperform humans on many reasoning benchmarks. The concept of 'understanding' itself is philosophically contested.

Sources: A Landscape of Conciousness, Kuhn et al · Taking AI Welfare Seriously
Debunked
The claimAI isn't real / it's all hype

Reality: Already in hospitals, logistics, media; denial wastes time & cedes ground.
Claims that AI is 'just hype' ignore its already substantial deployment in hospitals, logistics, media, and other sectors. Denial wastes valuable time and cedes ground in important policy discussions.

Sources: Wikipedia - The AI Effect · Transformer Explainer
Debunked
The claimQuantum AGI has secretly taken over

Reality: No AGI exists; confusion comes from sci-fi tropes.
Claims about 'quantum AGI' secretly taking over are unfounded. No AGI (Artificial General Intelligence) currently exists, and such concerns stem from science fiction tropes rather than technical reality.

Sources: Ian Goodfellow, Yoshua Bengio and Aaron Courville
Debunked
The claimAI art is always bad

Reality: Humans detect AI art only ~60% correct; audiences often prefer it.
The claim that AI-generated art is always inferior to human art is contradicted by research showing that humans can only correctly identify AI art about 60% of the time, and audiences often prefer AI-generated content.

Sources: Chein et al. 2024 · The AI Art Turing Test
Misunderstood
The claimAI art is theft; opt-out protects artists

Reality: Mass opt-out ≈ cultural erasure of marginalized styles from future models.
While concerns about AI art and copyright are legitimate, mass opt-out approaches can lead to the cultural erasure of marginalized artistic styles from future models, creating representation problems.

Sources: Stranisci & Hardmeier 2025 · AI & Copyright
Debunked
The claimCompanies know exactly what they're doing with AI

Reality: Enterprise adoption is high, but in-house expertise lags; most pilots yield modest gains.
Despite high enterprise adoption rates, most companies lack sufficient in-house AI expertise. The majority of AI pilot projects yield only modest gains, indicating a gap between adoption and effective implementation.

Sources: Stanford AI Index 2025
Ongoing
The claimStructural exploitation

Reality: Resource-rich nations stay poor due to extractive systems, not tech itself.
The persistence of poverty in resource-rich nations is primarily due to extractive economic and political systems, not technology itself. These structural inequalities exist independently of AI, though new technologies may interact with these existing dynamics.

Sources: Brevini 2024
Critical
The claimDirty power generation

Reality: Fossil fuel energy production is a global problem independent of specific technologies.
The environmental impact of fossil fuel-based power generation is a global challenge that affects all industries and technologies, not just AI. This is a broader energy policy and infrastructure issue.

Sources: IPCC Climate Change 2023

Part of the AI Problems Index · see the Risk Atlas and Environmental Impact.