Ethical Development and Use of Artificial Intelligence

Ethical Development and Use of Artificial Intelligence Presented by: Ada Chung Lai-ling, Privacy Commissioner for Personal Data, Hong Kong, China Date: 22 November 2021 What is Artificial Intelligence? - Artificial intelligence (AI) refers to technologies that involve the use of computer programs and machines to mimic the problem-solving or decision- making capabilities of human beings - Examples of Al applications include: • Facial recognition; • Speech recognition; • Chatbots; • Data analytics; and • Automated decision-making or recommendation. Facial Recognition - Workflow • 1. Capture the face of a target person • 2. Analyze and record distinguishable landmarks of the face, i.e. nodal points • 3. Convert location of nodal points into numerical data to formulate an unique face template • 4. Compare the face template to other templates in a database to identify the target person Increasing Use of Artificial Intelligence - Background • 1. Al has enormous potential to boost productivity and economic growth • 2. Al is gaining popularity in Hong Kong • 3. Privacy and ethical risks of Al are not negligible • 4. Consensus on legal regulation is yet to form Potential values of Al - Google's study in 2020 showed that 78% of Hong Kong businesses believe that Al is beneficial; 62% of Hong Kong businesses expect to increase Al investment in future - PwC's research in 2017 showed that global GDP could be up to 14% higher in 2030 as a net effect of using Al, mainly driven by: • a) Enhanced productivity; • b) Better personalization of services and products; and • c) Improved quality of services and products. Use of Al in the Financial Sector of Hong Kong - Financial Services • McKinsey's global survey in 2019 showed that 62% of the respondents in the financial sector have applied at least one Al capability • In 2019, the Hong Kong Monetary Authority reported that 89% of surveyed retail banks in Hong Kong adopted or planned to adopt Al applications • The Hong Kong Academy of Finance's research report released in 2021 also suggested that 71% of surveyed financial services companies in the APAC region adopted or planned to adopt Al and Big Data technologies in the next 12 months Examples of inaccurate and unfair Al - Three facial recognition programs shows consistently higher error rate for identification of dark-skinned women (Source: MIT News, 11 Feb 2018) - Amazon's Al recruitment program favored male candidates owing to its maledominated training data (Source: The Guardian, 11 Oct 2018) Metaverse - An Alternative World - Background • Metaverse refers to a highly immersive virtual world where people gather to communicate in shared virtual spaces with the aid of Al-powered avatars • It is envisaged that there will be a 'hyper-real' alternative world as the metaverse expands • Is it the next big thing? Metaverse - Privacy Risks - Excessive collection of data: • More advanced equipment and sensors will be put in homes and workplaces for replicating our actions - Data Profiling: • - Users are giving up more to live in digital worlds Spoofing • Personal data may become susceptible to being copied, stolen, erased or manipulated Impacts of Al on Data Privacy - Big Data Feed Machine Learning Develop/Improve Artificial Intelligence - Possible Impacts on Privacy • 1) Excessive Collection & Retention of Data • 2) Lack of Transparency • 3) Unpredictable Use • 4) Bias and Discrimination • 5) Re-identification Challenges to Data Protection Principle - Data Minimisation - Data Accuracy - Purpose - Specification & Limitation of Use - Transparency International Efforts on Promoting Ethical AI - OCT 2018 Global Privacy Assembly: • - JAN 2019 Singapore: • - First Draft of the Recommendation on the Ethics of Artificial Intelligence OCT 2020 Global Privacy Assembly: • - Recommendation of the Council on Artificial Intelligence SEP 2020 UNESCO: • - Ethics Guidelines for Trustworthy Al MAY 2019 OECD: • - Social Principles of Human-Centric Al APR 2019 European Commission: • - Model Artificial Intelligence Governance Framework (First Edition MAR 2019 Japan: • - Declaration on Ethics and Data Protection in Artificial Intelligence Resolution on Accountability in the Development and Use of Al APR 2021 European Commission: • Proposal for a Regulation Laying Down Harmonized Rules on Artificial Intelligence Guidance on the Ethical Development and Using Artificial Intelligence - Objectives • 1. To provide guidance to enable organizations to develop and use Al in compliance with the requirements under the Personal Data (Privacy) Ordinance (PDPO) and in an ethical manner • 2. To facilitate healthy development and use of Al in Hong Kong • 3. To facilitate Hong Kong to become an innovation and technology hub and world-class smart city - 3 Data Stewardship Values • Being RESPECTFUL  to the rights, interests and reasonable expectations of stakeholders • Being BENEFICIAL  by providing benefits and minimizing harm to stakeholders • Being FAIR  by avoiding bias and discrimination - 7 Ethical Principles for Al • 1. ACCOUNTABILITY  Organizations should: o Be responsible o Be able to provide sound justifications for actions • 2. HUMAN OVERSIGHT  The level of human involvement should: o Be proportionate to the risks and impact of using Al • 3. TRANSPARENCY & INTERPRETABILITY  Organizations should: o Be proportionate to the risks and impact of using Al o Disclose their use of Al and relevant policies o Strive to improve the interpretability of automated decisions • 4. DATA PRIVACY  Organizations should: o Put effective data governance in place to protect personal data privacy • 5. FAIRNESS  Organizations should: o Avoid bias and discrimination in the use of Al • 6. BENEFICIAL AI  The use of Al should: o Provide benefits o Minimize harm to stakeholders • 7. RELIABILITY, ROBUSTNESS & SECURITY  Al systems should: o Operate reliably o Be resilient to errors o Be protected against attacks - Practice Guide on 4 Major Business Processes for Ethical Development and Use of Al • 1. ESTABLISH  Al Strategy and Governance • 2. CONDUCT  Risk Assessment and Human Oversight • 3. EXECUTE  Development of Al Models and Management of Al Systems • 4. FOSTER  Communication and Engagement with Stakeholders - - ESTABILSH AI Strategy and Governance • Formulate Al Strategy • Establish Al Governance Committee • Provide adequate staff training CONDUCT Risk Assessment and Human Oversight • STEP 1:  Conduct a comprehensive and early risk assessment • STEP 2:  Take a risk-based approach to adopting appropriate risk mitigation measures • STEP 3:  Determine appropriate level of human oversight - CONDUCT Risk Assessment and Human Oversight • After assessing the risk:  Adopt the risk-based approach to human oversight- • Human-in-the-loop  Human actors retain control in the decision- making process to prevent any errors by Al • Human-in-command  Human actors oversee the operation of Al and intervene whenever necessary • Human-out-of-the-loop  Al makes decision without human intervention - EXECUTE Development of Al Models and Management of Al Systems • STAGE 1: Data Preparation  Ensure compliance with the requirements under the relevant data protection law (e.g. principles of data minimisation, accuracy) • STAGE 2: Model Development  Select appropriate machine learning algorithms  Test Al models before use • STAGE 3: Ongoing Monitoring and Management  Conduct human oversight  Conduct evaluation and adjustments of Al strategy - FOSTER Communication and Engagement with Stakeholders • Being transparent with stakeholders • Channels for engagement with stakeholders • Comprehensible communications using layman's language • Communication with Stakeholders • Engagement with Stakeholders • Disclose the Use of the Al System • Allow Opt-out and Correction of Inaccuracies • Provide Adequate Information • Disclose the Risks • Provide an Option of Human Intervention • Provide Explanation upon Request Way Forward - Issuing guidance and best practices • a) HKMA: High-level Principles on Artificial Intelligence (2019) • b) OGCIO: Ethical Artificial Intelligence Framework (2021) - Raising awareness locally - Contributing to discussions locally and internationally • a) Public webinars/seminars on Al • b) Roundtable events and research initiatives • c) Collaborating with overseas counterparts, • - e.g. Al Working Group of Global Privacy Assembly, PDPC Singapore Legislation?