As AI adoption increases, so does the need for ethics, or ethical AI. Ethical AI refers to a set of morals and ideas that guide the development and accountability of AI technology.
According to a 2020 Capgemini survey, 49% of participants said they trust AI systems, up from 30% in 2018. AI systems must be transparent and advance social values, as unethical practices can cause significant threats to society.
As AI continues to influence decision-making in healthcare, media, and manufacturing, and others, the need for ethical AI increases. IBM Watson detects treatment for cancer patients. BMW uses automated image recognition for quality checks, inspections and elimination of pseudo-defects. Erroneous judgments in AI applications can harm high stakes decision making. Amazon’s AI recruiting tool showed biased reviews of women, providing them with few points.
Some organizations are beginning to create and adopt AI codes of ethics. Google released their ethical AI principles in 2018, stating their technology is socially beneficial, accountable and adheres to ethical standards.
Despite efforts, there is work to do. Ask Delphi, a newly launched AI text system created to deliver ethical judgment, boasted a unique approach to ending the ethical dilemmas of biased decisions, inequality, racism, etc. Unfortunately, Delphi’s suggestions don’t project an ethical outcome and came under fire recently for conveying racist and biased statements, reflecting users’ biases disguised as machine objectivity.
Ethical governance should be at the core of the AI ecosystem. On April 21st 2021, the European Union (EU) introduced rules and actions for excellence and trust in AI. Based on the risk based approach, the EU has classified the AI system into four categories: unacceptable, high, limited, and minimal risk. Unacceptable AI is considered a clear threat to safety, livelihood, and rights. All remote biometric identification systems, such as critical infrastructure, employment management, law enforcement and so on, are considered high risk. The document defines an AI system with transparency obligations such as chatbots as limited risk. AI-enabled video games are considered minimal risk.
The Biden Administration’s Bill of Rights for AI is a step in the right direction. In a recent op-ed, White House officials said the bill focuses on reducing bias in AI systems, empowering individuals’ to govern their data and creating awareness about the uses of AI technologies and ethical AI concerns.
New York City introduced its first AI strategy last October. The 116-page AI Strategy focuses on using AI to serve New Yorkers better, building AI know-how in government, modernizing data infrastructure and AI policy, developing partnerships and promoting equitable access to opportunities.
Most recently,193 UNESCO countries adopted the first-ever global agreement on AI ethics. The document defines the shared principles which will guide construction of the necessary legal infrastructure to ensure ethical AI deployment. The 28 page document states AI technologies could be a great service to humanity and countries can benefit from them while averting biases, privacy threats and mass surveillance.
Companies, Fusemachines included, have a role to play in establishing ethical AI. Using educational tools and curricula that focus on ethical AI or providing bias training are two important steps. From data sourcing to model creation, we can establish checks and balances for internal auditing. An organization should create proper guidelines for team members designing AI systems and incorporate audits to maintain transparency and accountability.
Embedding ethics into AI-assisted professional disciplines such as manufacturing, healthcare, education, HR etc. will be a long journey. Nevertheless, with a synergistic approach from technology organizations, governments, and educational institutions, we can drive inclusive growth, sustainable development, and wellbeing, contributing to a fair innovative ecosystem.
To learn more about our AI programs for enterprises and government, visit https://fusemachines.com/ai-education/.