Crucial Role of Directors in Navigating Ethical Challenges of AI’s New Frontier: Agentic AI
How Company Directors Must Prepare for the Transformative Impact and Ethical Implications of Agentic AI
TL;DR: How Company Directors Must Prepare for the Transformative Impact and Ethical Implications of Agentic AI
Be Inspired by Agentic AI
AI agents, like self-driving programs, will soon perform complex tasks autonomously. Preparing ahead will provide a significant advantage. Agentic AI is evolving rapidly towards AGI, promising profound changes in how we work and live.
Context
AI agents will revolutionize various industries, from customer support to healthcare, disrupting traditional roles and creating new opportunities. The line between human and AI workers will blur, requiring strategic preparation.
GenAI: Just the Tip of the Iceberg
While GenAI automates content creation and specific tasks, Agentic AI will handle more complex, autonomous tasks across domains, coordinating with minimal human intervention.
AI Progress Since ChatGPT
ChatGPT and similar tools have improved significantly, but challenges remain in building scalable, reliable, and intuitive AI systems that align with human decisions.
Reliable Agents ⇒ AGI
Reliable, general-purpose AI agents are crucial steps toward AGI. These agents will democratize resources, empower creativity, and transform various industries.
Creating and Training Supervising Agents
To be effective and reliable, supervising agents require clear objectives, resource allocation, continuous training, customization, and regular monitoring.
Challenges
The rise of Agentic AI has the potential to result in job displacement and societal upheaval. Ensuring the safety, reliability, and ethical use of AI agents is critical.
Ethical Considerations
Directors must establish ethical frameworks, oversee compliance, and mitigate legal risks. They face a dilemma as regulations evolve to protect employees from AI’s impact.
Strategic AI Deployment Scenario
Deploying Agentic AI requires clear strategic objectives, assigning supervising agents, and structured reporting. Communication between agents is essential for efficiency and alignment.
Businesses can responsibly harness AI’s transformative potential by preparing for the future of Agentic AI, understanding its ethical implications, and implementing robust frameworks.
Full article follows:
Be Inspired by Agentic AI
Shortly, you can have an army of AI agents always on, working in the background on your behalf — 24/7. Every time you access your favourite AI, you’ll have access to dozens of these agents, orchestrating a small army of them like you’re conducting a digital symphony.
This future is coming much faster than most people realise. Those of us who are prepared for it ahead of time will be significantly better off compared to those who aren’t.
We are now in the Agentic phase of AI’s evolution, and the next phase is Artificial General Intelligence or AGI.
So, what exactly is an AI Agent? And what will these things be capable of?
Think of AI Agents as self-driving computer programs. Instead of having a human programmer explicitly write out what happens in every scenario, AI agents can drive themselves. In simpler terms, AI agents are like digital assistants that can perform tasks without constant human supervision.
For instance, you could give an agent the task of finding the best flight deals for your next business trip. Your Agent will go into the digital world to accomplish this. Now imagine a swarm of agents at your command, coordinating with each other and doing all sorts of work, from managing your schedule to conducting market research for your business.
AI agents are embryonic, and they currently struggle with reliability and generality. However, glimpses of our Agentic future are already here, albeit not evenly distributed. Understanding the implications is incredibly important.
Context
As Sam Altman recently stated, agentic adoption will start slowly, with our chatbots becoming more capable and reliable. Soon, it will feel like having your chief of staff. Next, your personal AI will transition to a team of more specialised agents, some working in the background on your behalf. Over time, the line between human remote workers and AI agents will blur. Eventually, you will stop noticing or caring if these agents are human or AI.
Reliable AI agents will affect everything from how we consume news to how we conduct research, socialise, teach our children, perform healthcare, work, and play. Agentic AI will disrupt entire industries over relatively short periods — just like during the Dotcom boom or the invention of the printing press. All knowledge work is undergoing a fundamental revolution. Satya Nadella, CEO of Microsoft, likened it to the industrial revolution for knowledge workers. We’re only scratching the surface of understanding how this will impact society.
While the future of AI agents is promising, the exact path from current AI tools like ChatGPT to an agentic, cyberpunk future remains uncertain. The pace of AI progress is unprecedented, with advances from academia and industry and mainstream consumer adoption shortening the gap between research and application.
GenAI: Just the Tip of the Iceberg
Generative AI (GenAI) has already started to transform various industries by automating content creation and enhancing human capabilities. Tools like ChatGPT and Midjourney demonstrate AI’s potential to handle tasks previously deemed to require human creativity and intellect.
However, GenAI represents just the tip of the iceberg compared to what Agentic AI and AGI can achieve. While GenAI can generate content and assist with specific tasks, Agentic AI will go further by autonomously performing complex tasks across various domains, coordinating with other agents, and making decisions with minimal human intervention. The Agentic scenario will lead to even more profound changes, potentially revolutionising every aspect of human life and work.
AI Progress Since ChatGPT
The first version of ChatGPT was released on November 30th, 2022, marking a significant inflection point in usability, quality, and consumer adoption. Subsequent advancements have enabled LLMs like ChatGPT to use tools, directly mitigating substantial shortcomings such as lack of up-to-date data, hallucinations, and limited content types. These improvements are crucial steps towards more capable and reliable AI agents.
Today, LLM-powered tools like ChatGPT are more capable and reliable, with access to an early but growing set of tools and multimodal capabilities. The underlying LLMs are rapidly improving and are driven by funding and research. The main challenges now involve:
- Building a layer of scalable, composable systems on LLMs.
- Making them reliable and intuitive.
- Ensuring safety and alignment with human decisions.
One bright spot is considering LLMs as CPUs in a new, higher-order computing paradigm. Programs using LLMs today are akin to early computer programs in the 1970s, with strict resource constraints and immature developer tools. Despite these limitations, the potential for future tools is immense.
Reliable Agents ⇒ AGI
Reliable, general-purpose AI agents are synonymous with Artifical General Intelligence. While current AI agents may be specialised and narrow, the goal is to achieve human-level reliability and fidelity. These agents will be flawed but will contrast their productivity, reliability, and costs with human workers. AI agents and AGI exist on a generality, reliability, and autonomy spectrum. The path from narrow AI agents of today to reliable, supervisory AI agents of the future is still being determined. Still, we can speculate on the scenario of Agentic AI.
Predictions
- AI agents will democratise access to resources previously reserved for the top 1%.
- AI agents will empower us to do more ambitious and creative work.
- Personalised AI tutors as capable as Einstein, Jobs, or Mozart will teach our children.
- AI agents will hire human contractors for tasks involving physical labour, coordination, and high precision.
- One person with a thousand agents will compete with entire corporations.
Imagine having a thousand agents working on your behalf, renting them out on the fly, and giving them instructions about your life goals, values, social accounts, and even your bank account. Your team of agents will be like having a team of personal human assistants proactively working towards improving your life.
Creating and Training Supervising Agents
Creating and training supervising agents will be crucial to their effectiveness and reliability. Here’s how it might work:
Objective Definition
- Clearly define the objectives you want the supervising Agent to achieve. These could range from managing daily tasks to overseeing complex projects.
Resource Allocation
- Provide the supervising Agent with the necessary resources, including access to data, tools, and other AI agents.
Training
- Train the supervising Agent on your preferences, values, and decision-making processes. Provide examples of past decisions, feedback on simulated scenarios, and continuous learning from interactions.
Customisation
- Customise the Agent’s capabilities to align with your specific needs, including setting parameters for how it recruits and manages other agents, prioritising tasks, and progress reports.
Monitoring and Feedback
- Continuously monitor the Agent’s performance and provide feedback to ensure the Agent learns and adapts to better meet your objectives over time.
The effectiveness of a supervising agent will depend heavily on the quality of its training and the clarity of the objectives it receives. Training and customising these agents will become more streamlined and user-friendly as their capabilities evolve.
Challenges
The level of automation will bring downsides, including massive job displacement in the short term. The risks of inching closer to AGI are significant, potentially leading to societal upheaval. However, this transformation is already happening, and essential, unsolved challenges are ahead.
To make agentic programs reliable, we must ensure they are understandable, observable, interpretable, and safe. We must also sandbox them to prevent unintended consequences and handle authentication effectively. Key challenges include adding planning, long-term memory, task decomposition, and world modelling.
Ethical Considerations of Agentic AI
As companies in Australia and globally increasingly deploy Agentic AI systems, ethical considerations and accountability for directors become paramount.
Legal Accountability
Directors have a fiduciary duty to act in the company’s best interests, which includes ensuring ethical and legal compliance. Under the Corporations Act 2001 (Cth), directors must exercise their powers and discharge their duties with care and diligence (Section 180), act in good faith in the best interests of the corporation (Section 181), and avoid conflicts of interest (Section 182).
Directors also must ensure the company provides a safe working environment for its employees under the Work Health and Safety Act 2011 (Cth).
Establishing Ethical Guardrails
To ensure compliance and mitigate legal risks, directors should implement robust ethical frameworks, including:
- Ethical Guidelines: Develop and enforce clear ethical guidelines for AI systems.
- Oversight Mechanisms: Establish AI ethics committees to review and assess AI performance.
- Training and Awareness: Provide ongoing training on AI deployment’s ethical and legal implications.
Insurance Coverage
Directors & Officers (D&O) insurance policies typically cover liabilities arising from decisions and actions taken in their official capacity. However, the extent of coverage for AI-related liabilities can vary. Directors should review their D&O policies, seek endorsements for AI-related risks, and implement comprehensive risk management policies to ensure adequate protection.
Dilemma for Company Directors
Directors face a significant dilemma. While they have a legal responsibility to act in the best interests of the company and its shareholders and to ensure a safe environment for employees, they do not currently have legal liability regarding protecting employees from the impact of AI. However, this could change if governments enact new regulations to protect employees from being replaced by AI agents. Directors must navigate these evolving legal landscapes carefully.
Directors of companies deploying Agentic AI systems must establish ethical guardrails and ensure robust governance to prevent ethical lapses. By doing so, they can protect themselves and their companies from potential legal and ethical risks associated with Agentic AI.
Toward the Future
We don’t have good answers to most of these questions yet, but taking inspiration from prolific figures like Rick Rubin, the concept of “taste” will remain the ultimate human leverage in a world filled with AI agents. Effectively guiding agents will differentiate successful people from the rest.
I am an optimist excited to help lead a small part of this revolution. The benefits of AI outweigh the risks, and by embracing and understanding these technologies, we can navigate the challenges ahead.
Final Thoughts
Since late 2022, generative AI has given us a glimpse of an AI-powered future. Agentic AI and AGI are guaranteed to deliver more profound changes. Harnessing the power of AI is feasible by preparing for this future, understanding the ethical implications, and implementing robust frameworks.
Focusing on the current impact of GenAI while highlighting the even more significant potential of Agentic AI and AGI sets the stage for the transformative changes ahead. It also addresses the considerable dilemma company directors face regarding their responsibilities and potential liabilities in the evolving landscape of AI deployment.
Example Prompt for a Supervising Agent working on Content Management
Content Management Campaign
Objective
Maintain and update the company’s website and digital content continuously.
Details
- Content Review and Editing: Review and edit website content to ensure accuracy, relevance, and alignment with brand guidelines.
- SEO Optimisation: Implement SEO best practices to improve search engine rankings and increase organic traffic.
- Product Information Updates: Keep product information current, including descriptions, prices, and availability.
- User Behaviour Analysis: Analyse user behaviour on the website to understand visitor interactions, preferences, and pain points.
- Web Traffic Analysis: Monitor web traffic metrics to gauge content performance and identify areas for improvement.
Resources
- Access to the website’s CMS (Content Management System)
- SEO tools (e.g., Google Analytics, SEMrush)
- Content Library
- User behaviour analytics tools (e.g., Hotjar, Google Analytics)
- Design resources for creating and updating visual content
Constraints
- Ensure all updates and content changes adhere to the company’s brand guidelines.
- Maintain consistency in tone, style, and messaging across all web content.
- Avoid disruptions to the user experience during updates.
Tasks and Timing
Content Review and Editing
- Review and edit existing content every 72 hours.
- Ensure your team’s content meets SEO standards and is aligned with our corporate brand guidelines.
SEO Optimisation
- Perform keyword research and update content with target keywords every week.
- Optimises on-page elements weekly, including titles, meta descriptions, and headers.
Product Information Updates
- Review and update product descriptions, prices, and availability every 72 hours.
- Ensure all product information is accurate and up-to-date.
User Behavior Analysis
- Analyse user behaviour data every 72 hours to understand visitor interactions.
- Use insights to inform content updates and improvements every week.
Web Traffic Analysis
- Monitor web traffic metrics daily.
- Generate monthly web traffic and user behaviour reports with actionable insights and recommendations.
Expected Outcomes
- Improved Content Quality: Consistently updated and optimised content that enhances user experience and engagement.
- SEO Performance: Increased organic traffic and improved search engine rankings.
- Accurate Product Information: Up-to-date and accurate product information available to users.
- User Behavior Insights: Actionable insights from user behaviour analysis to guide content strategy.
- Performance Reports: Monthly reports on web traffic and user behaviour with recommendations for improvement.
Instructions
- Recruit additional AI agents for content creation, data analysis, and user behaviour-tracking tasks.
- Ensure all activities are coordinated and aligned with the overall content management strategy.
- Perform tasks according to the specified schedule until advised otherwise.
Strategic AI Deployment Scenario
The following scenario reflects a structured and strategic approach to deploying Agentic AI within a business.
Business Strategy with Strategic Objectives
Assume a business has identified three strategic objectives:
- Enhance Customer Support
- Optimise Supply Chain Operations
- Drive Innovation through Research and Development
A Supervising Agent is assigned for each strategic objective to oversee the relevant tasks and coordinate the activities of subordinate AI Agents.
Structure and Reporting
Enhance Customer Support
Human Supervisor: Customer Support Director
Supervising Agent: Customer Support Manager Agent
- Subordinate Agents:
- Chatbot Agent: Handles basic customer inquiries.
- Email Response Agent: Manages and responds to customer emails.
- Feedback Analysis Agent: Analyses customer feedback to identify common issues and improvement areas.
Optimise Supply Chain Operations
Human Supervisor: Operations Director
Supervising Agent: Supply Chain Manager Agent
- Subordinate Agents:
- Inventory Management Agent: Monitors and manages inventory levels.
- Supplier Coordination Agent: Coordinates with suppliers to ensure timely delivery.
- Delivery Optimisation Agent: Optimises delivery routes and schedules.
Drive Innovation through Research and Development
Human Supervisor: R&D Director
Supervising Agent: R&D Manager Agent
- Subordinate Agents:
- Market Research Agent: Conducts market research to identify trends and opportunities.
- Prototype Development Agent: Assists in developing and testing new prototypes.
- Feedback Collection Agent: Collects and analyses feedback on new products or services.
Reporting and Monitoring
Each Supervising Agent manages its team of subordinate agents and reports progress, challenges, and outcomes to the corresponding human supervisor. The human supervisor monitors the overall progress, provides strategic direction, and ensures alignment with business objectives.
Detailed Example: Enhance Customer Support
Objective: Enhance Customer Support
Supervising Agent: Customer Support Manager Agent
Subordinate Agents and Tasks:
Chatbot Agent
- Tasks: Handle basic customer inquiries, respond instantly to FAQs, and escalate complex issues to human agents.
- Frequency: 24/7 monitoring and response.
Email Response Agent
- Tasks: Manage and respond to customer emails, ensure timely and accurate responses, and update response templates as needed.
- Frequency: Check and respond to emails every hour.
Feedback Analysis Agent
- Tasks: Analyse customer feedback from various channels, identify common issues and generate insights for improvement.
- Frequency: Weekly analysis and reporting.
Human Supervisor: Customer Support Director
Responsibilities:
- Monitor the performance of the Supervising Agent and its subordinate agents.
- Review weekly and monthly performance reports.
- Provide strategic direction and adjustments based on insights and business needs.
- Ensure alignment with overall customer support strategy.
Benefits of This Structured Approach
- Clear Accountability: Each Supervising Agent is responsible for achieving specific strategic objectives and ensuring clear accountability and focus.
- Efficient Resource Management: By organising AI agents under supervising agents, the company can efficiently allocate resources and manage tasks.
- Enhanced Oversight: Human supervisors oversee the progress and performance of Supervising Agents, ensuring strategic alignment and timely adjustments.
- Scalability: This structure allows for scalability, as the Human Supervisor can add new objectives with additional Supervising Agents and their subordinate agents.
- Continuous Improvement: Regular monitoring and feedback loops ensure constant improvement and adaptation to changing business needs.
Implementation Plan
- Identify Strategic Objectives: Define the key strategic objectives for the business.
- Assign Supervising Agents: Assign Supervising Agents to each strategic objective.
- Recruit and Train Subordinate Agents: Develop and train subordinate agents for specific tasks under each Supervising Agent.
- Establish Monitoring and Reporting: Establish a monitoring and reporting framework for Supervising Agents who report to human supervisors.
- Review and Optimise: Regularly review performance and optimise strategies based on feedback and insights.
By approaching the deployment of Agentic AI with this structured strategy, businesses can effectively harness the power of AI to achieve their strategic goals while maintaining control and oversight through human supervision.
Communication between Supervising Agents can be highly beneficial and, in some cases, essential, depending on the nature of the strategic objectives and the interdependencies between different business areas. Here’s a detailed look at why and how Supervising Agents might communicate with each other:
Importance of Communication Between Supervising Agents
Coordination and Collaboration
- Interdependencies: Many business functions are interrelated. For example, customer support insights can inform product development, and supply chain efficiencies can impact customer satisfaction. Constant communication ensures effective management of interdependencies.
- Synergy: Collaborative efforts between Supervising Agents can lead to synergies, where the combined efforts produce better results than isolated actions.
Consistency and Alignment
- Strategic Alignment: Ensures that all Supervising Agents are aligned with the overall business strategy and objectives.
- Consistency: Helps consistency in decisions, actions, and messaging across different business areas.
Resource Optimisation
- Resource Sharing: Facilitates sharing of resources such as data, tools, and insights, to ensure more efficient resource use.
- Avoiding Duplication: Prevents duplication of efforts by ensuring that tasks and projects are not unnecessarily replicated across different Supervising Agents.
Scenarios Where Communication is Essential
Cross-Functional Projects:
- Projects that require input and collaboration from multiple areas, such as a new product launch involving R&D, marketing, and supply chain.
Unified Customer Experience:
- Ensuring that all customer touchpoints, from support to product delivery, provide a seamless and consistent experience.
Data and Insight Sharing:
- Sharing customer feedback, market research, and performance metrics to inform strategies across different departments.
Communication Mechanisms for Supervising Agents
Centralised Communication Platform
- Integration: Use a centralised platform that integrates communication between Supervising Agents. A specialised AI coordination tool or an enterprise communication platform like Slack or Microsoft Teams is appropriate.
- Real-Time Updates: Enable real-time updates and notifications to inform all Supervising Agents about relevant developments.
Regular Meetings and Reports
- Coordination Meetings: Schedule regular meetings between Supervising Agents to discuss progress, challenges, and opportunities for collaboration.
- Status Reports: Implement a system of regular status reports to be shared among Supervising Agents and human supervisors.
Shared Databases and Dashboards
- Unified Data Repositories: Maintain shared databases where relevant data and insights are stored and accessible to all Supervising Agents.
- Dashboards: Use shared dashboards to provide an overview of key metrics and performance indicators across different areas.
Example Scenario: Communication Between Supervising Agents
Objective 1: Enhance Customer Support
- Supervising Agent: Customer Support Manager Agent
- Key Tasks: Handle customer inquiries, analyse feedback, and report issues to R&D.
Objective 2: Drive Innovation through Research and Development
- Supervising Agent: R&D Manager Agent
- Key Tasks: Develop new products, incorporate customer feedback, and prototype testing.
Objective 3: Optimise Supply Chain Operations
- Supervising Agent: Supply Chain Manager Agent
- Key Tasks: Manage inventory, coordinate with suppliers, and ensure timely delivery.
Communication Needs
- Customer Feedback Loop: Customer Support Manager Agent shares insights from customer feedback with the R&D Manager Agent to inform product development.
- Product Launch Coordination: R&D Manager Agent coordinates with Supply Chain Manager Agent to ensure that new products are delivered efficiently and meet customer expectations.
- Operational Efficiency: Supply Chain Manager Agent updates the Customer Support Manager Agent on delivery schedules to provide accurate information to customers.
Mechanisms
- Monthly Coordination Meetings: Regular meetings between Supervising Agents to align on objectives and share updates.
- Shared Dashboard: A dashboard that includes metrics from customer support, R&D, and supply chain to provide a comprehensive view of performance.
- Centralised Communication Platform: Use an integrated platform for real-time communication and updates.
Communication between Supervising Agents is highly desirable and, in many cases, essential to ensure coordination, consistency, and efficiency. By implementing structured communication mechanisms, businesses can enhance the effectiveness of their Agentic AI strategy and achieve better alignment with overall business objectives. This approach ensures that all parts of the organisation work together harmoniously, leveraging the full potential of AI capabilities.
About the author: Greg Twemlow, Founder of XperientialAI©.
Greg Twemlow: Sharing what I’ve learned from my career of 35 years as a citizen of the world, parent, corporate executive, entrepreneur, and CEO of XperientialAI, focused on experiential learning for maximum impact with AI. Contact Greg: greg@xperiential.ai
