In this episode of AI Ethics with Fexingo, Lucas and Luna explore a surprising source of bias: when AI models are trained on data generated by older AI systems. They dive into a 2025 study from researchers at Oxford and MIT showing that language models fed synthetic text from earlier models develop 'model collapse'—becoming less diverse and more error-prone over generations. Lucas explains why this matters for everything from customer service chatbots to medical diagnostic tools, and Luna pushes him on whether this is a fixable technical problem or a fundamental limit. They also discuss the proposed 'data provenance' tagging system some companies are experimenting with to label AI-generated content in training sets. This episode ends with the hosts reflecting on how the solution might ironically require more human oversight, not less.

#AIEthics #ModelCollapse #BiasInAI #SyntheticData #DataProvenance #OxfordStudy #MITStudy #LanguageModels #MachineLearning #AISafety #ResponsibleAI #Tech #ArtificialIntelligence #DataQuality #GenerativeAI #FexingoBusiness #BusinessPodcast #Technology

Keep every episode free: buymeacoffee.com/fexingo

In this episode, Lucas and Luna explore how AI hiring tools don't just inherit bias from the resumes they screen—they learn it from the job postings themselves. They walk through a 2025 study from the University of Chicago that analyzed 10,000 job descriptions on LinkedIn and found that postings using masculine-coded language (like 'aggressive,' 'dominant,' or 'ninja') caused AI models to rank male candidates higher even when qualifications were identical. The hosts discuss how this 'posting bias' creates a feedback loop: biased postings attract biased applicants, which then reinforce the model's skewed scoring. They also talk about how companies like Unilever and Hilton have restructured their job ads to reduce gender-coded language, with Hilton seeing a 12% increase in female applicants for technical roles after a rewrite. Lucas brings in a startling stat: 88% of Fortune 500 companies now use some form of AI screening, yet fewer than 15% audit their job descriptions for language bias. Luna asks whether the solution is better training data or a fundamental redesign of how we write job ads. The episode closes on a forward-looking note: as more states pass algorithmic accountability bills, the real question may not be whether AI can be fair, but whether companies are willing to change the human inputs they feed it.

#AIEthics #HiringBias #JobPostings #GenderBias #ResumeScreening #AlgorithmicBias #HRTech #LinkedInStudy #UniversityOfChicago #Unilever #Hilton #Fortune500 #AIandEmployment #FairHiring #Technology #FexingoBusiness #BusinessPodcast #AIAccountability

Keep every episode free: buymeacoffee.com/fexingo

In this episode of AI Ethics with Fexingo, Lucas and Luna explore how AI models used in court sentencing can inherit racial and socioeconomic biases from historical sentencing data. They focus on the COMPAS algorithm, a risk assessment tool used in U.S. courts, which was found to falsely flag Black defendants as high-risk at nearly twice the rate of white defendants. The hosts discuss a 2024 study from the University of Chicago that showed how retraining the model on clean data reduced this disparity by 60 percent. They also examine the broader implications for AI in the justice system, including recent legislative efforts in California and the EU AI Act. The episode includes a brief, listener-supported funding segment that helps keep the show ad-free.

#AIEthics #BiasInAI #CriminalJusticeAI #COMPAS #AlgorithmicFairness #RacialBias #SentencingAI #PredictivePolicing #EUAIAct #CaliforniaAI #UniversityOfChicago #RiskAssessment #MachineLearning #FairnessInAI #Technology #AIandLaw #FexingoBusiness #BusinessPodcast

Keep every episode free: buymeacoffee.com/fexingo

In this episode of AI Ethics with Fexingo, Lucas and Luna explore how reinforcement learning from human feedback (RLHF) can introduce subtle biases into language models. They dive into a real-world example from early 2026, when a major chatbot began refusing to answer questions about certain historical figures after receiving thousands of user flaggings — even when those flaggings were incorrect or politically motivated. The hosts explain how RLHF works, why crowd-sourced feedback amplifies majority viewpoints, and what companies like Anthropic and Google are doing to fix it. They also discuss the tension between safety and overcorrection, and whether users should have more say in shaping model behavior without undermining fairness. A concrete look at why your chatbot's polite refusal to engage might itself be a form of bias.

#AIEthics #RLHF #Bias #Chatbots #LanguageModels #HumanFeedback #Anthropic #Google #Safety #Fairness #AIAlignment #Crowdsourcing #Technology #FexingoBusiness #BusinessPodcast #ArtificialIntelligence #ModelTraining #ContentModeration

Keep every episode free: buymeacoffee.com/fexingo

In this episode of AI Ethics with Fexingo, Lucas and Luna explore a surprising source of algorithmic bias: satellite imagery. They discuss a 2025 study from Stanford's Center for International Development that found object detection models trained on satellite photos systematically misidentify certain infrastructure in low-income and rural areas—like mislabeling solar panels as rooftops or failing to detect paved roads in regions with darker soil. The hosts walk through how training data from high-res urban environments creates a 'geographic blind spot,' leading to downstream impacts on disaster response, resource allocation, and insurance pricing. They also touch on what's being done to fix it, including new datasets deliberately balanced across income levels and latitudes. A specific, grounded look at how the view from space can reinforce inequality.

#AIEthics #BiasInAI #SatelliteImagery #ObjectDetection #Stanford #GeographicBias #InfrastructureMapping #DisasterResponse #TechInequality #MachineLearning #ComputerVision #TrainingData #ResponsibleAI #Technology #FexingoBusiness #BusinessPodcast #AIEthicsPodcast #SatelliteAI

Keep every episode free: buymeacoffee.com/fexingo

Episode 119 of AI Ethics with Fexingo examines how credit scoring models can inherit and amplify racial and socioeconomic bias. Lucas and Luna dive into a 2025 Federal Reserve study showing that alternative data like utility payments and rental history can reduce bias when used carefully, but also risk creating new forms of discrimination. They discuss the case of Upstart, a lending platform that uses AI to approve more borrowers while maintaining lower default rates, yet still faces scrutiny from the Consumer Financial Protection Bureau. The hosts explore the tension between financial inclusion and fair lending laws, and ask whether regulators are equipped to audit complex machine learning models. A timely look at the hidden biases in the algorithms that decide who gets access to credit.

#CreditScoring #AlgorithmicBias #FairLending #FederalReserve #Upstart #CFPB #AlternativeData #FinancialInclusion #MachineLearning #AIEthics #Technology #FexingoBusiness #BusinessPodcast #LucasAndLuna #Discrimination #SocioeconomicBias #LendingAlgorithms #RegulatoryChallenge

Keep every episode free: buymeacoffee.com/fexingo

In episode 118 of AI Ethics with Fexingo, Lucas and Luna dive into a specific, overlooked source of algorithmic bias: the resume itself. They trace how AI hiring tools inherit prejudice not just from historical decisions or biased training data, but from the very format and language of resumes—subtle cues like name length, font choice, and the presence of hobbies that signal socioeconomic class. The episode focuses on a 2025 study from the University of Toronto, which found that AI models trained on a dataset of 1.2 million resumes developed a systematic preference for candidates with 'polished' formatting and 'professional' activity keywords, penalizing applicants who listed non-traditional work experience or omitted a LinkedIn URL. Lucas and Luna unpack the specific failure mode: the models learned to associate certain resume structures with higher callback rates from past human recruiters, embedding those human biases into automated screening. They also explore a proposed fix—'resume anonymization 2.0' that strips stylistic markers—and why companies are slow to adopt it. No sweeping claims; just one concrete bias pathway and what it means for job seekers today.

#AIEthics #BiasInAI #HiringAlgorithms #ResumeFiltering #JobSearch #AlgorithmicFairness #UniversityOfToronto #ResumeBias #AIBias #HRTech #ResponsibleAI #AutomatedHiring #JobMarket2026 #Discrimination #FairHiring #Technology #FexingoBusiness #BusinessPodcast

Keep every episode free: buymeacoffee.com/fexingo

Episode 117 of AI Ethics with Fexingo examines how AI models used in home appraisal and mortgage underwriting have absorbed and amplified historical housing discrimination. Lucas and Luna walk through a specific 2025 study from the Urban Institute that found automated valuation models systematically undervalued homes in majority-Black neighborhoods by 8 to 12 percent compared to similar homes in white neighborhoods. They trace the root cause: training data from the 2010s that encoded decades of redlining and steering practices. The conversation also covers the 2024 FHFA rule requiring appraisal algorithm audits, and why simply removing race labels from input data fails to fix the bias. A concrete look at how fairness in property AI is still catching up to policy intentions.

#AIEthics #Bias #Housing #Appraisal #AlgorithmicBias #Redlining #UrbanInstitute #FHFA #Mortgage #Technology #Podcast #FairHousing #RealEstate #DataBias #Homeownership #Regulation #FexingoBusiness #BusinessPodcast

Keep every episode free: buymeacoffee.com/fexingo

Episode 116 of AI Ethics with Fexingo investigates how dynamic pricing algorithms have evolved beyond surge pricing into personalized price discrimination. Lucas and Luna examine a 2026 case study where a major retailer's AI pricing model charged loyal customers higher prices based on browsing history and purchase frequency. They discuss the opaque algorithms that set prices in real time, the ethical lines between market efficiency and exploitation, and why regulators are starting to scrutinize 'price optimization' in everything from groceries to airline tickets. The hosts also explore how models trained on customer data can inadvertently encode wealth-based bias, making the same product cost more for people who can least afford it. With concrete examples and specific numbers, this episode offers a clear-eyed look at one of the most pervasive yet invisible forms of AI-driven bias.

#AIEthics #DynamicPricing #PriceGouging #AlgorithmicBias #ConsumerProtection #PersonalizedPricing #PriceDiscrimination #RetailTech #Regulation #FexingoBusiness #BusinessPodcast #Technology #LucasAndLuna #SurgePricing #CustomerData #AIandSociety #EthicalAI #Transparency

Keep every episode free: buymeacoffee.com/fexingo

In this episode of AI Ethics with Fexingo, Lucas and Luna explore how AI systems are inadvertently learning bias from public transit data. Using a concrete example from the Los Angeles Metro, they show how predictive maintenance models trained on historical ridership data can systematically under-serve lower-income neighborhoods. They discuss the feedback loop where reduced service leads to less data, leading to even more reduced service, and what transit authorities can do to break it. The hosts also touch on the broader implications for urban planning and equity, and briefly note how listener support keeps the show ad-free. This episode drills into a specific case—the Metro's 2024 data bias audit—and offers a tangible takeaway for anyone working with geospatial or infrastructure data.

#PublicTransit #AlgorithmicBias #LosAngelesMetro #PredictiveMaintenance #DataFeedbackLoop #UrbanPlanning #TransportationEquity #GeospatialData #InfrastructureAI #BiasAudit #AIEthics #Technology #FexingoBusiness #BusinessPodcast #Podcast #DataScience #MachineLearning #Equity

Keep every episode free: buymeacoffee.com/fexingo