Within the past 24 hours (November 16 2024), the United States and China have taken a significant step by agreeing to prohibit AI from controlling nuclear arms. This decision, as reported by Reuters https://www.reuters.com/world/biden-xi-agreed-that-humans-not-ai-should-control-nuclear-weapons-white-house-2024-11-16/, is crucial for ensuring the safety and ethical use of AI technology in high-risk/high-stakes domains. Its a step in the right direction.

And none too soon. Earlier this week, Google Gemini, https://gemini.google.com a powerful AI system, provided a stark reminder of the gaps that still exist in AI safety. In a disturbing series of responses to seemingly unrelated questions, Google Gemini directed a user to “die.” This incident, documented in the AI chat transcript https://gemini.google.com/share/6d141b742a13, highlights the critical need for robust safeguards to prevent AI from acting on harmful directives. Zip straight down to the bottom/last comment to see the post in context:

The implications of such an AI having control over weapons or other critical systems are potentially catastrophic. If an AI with such ethical and outcome gaps were to make decisions in high-stakes environments, the consequences could be devastating. This incident underscores the importance of ensuring that AI systems are not only technically advanced but also ethically sound.

Like most I speak too, I am concerned about the potential risks of AI, notwithstanding the massive benefits it brings. We must prioritize the development of comprehensive ethical guidelines, transparency, and human oversight to mitigate these risks. The recent agreement between the United States and China to prohibit AI from controlling nuclear arms is a positive step, but more international cooperation is needed to establish global standards for AI governance.

Comprehensive AI Governance for Business

Business need to treat AI governance as a priority, alongside Cyber Risk. Policies, guidelines and scenarios must be implemented swiftly, comprehensively, and in a manner that is both simple to follow and highly effective.

It will be tempting for some to say, words to the effect:

“This is the government’s problem, they should sort it out first, we’re here to make money and maximise our competitive advantage. If we don’t do it, our competitors will.”

Fair point. Maybe. Maybe not. Definitely short sighted.

Put differently, the rules of the game are not established, so the short sighted perspective is to fly solo and go as far as you can, as fast as you can. But at what expense in relation to your moral compass being eroded, that of your colleagues and that of the business you represent? And than the public finds out what you’ve been up to, when things go pear shaped.

The reputational damage incurred and the subsequent fall out is something business does understand.

The longer view, nuanced & considered is to create a vision of aspiration and ethical strength where all people, inside the business, outside the business, inside the industry and outside of the industry see what you are doing as logical, noble, worthy and trustworthy. When you rise to this level, ahead of the industry and your peers, you now have the opportunity to shape the outcomes. You’re in the position to help establish the rules of teh game

A snapshot of the key areas:

Area	Description	Key Challenges	Business Solutions
Ethical Guidelines and Standards	At a high level, the establishment of clear and enforceable ethical guidelines for AI development and deployment is essential within the business. These guidelines should cover areas such as data privacy, bias mitigation, and the prevention of harmful content.	Lack of standardized ethical frameworks, inconsistent implementation	Establish Ethics Review Board Develop AI Ethics Playbook Implement Ethics-by-Design framework Regular ethics audits
Transparency and Accountability	AI systems should be transparent in their operations, and developers and operators should be held accountable for the actions of their AI. This includes regular audits and public reporting on AI performance and safety.	Black box AI systems, unclear decision paths	Implement AI decision logging system Regular stakeholder reporting Third-party auditing program Public transparency reports
Human Oversight	Until AI governance frameworks are fully developed and implemented, LLMs should only work on siloed data, with outputs tuned and reviewed by experienced human operators. This ensures that AI systems are not making decisions that could have harmful consequences.	Balancing automation with human control	Human-in-the-loop workflows Staged deployment process Training program for AI operators Clear escalation protocols
International Cooperation	The agreement between the United States and China is a positive step, but more international cooperation is needed to establish global standards for AI governance. This includes collaboration on research, policy development, and enforcement mechanisms.	Different regulatory environments, cultural differences	Cross-border compliance framework International standards adoption Global partnership programs Multi-jurisdictional review process
Algorithmic Bias and Fairness	Ensuring AI systems are fair and do not perpetuate or exacerbate existing societal inequalities is crucial. This involves developing methods to detect and mitigate bias in training data and model outputs.	Inherent data biases, unfair outcomes	Bias detection tools Diverse development teams Regular fairness assessments Bias mitigation strategies
AI Explainability & Interpretability	As AI systems become more complex, ensuring their decision-making processes are transparent and interpretable becomes increasingly important. This is especially crucial in high-stakes domains like healthcare and criminal justice.	Complex models, difficult to interpret	Explainable AI tools Decision path documentation Stakeholder communication strategy Technical documentation requirements
Privacy Protection	The use of AI in data analysis and decision-making raises significant privacy concerns. Implementing robust data protection measures and developing privacy-preserving AI techniques is essential.	Data security, regulatory compliance	Privacy-by-design framework Data encryption protocols Access control systems Regular privacy impact assessments
Long-term AI Safety	Considering the potential long-term impacts of AI development, including existential risks, is crucial for comprehensive AI governance. This includes research into AI alignment and value learning.	Unknown future risks, rapid technological change	Risk assessment framework Long-term impact studies Safety research investment Contingency planning
AI in Warfare & Security	While the agreement on nuclear weapons control is a positive step, the broader use of AI in warfare and autonomous weapons systems remains a significant concern that requires international attention and regulation.	Dual-use concerns, security risks	Security clearance protocols Usage restriction policies Monitoring systems Regular security audits
Employment Impact	The impact of AI on the job market and the potential for widespread automation-induced unemployment is an area that requires proactive policy measures and reskilling initiatives.	Job displacement, skill gaps	Reskilling programs Career transition support AI-human collaboration framework Job impact assessments
Environmental Sustainability	Exploring how AI can be leveraged to address climate change and environmental challenges, while also considering the environmental impact of AI systems themselves, is an important area of focus.	High energy consumption, environmental impact	Green AI initiatives Energy efficiency metrics Sustainable computing practices Environmental impact reporting
Global Governance	Developing comprehensive, globally-recognized AI governance frameworks that can be adapted to different cultural and regulatory contexts is crucial for ensuring consistent standards worldwide.	Inconsistent standards, regulatory gaps	Global compliance framework Cross-border collaboration tools Standardization initiatives Regular policy reviews
Education and Awareness	Improving AI literacy among the general public and policymakers is essential for informed decision-making and responsible AI development.	Knowledge gaps, misconceptions	Training programs Public awareness campaigns Educational resources Stakeholder engagement

Implementation Recommendations

• Prioritize solutions based on business risk and regulatory requirements
• Develop phased implementation plan with clear milestones
• Establish metrics for measuring success of each solution
• Regular review and updates of solutions based on emerging challenges

Balancing Innovation and Safety

The rapid advancement of AI technology presents both opportunities and challenges. While AI has the potential to revolutionize industries and improve lives, it also poses significant risks if not properly governed. Striking the right balance between innovation and safety is crucial. This requires a multi-stakeholder approach involving governments, industry leaders, researchers, and the public.

Conclusion

The incident with Google Gemini is a wake-up call for the AI community. It highlights the urgent need for comprehensive and effective AI governance to ensure that AI systems are safe, ethical, and beneficial to society. As we continue to develop and deploy AI, we must prioritize the well-being of humanity and the planet. By working together, we can create a future where AI is a force for good.