The Paris Olympics 2024 isn't just making headlines as a global sporting event—it's also pioneering how AI is being used to enhance security and improve the overall experience for athletes and spectators alike. There is a catch though: While using AI-powered mass surveillance does promise to be a security boon, it also raises concerns about potential privacy threats and invasions. 

The Olympics as well as the US presidential elections’ campaign have also raised concerns about sports people and politicians’ images and videos being doctored by AI for entertainment or for memes. This brings us to the important question: what is and what is not allowed in AI content and who should regulate it?

According to KPMG's 2023 global study, 85% of respondents recognize the benefits of AI, yet 61% remain wary about placing their trust in these systems. Even more telling is the fact that 67% report only low to moderate acceptance of AI.

Challenges of Implementing Ethical AI in Content

Deepfakes and Misinformation

“Deepfakes” are AI-generated videos adept at manipulating visual and auditory components to construct deceptively authentic footage. The paper "Deepfakes, Misinformation, and Disinformation in the Era of Frontier AI, Generative AI, and Large AI Models” by Shoaib et al. explores the risks to privacy and personal security, the challenges posed to journalism, and the loss of public trust. As these technologies continue to become more sophisticated, the legal and ethical challenges posed by them have exposed the gap in the current regulations.

Algorithmic Bias

Algorithmic bias in AI generated content occurs when AI systems reflect biases present in their training data or design. This can lead to unfair or discriminatory responses, which can have severe consequences especially in high-stake areas like law enforcement, healthcare and recruitment.

Data Privacy and Security in AI

Training the machine learning algorithms requires vast amounts of data which can include sensitive information. This information could inadvertently be served as content, thus violating data privacy and security.  It is imperative to ensure that the data is handled securely and privacy is not compromised to maintain public trust. 

Accountability and Transparency 

Ensuring accountability in AI requires clear mechanisms for tracing AI decisions, disclosing data sources of the content, and holding developers responsible. Transparent practices help stakeholders understand AI behavior, fostering trust and accountability.

Copyright Infringement and GenAI

Generative AI has sparked both excitement and apprehension for its creative and powerful impact on content creation. With several high profile lawsuits already underway, it has prompted deeper scrutiny in relation to copyright protection for both “data used for training” and AI generated content.

AI Guardrails and Ethical Frameworks

An AI guardrail is a safety mechanism intended to prevent artificial intelligence (AI) from causing harm. Just as highway guardrails are installed to protect drivers and promote safe travel, AI guardrails are established to ensure that AI systems function safely and produce positive and non-hostile content. 

Implementing Guardrails in Practice

Practical implementation of ethical AI requires more than just guidelines; it necessitates actionable steps. Companies and developers can use open source tools like NVIDIA NeMo-Guardrails and Guardrails AI to integrate ethical guardrails into their AI applications. These tools provide essential frameworks for monitoring and controlling AI outputs, ensuring that content is deployed responsibly and ethically.

Responsible AI Initiatives by Major Players

The primary responsibility of implementing ethical AI frameworks falls on the shoulders of those who develop them. Here, we briefly explore the responsible AI initiatives by the two tech stalwarts: Google and Meta. 

Google 

Google’s 2024 AI responsibility Update outlines four phases in their responsible AI roadmap: Research, Design, Govern and Share, known as the “AI responsibility lifecycle”. 

They emphasize the importance of transparency and collaboration by sharing knowledge through research publications, model cards, and partnerships with external organizations and governments. 

Google's Secure AI Framework (SAIF) is a critical component of their rigorous risk assessments and security measures. This includes addressing challenges such as adversarial testing and prompt injection attacks.

Google's AI principles guide the development of policies, including prohibited uses for generative AI and requirements for disclosing synthetic content in various contexts.

Meta's LLaMA 3.1 

Meta's approach to responsible AI is pioneering, particularly as it opens Llama 3.1 to the public. The open-source nature of this tool requires rigorous ethical considerations to prevent misuse. 

The company has introduced several safety and security tools, such as Llama Guard 3 for content moderation and Prompt Guard for preventing prompt injections, to mitigate risks associated with open-source AI. 

Additionally, Meta collaborates with global organizations to establish AI safety standards and conducts rigorous pre-deployment assessments, red teaming, and fine-tuning to identify potential risks.

The Role of Regulation and Global Standards

On a global scale, there's a growing momentum for collaboration on AI governance. A recently released report by EY assessed the AI regulatory landscape across eight jurisdictions: Canada, China, the European Union (EU), Japan, Korea, Singapore, the United Kingdom (UK), and the United States (US). The report identified common regulatory trends across these regions, all aimed at achieving a shared goal: minimizing the risks associated with AI while also enhancing its potential to deliver social and economic benefits for their citizens.

The UNESCO Recommendations on the Ethics of AI provide a universal framework that their 193 member nations can follow, ensuring that AI serves humanity and aligns with global ethical standards. It outlines Ten Core Principles with a human-rights centered approach. The principles aim to establish ethical global standards, particularly addressing the challenges and considerations related to AI-generated content.

The EU AI Act is a risk based regulation legislation. It categorizes AI systems based on their risk level. The most stringent compliance obligations would be applicable to AI systems classified as “high risk” while general purpose AI models (like foundation model and Gen AI) would pose “systemic risk”. The Act is a first of a kind which is legally binding across the EU.  It emphasizes the need for AI systems to respect intellectual property rights, particularly in content creation and distribution.

The United States has adopted a more voluntary, guideline-based approach, focusing on risk management frameworks and ethical guidelines without enacting specific legislation​.

Corporate Responsibility

It is advisable for companies to take proactive steps to self-regulate, ensuring their AI systems adhere to ethical standards. This involves regular audits, transparency reports, and engaging with stakeholders to address concerns and improve trust in AI technologies. 

The future of AI hinges on trust, and the stakes couldn't be higher. With 97% of respondents in the KPMG study backing trustworthy AI principles, and three-quarters ready to embrace AI if robust safeguards are in place, it is clear that ethical AI isn't just an option; it's a necessity. This rings true especially in content creation, where ethical AI ensures the responsible use of sensitive information and respects copyright laws. The adoption of ethical AI practices is not a solitary endeavor, it requires collaboration and support from all the stakeholders including governments, organizations, developers, and the public. Implementing AI guardrails, fostering transparency, and adhering to global standards are imperatives for building a future where AI serves humanity responsibly. 

References

International Olympic Committee. (2023). AI and tech innovations at Paris 2024: A game-changer in sport. Retrieved from https://olympics.com/ioc/news/ai-and-tech-innovations-at-paris-2024-a-game-changer-in-sport

Becker, B. (2024). AI mass surveillance at Paris Olympics: A legal scholar on the security boon and privacy nightmare. The Conversation. Retrieved from https://theconversation.com/ai-mass-surveillance-at-paris-olympics-a-legal-scholar-on-the-security-boon-and-privacy-nightmare-233321

Meta. (2024). Meta LLaMA 3.1: AI responsibility. Retrieved  fromhttps://ai.meta.com/blog/meta-llama-3-1-ai-responsibility/

UNESCO. (2023). Recommendation on the ethics of artificial intelligence. Retrieved from https://www.unesco.org/en/artificial-intelligence/recommendation-ethics

Google. (2023). Our responsible approach to building guardrails for generative AI. Retrieved from https://blog.google/technology/ai/our-responsible-approach-to-building-guardrails-for-generative-ai/

KPMG. (2023). Trust in AI: Global study 2023. Retrieved from https://assets.kpmg.com/content/dam/kpmg/xx/pdf/2023/09/trust-in-ai-global-study-2023.pdf

Shoaib, M. R., Wang, Z., Ahvanooey, M. T., & Zhao, J. (2023). Deepfakes, Misinformation, and Disinformation in the Era of Frontier AI, Generative AI, and Large AI Models. Retrieved from https://arxiv.org/pdf/2311.17394

Google. (2024). AI responsibility 2024 update. Google. https://ai.google/static/documents/ai-responsibility-2024-update.pdf

Towards Data Science. (2024). Safeguarding LLMs with guardrails. Towards Data Science. https://towardsdatascience.com/safeguarding-llms-with-guardrails-4f5d9f57cff2

EY. (2024). The artificial intelligence global regulatory landscape: 2024 insights and trends. Retrieved from https://www.ey.com/content/dam/ey-unified-site/ey-com/en-gl/insights/ai/documents/ey-gl-the-artificial-intelligence-global-regulatory-07-2024.pdf

Techopedia. (n.d.). AI guardrail. Retrieved from https://www.techopedia.com/definition/ai-guardrail

‍