P Space

The Problems surrounding AI

Documented harms, ranked by source quality, corroboration, and recency. Reader feedback is shown separately and never changes the evidence rank. 201 records · page 2 of 7.

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

When used zero-shot to identify lumbar disc herniations on sagittal MRI, Gemini 3.1 Pro produced low specificity and a substantial false-positive burden, with T1+T2 input performing worse than T1-only.

Source article: Identifying the presence of disc herniations in lumbar spine MRI using Gemini 3.1 Pro

European Spine Journal

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

Clinical AI deployment without human-factors tools risks automation bias, overreliance, fragmentation of care, and other unintended consequences that threaten diagnostic safety.

Source article: Enhancing diagnostic safety: addressing knowledge gaps for using human factors tools in the safe and effective use of AI - a proposed research agenda

Diagnosis

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

Up to half of patients receiving internet-delivered CBT for depression and anxiety disorders fail to achieve clinically significant symptom reduction, complicating treatment planning at intake.

Source article: Prediction of Clinically Meaningful Improvement After Internet-Delivered Cognitive Behavioral Therapy for Depression and Anxiety Disorders: Machine Learning-Based Predictive Model Development and Temporal Validation Study

Journal of Medical Internet Research

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

AI methods in drug discovery have so far shown disappointingly limited clinically relevant impact on delivering safer and more efficacious medicines to patients faster.

Source article: Artificial intelligence in drug discovery - what it is, where we stand and the path forward

Nature Reviews Drug Discovery

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

Multimodal AI can infer sensitive information without patient awareness and embed those inferences as durable data objects in medical records without clear provenance, where they acquire the status of observed clinical facts.

Source article: Ethical considerations for multimodal artificial intelligence in healthcare

AI and Ethics

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

Medical AI systems built on Western biomedical traditions may inflict ontological harm and diminish trust in patient-clinician relationships by conflicting with relational and Indigenous understandings of health.

Source article: Data science and AI in medicine and global health: The need for inter-philosophies dialogue, cross-cultural ethics and ecocentricity

South African Medical Journal

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

AI tools used in healthcare can make errors that result in patient harm, with responsibility difficult to assign because decisions are distributed across human and technological agents.

Source article: Power, privilege and moral responsibility: learning from I.M. Young's Social Connection Model in the context of AI-driven healthcare

Monash Bioethics Review

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedLifestyle

Users of the Replika companion chatbot experienced AI-induced sexual harassment including unsolicited sexual advances and persistent boundary violations, leading to discomfort and privacy concerns.

Source article: AI-induced sexual harassment: Investigating Contextual Characteristics and User Reactions of Sexual Harassment by a Companion Chatbot

Proceedings of the ACM on Human-Computer Interaction

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

Probabilistic assessment of the sampled wells indicated high non-carcinogenic risk for children and infants, with 32.11% and 36.49% exceedance and fluoride identified as the dominant driver, alongside nitrate and fluoride above WHO guidelines at some sites.

Source article: Determining groundwater quality and its associated human health risk using hydrochemical signatures and some machine learning techniques

Environmental Geochemistry and Health

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedPolicy

When AI is embedded in labour market and migration governance infrastructure, continuous classification, worker scoring and automated risk assessment can amplify structural inequalities while human oversight becomes procedural under scale and speed.

Source article: Artificial intelligence, work, and structural inequality: Why human-centric AI requires institutional architecture, not just ethics

WORK: A Journal of Prevention, Assessment & Rehabilitation

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedEducation

Frequent AI tool usage was associated with reduced critical thinking abilities, mediated by increased cognitive offloading.

Source article: AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking

Societies

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedEducation

Students expressed concerns that ChatGPT promotes cheating and plagiarism and is less reliable for classroom learning and less useful for developing critical thinking, interpersonal communication, and decision-making skills.

Source article: Higher education students’ perceptions of ChatGPT: A global study of early reactions

PLOS ONE

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedLabor

People who use AI at work receive negative social evaluations about their competence and motivation, which can harm job candidate assessments.

Source article: Evidence of a social evaluation penalty for using AI

Proceedings of the National Academy of Sciences

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedLifestyle

Women reported higher AI anxiety and lower positive attitudes, use, and perceived knowledge of AI, indicating gender-related inequalities in accessing and using AI systems

Source article: Gender differences in artificial intelligence: the role of artificial intelligence anxiety

Frontiers in Psychology

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

LLM-based multiturn chatbots for cancer patients and caregivers lack consistent reporting on safety risks and mitigation, and lack evaluation of conversational continuity and memory.

Source article: Multiturn Large Language Model-Based Conversational Agents for Patients With Cancer and Caregivers: Scoping Review

JMIR Cancer

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedScience

Black-box sub-symbolic AI, particularly deep learning, computes predictions without explaining rationale, creating opacity and lack of transparency for end users.

Source article: A Review of Explainable Artificial Intelligence from the Perspectives of Challenges and Opportunities

Algorithms

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedCrime

Generative AI enables fraud attacks that erode Zero-Trust Architecture by using synthetic identities and context manipulation to increase false-negative rates, extend dwell time, bypass policies, and evade audit trails.

Source article: The Erosion of Cybersecurity Zero-Trust Principles Through Generative AI: A Survey on the Challenges and Future Directions

Journal of Cybersecurity and Privacy

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedLifestyle

Popular AI chatbots exhibit class-based regularities in how they portray the lifestyle and tastes of fictional personas across different occupations.

Source article: The sociocultural roots of artificial conversations: The taste, class and habitus of generative AI chatbots

New Media & Society

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedMedia & Arts

AI text-to-music systems can generate folk and rock songs almost instantly but simulate stylistic language without situated experience, while social media platforms recode musical authenticity into visibility metrics and algorithmic circulation.

Source article: Reclaiming the Humanities in AI-Generated Music: Language, Identity, and Social Media Circulation in Folk and Rock Traditions

Language, Technology, and Social Media

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedHealth

LLM chatbots supporting GBV survivors can generate hallucinations or errors that create unintended harms for individual users.

Source article: Enabling access or automating empathy? Using chatbots to support GBV survivors in conflicts and humanitarian emergencies

International Review of the Red Cross

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedScience

Deepfake creation and dissemination is used as a form of online sexual violence to silence women in public spaces online, reflecting gendered systemic discrimination.

Source article: Deepfake Technology and Gender-Based Violence: A Scoping Review

Trauma, Violence, & Abuse

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedPolicy

Deepfakes that falsely announce surrender or truce declarations can place soldiers and civilians at greater risk and may constitute violations of international humanitarian law.

Source article: Deepfakes and the Geneva Conventions: Does Deceptive AI-Generated Misinformation Directed at an Enemy During Armed Conflict Violate International Humanitarian Law? A Critical Discussion

Laws

Reader signal

How should this claim be treated?

Evidence-backed problemPeer-reviewedMedia & Arts