​Human-Centered AI (BAI613)​

​MODULE 5​
​GOVERNANCE STRUCTURES - 2​
​ opics:​ ​Government​ ​Interventions​ ​and​ ​Regulations,​ ​Introduction:​ ​Driving​ ​HCAI​ ​Forward,​
T
​Assessing Trustworthiness, caring for and Learning from Older Adults​

​1. Government Interventions and Regulations​

​Introduction​
​ overnments​ ​across​ ​the​ ​world​​now​​recognize​​AI​​as​​a​​critical​​technology​​for​​economic​​growth,​
G
​national​ ​power,​ ​and​ ​societal​ ​development.​ ​Earlier,​ ​there​​was​​resistance​​to​​regulation​​because​​it​
​might​​slow​​innovation,​​but​​now​​it​​is​​understood​​that​​without​​regulation,​​AI​​can​​cause​​harm,​​bias,​
​and misuse. Hence, controlled intervention is necessary.​
​Global AI Leadership and Policies​
​Different countries are shaping AI in different ways:​
​●​ ​ hina​ ​focuses​ ​on​ ​rapid​ ​development​ ​with​ ​strong​ ​government​ ​support​ ​and​ ​large-scale​
C
​data collection.​
​●​ ​ nited​ ​States​ ​emphasizes​ ​innovation​ ​through​ ​private​ ​companies​ ​and​ ​research​
U
​institutions.​
​●​ ​European Union​​focuses on strict privacy laws and​​ethical AI practices.​
​This shows that AI governance depends heavily on cultural values and political systems.​
​Need for Regulation​
​AI systems are different from traditional software because they are:​
​●​ ​Adaptive (learn from data)​
​●​ ​Autonomous (make decisions)​
​●​ ​Intelligent (simulate human thinking)​
​ ecause​​of​​these​​characteristics,​​existing​​laws​​are​​not​​sufficient,​​and​​new​​regulatory​​frameworks​
B
​are required to ensure safety and fairness.​
​Human-Centered AI in Policies​
​Modern policies emphasize​​Human-Centered AI (HCAI)​​,​​meaning:​

​HARSH VERMA​ ​1​

​Human-Centered AI (BAI613)​

​●​ ​Humans must remain in control​

​●​ ​AI decisions should be explainable​
​●​ ​Systems should be fair and unbiased​
​The goal is to build trustworthy AI systems that benefit society.​
​Government Role in Research​
​ overnments influence AI through funding programs.​
G
​For example, research initiatives now require:​
​●​ ​Real-world impact (not just theory)​
​●​ ​Collaboration with industry and society​
​This shift is called use-inspired research, focusing on solving real problems.​
​International Principles​
​European Union Principles​
​EU defines 7 key principles like:​
​●​ ​Human oversight​
​●​ ​Transparency​
​●​ ​Fairness​
​●​ ​Accountability​
​These principles guide developers in building ethical AI.​
​OECD Principles​
​These emphasize:​
​●​ ​Inclusive growth​
​●​ ​Sustainability​
​●​ ​Transparency​
​●​ ​International cooperation​
​These are widely accepted globally.​
​Role of International Organizations​

​HARSH VERMA​ ​2​

​Human-Centered AI (BAI613)​

​Organizations like the​​United Nations (UN)​​promote AI for global good through initiatives like:​
​●​ ​Sustainable Development Goals​
​●​ ​AI for Good programs​
​However, they currently lack strong enforcement power.​
​Real-World Regulation Examples​
​Tesla Accident​
​A fatal accident showed that over-reliance on automation is dangerous.​
​Lesson: Human supervision is always necessary.​
​Medical AI (FDA)​
​The FDA regulates AI-based medical devices to ensure:​
​●​ ​Safety​
​●​ ​Effectiveness​
​This is important because mistakes can be life-threatening.​
​Aviation AI​
​AI in aircraft systems must be:​
​●​ ​Predictable​
​●​ ​Transparent​
​●​ ​Controlled​
​Even small errors can lead to disasters.​
​Key Challenges​
​Surveillance Capitalism​
​Companies collect user data and use AI to:​
​●​ ​Predict behavior​
​●​ ​Target ads​
​Raises serious privacy concerns.​
​Misinformation​

​HARSH VERMA​ ​3​

​Human-Centered AI (BAI613)​

​AI helps spread:​

​●​ ​Fake news​
​●​ ​Hate speech​
​Governments may regulate platforms to protect society.​
​Regulatory Capture​
​ ometimes companies influence regulators, leading to weak rules.​
S
​This reduces the effectiveness of governance.​

​2. Introduction: Driving HCAI Forward​

​Core Idea​
​AI should not only be powerful but also:​
​●​ ​Safe​
​●​ ​Reliable​
​●​ ​Human-friendly​
​HCAI ensures that technology aligns with human values.​
​Research Directions​

​HARSH VERMA​ ​4​

​Human-Centered AI (BAI613)​

​Interdisciplinary Approach​
​HCAI combines:​
​●​ ​Technology​
​●​ ​Ethics​
​●​ ​Design​
​●​ ​Social sciences​
​This ensures AI works well in real-world environments.​

​HARSH VERMA​ ​5​

​Human-Centered AI (BAI613)​

​Grand Challenges​
​Citizen Science​
​ I helps ordinary people contribute to research (e.g., environmental monitoring).​
A
​This increases public participation and awareness.​
​Misinformation Control​
​ I + human moderation is needed to detect and reduce fake content.​
A
​This is crucial for protecting democracy and society.​
​Healthcare Innovation​
​AI helps in:​
​●​ ​Drug discovery​
​●​ ​Disease prediction​
​ xample: Protein folding (AlphaFold).​
E
​Improves medical research significantly.​
​Education and Industry​
​●​ ​Universities are adding AI ethics courses​
​●​ ​Companies use ethical AI as a competitive advantage​
​Trustworthy systems attract more users.​
​Key Insight​
​Best innovation happens when:​
​●​ ​Research + real-world problems combine​
​Practical impact improves both research and society.​

​3. Assessing Trustworthiness​

​Core Idea​
​Trustworthiness means an AI system deserves trust, not just appears trustworthy.​
​1.​ ​General Virtues of the System​

​HARSH VERMA​ ​6​

​Human-Centered AI (BAI613)​

​Attribute​ ​Description / Meaning​

​Trustworthy​ ​Can users trust the system to perform correctly?​

​Responsible / Humane​ I​ s​ ​the​ ​system​ ​designed,​ ​developed,​ ​and​ ​tested​

​responsibly?​

​Ethical Design​ ​ ere​ ​stakeholders​ ​involved​ ​during​ ​the​ ​design​

W
​process?​

​Ethical Data​ ​Was the data collected in a fair and ethical manner?​

​Ethical Use​ ​ ill​ ​the​ ​system​ ​be​ ​used​ ​ethically​ ​in​ ​real-world​
W
​scenarios?​

​ ell-being​
W ​/​ ​ oes​​the​​system​​support​​human​​health,​​comfort,​​and​
D
​Benevolence​ ​values?​

​Secure​ ​ ow​ ​resistant​ ​is​ ​the​ ​system​ ​to​ ​cyber​ ​attacks​ ​or​
H
​misuse?​

​Private​ ​ oes​ ​the​ ​system​ ​protect​ ​user​ ​identity​ ​and​ ​personal​

D
​data?​

​2.​ ​Performs Well in Practice​

​Attribute​ ​Description / Meaning​

​Robust / Agile​ ​ oes​ ​the​ ​system​ ​perform​ ​well​ ​even​​when​

D
​inputs change?​

​Reliable / Dependable​ ​ oes​ ​the​ ​system​ ​consistently​ ​produce​

D
​correct results?​

​Available​ I​ s​ ​the​ ​system​ ​accessible​ ​and​ ​functioning​

​when needed?​

​Resilient / Adaptive​ ​ an​ ​the​ ​system​ ​recover​ ​from​ ​failures​ ​or​

C
​disruptions?​

​ estable​
T ​/​ ​Verifiable​ ​/​ ​ an​ ​the​ ​system​ ​be​ ​tested​ ​and​ ​validated​
C
​Validatable / Certifiable​ ​against requirements?​

​HARSH VERMA​ ​7​

​Human-Centered AI (BAI613)​

​Safe​ ​ oes​ ​the​ ​system​ ​have​ ​a​​proven​​history​​of​

D
​safe operation?​

​3.​ ​Clarity to Stakeholders​

​Attribute​ ​Description / Meaning​

​Accurate​ ​ oes​​the​​system​​provide​​correct​​results​​in​​both​
D
​tests and real-world use?​

​Fair / Unbiased​ ​ re​ ​biases​

A ​identified,​ ​minimized,​ ​and​
​reported?​

​Accountable / Liable​ I​ s​ ​responsibility​ ​clearly​ ​assigned​ ​for​ ​system​

​outcomes?​

​Transparent​ ​ an​ ​external​ ​observers​ ​understand​ ​how​ ​the​

C
​system works?​

I​ nterpretable​​/​​Explainable​​/​ ​ an​ ​the​ ​system​ ​explain​ ​how​ ​it​ ​arrived​ ​at​

C
​Intelligible​ ​decisions?​

​Usable​ I​ s​ ​the​ ​system​ ​easy​​and​​efficient​​for​​humans​​to​

​use?​

​4. Enables Independent Oversight​

​Attribute​ ​Description / Meaning​

​Auditable​ ​Can the system be reviewed after failures for analysis?​

​Trackable​ ​ oes​​the​​system​​show​​current​​status​​and​​next​​actions​​for​​human​
D
​intervention?​

​Traceable​ ​Can outcomes be traced back to their root cause?​

​Redressable​ I​ s​ ​there​ ​a​ ​mechanism​ ​for​ ​users​ ​to​ ​report​ ​harm​ ​and​ ​seek​
​compensation?​

​Insurable​ ​Can insurance be provided based on system reliability?​

​Recorded​ ​Does the system maintain logs for future analysis?​

​Open​ ​Are code and data accessible for review and transparency?​

​HARSH VERMA​ ​8​

​Human-Centered AI (BAI613)​

​Certifiable​ ​Can the system be officially approved for use?​

​5.​ ​Complies with Accepted Practices​

​Attribute​ ​Description / Meaning​

​Compliant with Standards​ ​ oes​ ​the​ ​system​ ​follow​ ​recognized​

D
​standards (e.g., IEEE guidelines)?​

​ ompliant​ ​with​ ​Software​

C ​ as​​the​​system​​developed​​using​​proper​​and​
W
​Engineering Workflows​ ​trusted engineering processes?​

​Table 1: Attributes of HCAI Systems (Organized into Five Categories)​

​Step​ C
​ omponen​ ​Description​
​t​

​1​ ​Definitions​ A
​ gree​ ​on​ ​a​ ​clear​ ​and​ ​limited​ ​set​ ​of​ ​terms​ ​(e.g.,​ ​fairness,​ ​transparency,​
​reliability)​​that​​can​​be​​assessed.​​These​​definitions​​should​​be​​precise​​and​
​may use scoring methods for evaluation.​

​2​ ​Methods​ ​ ecide​​how​​trustworthiness​​will​​be​​assessed​​using​​multiple​​approaches:​

D
​•​ ​Static​ ​analysis​ ​of​ ​code​ ​and​ ​training​ ​data​ ​•​ ​Benchmark​ ​testing​
​performance​​•​​Real-world​​system​​performance​​•​​Stakeholder​​perceptions​
​(subjective evaluation) • Reviews by internal/external experts​

​3​ ​Assessors​ ​ ecide​ ​who​ ​will​ ​evaluate​ ​the​ ​system​ ​and​ ​how​ ​many​ ​evaluators​ ​are​
D
​needed.​​Possible​​assessors​​include:​​•​​System​​developers​​(may​​have​​bias)​
​•​​Independent​​developers​​with​​similar​​expertise​​•​​Project​​managers​​using​
​the​ ​system​ ​•​ ​End​ ​users​ ​affected​ ​by​ ​the​ ​system​ ​•​ ​Independent​ ​oversight​
​experts​

​4​ ​Audience​ I​ dentify​ ​who​ ​will​ ​use​ ​or​ ​benefit​ ​from​ ​the​ ​assessment​ ​results:​ ​•​
​Developers​ ​(to​ ​improve​ ​the​ ​system)​ ​•​ ​Other​ ​developers​ ​(to​ ​learn​ ​from​
​findings)​ ​•​ ​Managers​ ​(to​ ​refine​ ​usage​ ​strategies)​ ​•​ ​Users​ ​or​ ​affected​
​stakeholders (for awareness and trust)​

​Table 2: HCAI Trustworthiness Assessment Process​

​Trust vs Trustworthiness​
​●​ ​Trust → User belief​

​HARSH VERMA​ ​9​

​Human-Centered AI (BAI613)​

​●​ ​Trustworthiness → System reliability​

​A system may look trustworthy but fail in reality.​

​Sl. No.​ ​Evaluation Criteria​

​1​ ​Openly available, thorough report​

​2​ ​Report contains positive results​

​3​ ​Internal independent review before implementation​

​4​ ​Audit trail implemented​

​5​ ​Training data collected and properly assessed (e.g., datasheets)​

​6​ ​Software verified and validated​

​7​ ​Fairness testing conducted​

​8​ ​Explainability implemented and tested​

​9​ ​Performance evaluated in first two months of usage​

​10​ ​Performance evaluated in first six months of usage​

​11​ ​Stakeholder involvement (feedback, incident reporting, suggestions)​

​12​ ​Internal review process with redress mechanisms​

​13​ ​Continuous reviews for system improvement​

​14​ ​External review by independent oversight body​

​Table 3: Proposed HCAI Trustworthiness Scale​

​Important Attributes​
​System Quality​
​AI should be:​
​●​ ​Safe​
​●​ ​Reliable​
​●​ ​Secure​
​Performance​

​​
1
​HARSH VERMA​
​0​
​Human-Centered AI (BAI613)​

​System must:​
​●​ ​Work correctly​
​●​ ​Adapt to changes​
​Transparency​
​Users should understand:​
​●​ ​How decisions are made​
​Oversight​
​System should allow:​
​●​ ​Auditing​
​●​ ​Tracking​
​●​ ​Accountability​
​Compliance​
​System must follow:​
​●​ ​Standards​
​●​ ​Regulations​
​Evaluation Methods​
​Objective Measurement​
​Difficult because ethics cannot be measured like numbers.​
​Scoring by Components​
​Similar to rating systems:​
​●​ ​Divide into factors​
​●​ ​Assign scores​
​Helps compare systems.​

​​
1
​HARSH VERMA​
​1​
​Human-Centered AI (BAI613)​

​Social Processes​
​Includes:​
​●​ ​Expert reviews​
​●​ ​Committees​
​●​ ​Audits​
​Important for critical systems.​
​Trustworthiness Scale​
​Evaluation includes:​
​●​ ​Data quality​
​●​ ​Fairness​
​●​ ​Explainability​
​●​ ​Real-world performance​
​Score helps judge system reliability.​

​​
1
​HARSH VERMA​
​2​
​Human-Centered AI (BAI613)​

​Challenge​
​AI systems:​
​●​ ​Change over time​
​●​ ​Depend on data​
​So trustworthiness is not fixed it evolves.​

​4. Caring for and Learning from Older Adults​

​Importance​
​Older adults:​
​●​ ​Need support​
​●​ ​Also provide knowledge and experience​
​AI should support both roles.​
​HCAI Approach​
​Instead of focusing on technology, focus on:​
​●​ ​User needs​
​●​ ​Real-life problems​
​Needs of Older Adults​
​Includes:​
​●​ ​Mobility​
​●​ ​Healthcare​
​●​ ​Communication​
​●​ ​Safety​
​●​ ​Daily tasks​
​Needs vary widely among individuals.​

​Older Adult Needs​ ​Example Tasks with Devices that Serve Needs​

​​
1
​HARSH VERMA​
​3​
​Human-Centered AI (BAI613)​

​Mobility​ ​ se​​canes,​​walkers,​​powered​​wheelchairs,​​public​​transportation,​
U
​car driving, taxi services​

​ ood​
F ​Preparation,​ U
​ se​ ​simple/safe​ ​preparation​ ​tools,​​cooking​​appliances,​​clearing​
​Cooking, and Cleaning​ ​tables, cleaning dishes​

​Personal Care​ ​ ash,​ ​bathe,​ ​care​ ​for​ ​teeth,​ ​mouth,​ ​and​ ​skin,​ ​brush​ ​hair,​
W
​dress/undress, manage personal appearance​

​Medical Care​ ​ nsure​ ​medication​ ​adherence;​ ​use​ ​medical​ ​devices​ ​like​​insulin​

E
​pumps, pacemakers, prosthetics; attend telehealth appointments​

​Medical Monitoring​ ​ se​ ​and​ ​record​​data​​from​​weight​​scale,​​thermometer,​​oximeter,​

U
​blood​​pressure​​monitor,​​heart​​rate​​monitor,​​insulin​​testing,​​ECG,​
​EEG​

​ ellness​ ​and​ ​Emotional​ T

W ​ rack​​nutrition,​​sleep,​​exercise,​​yoga,​​mood,​​social​​interactions,​
​Support​ ​cognitive and physical health​

​Shopping​ ​ rder​​food,​​medicines,​​clothes,​​and​​other​​essentials;​​receive​​and​
O
​store deliveries​

​Finances​ ​ rack​ ​income​ ​and​ ​expenses,​ ​prepare​ ​taxes,​ ​pay​ ​bills,​ ​manage​
T
​financial assets​

I​ nformation​ ​and​ A
​ ccess​​online​​information​​on​​medical,​​legal,​​financial​​services,​
​Education​ ​and learning platforms​

​News and Entertainment​ ​ ccess​​news,​​sports,​​music,​​movies,​​hobbies,​​games,​​books,​​and​

A
​photos​

​ ommunication​
C ​and​ U
​ se​ ​email,​​messaging,​​video​​calls,​​social​​media,​​calendars,​​and​
​Human Connection​ ​digital communication tools​

​Security​ ​ se​ ​intrusion​ ​detection​ ​systems,​ ​monitoring​ ​devices,​ ​fire​

U
​alarms; request help in emergencies​

​House Cleaning​ ​ se​ ​vacuum​ ​cleaners,​ ​mops,​ ​dusting​ ​tools,​ ​bathroom​ ​cleaning​
U
​devices, garbage disposal systems​

​House Maintenance​ ​ aintain​ ​appliances,​ ​furniture,​ ​plumbing,​ ​electrical​ ​systems,​

M
​heating, ventilation, and air conditioning​

​​
1
​HARSH VERMA​
​4​
​Human-Centered AI (BAI613)​

​Gardening and Pet Care​ ​ ater​ ​plants,​ ​gardening,​ ​maintaining​ ​outdoor​ ​spaces,​ ​pet​ ​care,​
W
​dog walking​

​Mentoring​ ​ uide​ ​younger​ ​people,​ ​tutor​ ​students,​ ​deliver​ ​lectures,​ ​mentor​

G
​professionals​

​Contributing​ ​ articipate​ ​in​ ​volunteering,​ ​community​ ​service,​ ​religious​

P
​activities, citizen science, writing memoirs​

​Creative Projects​ ​ ngage​ ​in​ ​writing,​ ​music,​ ​crafts,​ ​painting,​ ​photography,​

E
​cooking, and family/community history projects​

​Table 4: Older Adult Needs and Supporting Tasks/Devices​

​Design Philosophy​
​Technology-Centered (Wrong Approach)​
​●​ ​Focus only on robots​
​Human-Centered (Correct Approach)​
​●​ ​Support independence​
​●​ ​Improve quality of life​
​Issues with Care Robots​
​●​ ​Lack emotional connection​
​●​ ​Reduce human interaction​
​Not always suitable as replacements.​
​Better Solutions​
​Smart Devices​
​●​ ​Medication reminders​
​●​ ​ utomatic appliances​
A
​Help users stay independent.​
​Example 1: Smart Wheelchairs​
​●​ ​Combine human control + AI​
​●​ ​Provide navigation assistance​

​​
1
​HARSH VERMA​
​5​
​Human-Centered AI (BAI613)​

​●​

​Example 2: Smart Dining Solutions​

​●​ ​ uilt-in dishwasher tables​
B
​Reduce physical effort.​

​​
1
​HARSH VERMA​
​6​
​Human-Centered AI (BAI613)​

​Example 3: AI Yoga Apps​

​●​ ​Personalized exercise​
​●​ ​Health tracking​
​Improves physical and mental well-being.​
​Social Integration​
​Human support is essential:​
​●​ ​Family​
​●​ ​Caregivers​
​●​ ​Community​

​AI should support, not replace humans.​

​​
1
​HARSH VERMA​
​7​