Adedeji Awolesi

The adoption and advancement of Artificial Intelligence(AI) in today’s world is transforming industries, reshaping economies and influencing daily lives. From technology to education to healthcare and entertainment, AI tools is driving efficiency and innovation. Artificial Intelligence today is generally acceptable with more models and logic released. This progress has put much pressure on AI. Competitors are adopting it faster than you think. Consumers want their needs addressed and governments are adding public mandates to quickly adopt AI. I believe that AI tools must be backed with a strong ethical framework. This ensures that it benefits everyone equitably and do not cause harm.

In this blog post, I will be exploring the ethical consideration for AI tools, practical examples with key insights and how best we can implement responsible AI. This post is for the Azure Spring Clean 2025.

Fairness

AI tools should treat people fairly regardless of their gender, ethnicity or inequality. AI systems learn from historical data. If the data is biased, the AI system will amplify those biases. This amplification can lead to unfair treatment of people. For example, In 2018, BBC reported that Amazon scrapped an AI recruiting tool because it was biased against women. The AI system was trained on resumes submitted over a ten year period, most of which came from men. As a result, the AI penalized resumes that included words like women or referenced all girls colleges.

Best Practices

• Use of diverse and representative dataset to train AI models.
• Regularly test AI systems for discriminatory outcomes and bias audit.
• Implement fairness aware machine learning techniques to mitigate bias by AI ethics team.
• Ensure diverse teams are involved in the development process to find blind spots.

Reliability and Safety

Without reliable and safe AI tools, the potential for harm becomes significant. AI systems should be consistent and accurate without causing harm even in unexpected scenarios. This is critical for building trust in AI technologies. For example, According to New York Times, Uber autonomous vehicle struck and killed a pedestrian in Arizona. The in built AI system failed to correctly identify the pedestrian as a hazard which lead to public uproar. AI systems must be tested in diverse environments to ensure they can handle unexpected situations.

Best Practices

• Testing extensively in diverse and realistic scenarios to identify and solve potential failures.
• Train models on dataset that include edge cases and anomalies.
• Design systems that require human approval for critical decisions.
• Design systems to default to a safe state in case of uncertainty or failure.
• Conduct risk assessments to identify potential safety and reliability issues.

Privacy and Security

AI tools should be secure and respect privacy. AI systems rely on large volumes of data. These often include sensitive personal information, which raises concerns about privacy breaches and misuse. The Cambridge Analytica scandal is a prime example of AI driven data misuse. The company used personal data from millions of Facebook users without consent using AI powered analytics to influence elections. According to BBC, Meta settled the scandal case for $735m and paid $5b fine to Federal Trade Commission probe into its privacy practices.

Best Practices

• Use privacy preserving technique like federated learning and differential privacy.
• Use encryption and anonymization techniques to protect sensitive data.
• Seek informed consent from users before collecting or processing their data.
• Adhere to data protection compliance regulations like GDPR and CCPA.
• Adopt privacy first AI design principles ensuring minimal data collection.

Inclusiveness

AI tools need to be designed with the needs of diverse populations in mind. AI systems should empower everyone and engage people. AI should bring benefits to all parts of the society regardless of physical ability, gender, sexual orientation and ethnicity. For example, voice recognition AI systems have struggled to understand accents. They also struggle with dialects. This disadvantages non native speakers and certain ethnic groups.

Best Practices

• Involve diverse stakeholders in the development process of design.
• Ensure AI tools has accessibility features that are usable by people with disabilities.
• Make AI solutions accessible and affordable to underrepresented communities.
• Partner with local organizations to tailor AI solutions to specific regional need.

Transparency

AI tools should be understandable. Many AI systems especially deep learning models, are complex and opaque. Even the developers find it difficult to understand how decisions are made. People should be aware of the purpose of the system, how it works and what limitations to be expected. This lack of transparency can erase trust and make it difficult to hold systems accountable.

For example, In healthcare, AI tools are used to diagnose diseases and recommend treatments. Yet, if a doctor can not explain why an AI system recommended a specific treatment, this can lead to mistrust. This mistrust can also harm the patient. According to BBC in 2020, Apple’s credit card received criticism. An AI based algorithm offering lower credit limits to women compared to men with similar financial backgrounds. The decision making process lacked transparency. This made it difficult for affected individuals to contest the decisions. They did not understand why they received lower limit.

Best Practices

• Use Explainable AI (XAI) tools that offer clear explanations for AI decisions.
• Design user friendly systems that communicate results in an understandable way.
• Publish documentation of AI models including their functions, limitations and uncertainties.
• Create mechanisms for users to question or contest AI decision.

Accountability

People should be accountable for AI systems. When AI systems make mistakes or cause harm, it can be difficult to determine who is accountable. Is it the developer, the organization or the user.

In the Uber autonomous vehicle incident example I used earlier, the investigation revealed important findings. The AI system had detected the pedestrian but it failed to react appropriately. The incident raised questions about who was responsible – the AI system, the developers or Uber itself. This is why designers and developers of AI-based solutions should work within a framework of governance. These organizational principles ensures that the solution meets clearly defined ethical and legal standards.

Best Practices

• Establish clear AI governance frameworks outlining liability and responsibility.
• Define processes for addressing harm caused by AI systems.
• Create robust testing and validation processes to minimize error.
• Adopt industry wide standards for accountability.

Benefits

• It builds trust with users and stakeholders.
• It drives inclusive and fair outcomes.
• Allows businesses to stay ahead of regulatory requirements.
• It helps to manage risks – be it ethical risks, reputational risks, legal liabilities or regulations.
• It helps to make better AI powered decision.

Conclusion

The journey toward ethical AI is not without challenges. As technology professionals, we have a responsibility to ensure that AI tools are developed responsibly. Organizations must proactively address issues related to bias, transparency, privacy and accountability. By adopting ethical AI frameworks, businesses can create AI systems that are fair, safe and inclusive. Conducting regular audits helps to maintain these standards and staying compliant with regulations ensures ongoing fairness and safety.

For AI to truly be “for all,” it must be developed, deployed and used in ways that focus on ethical considerations. As I always say “Inasmuch that AI has lots of potential, it is our duty to make sure that it does good“