Pioneers may view strict ethical and governance frameworks as barriers to innovation that prevent timely competitive advantage. At this year’s Dell Technologies World conference, one executive demonstrated how these efforts can be balanced.
John Rose, Dell’s Global Chief Technology Officer and Chief AI Officer, set the record straight on AI agents, explained the nuances the company has discovered in different types of AI agents, and proposed an ethics and governance framework that drives competitive advantage.
Enterprises rely on the ethics and governance of agent AI to ensure the security and stability of their organizations. Understanding how to balance structure and innovation can provide businesses with secure, streamlined workflows that put agents in charge.
What actually is an enterprise AI agent?
Rose began his talk by highlighting one of the simplest truths about agent AI: AI agents are not chatbots. Rose believes this is a common misconception that can undermine an organization’s strategic planning.
”[Agentic AI] “It’s a technology focused on digitizing operations,” he said at the conference session “Agent AI: Unleashing the Power of Purposeful Intelligence.” “It’s autonomous, it can reason, it can use tools, it has memory, it has expertise, and it can interact with other agents.” Does anything sound like a chatbot?”
Companies can only begin to understand the human and organizational dynamics of agentic AI, he continued, when they understand how it differs from productivity tools, which have rational, autonomous entities capable of performing tasks. Agentic AI can change many organizational norms. It can optimize workflows, change the roles of humans and agents within an organization, and require new security and ethical considerations.
Additionally, companies need to understand the nuances between agent types to achieve optimal results and measurable ROI. Roese shared the following framework for classifying the operational capabilities of AI agents.
- Low autonomy, simple tasks: A simple productivity tool.
- High autonomy, simple work: Able to complete simple tasks with little human intervention.
- Hygiene manager: An autonomous entity that can accomplish its goals and use many tools.
- Coordination agent: Complete complex tasks without much autonomy (for example, overseeing complex workflows).
“This gives us some guardrails to connect this technology with what it’s doing,” he said.
These guardrails force companies to be thoughtful about the agents they deploy, their scope, and their intent with the agents, allowing them to see results faster. Having the right agent for a specific task ensures effectiveness and helps you get results faster while saving time and money.
According to Rose, faster time to market and the knowledge to effectively operate AI tools will be key differentiators for companies going forward. Now that anyone with any level of technical experience can copy or generate software code using open source AI tools, the product no longer differentiates the organization.
“In the software world, the only sustainable source of differentiation from a product and technology perspective is execution speed,” he said. “The intellectual ability to know what to do is very important and is a great asset, but if it is not accompanied by the ability to be very nimble and quick to move from idea to concept, you are at a huge loss.”
But speeding without guardrails can lead to disaster. Almost paradoxically, it is thanks to standardized ethics and governance frameworks that companies like Dell and IBM are able to continue to innovate with the agility they need to remain competitive.
A protocol-based approach to agentic AI ethics
Mr Rose said the traditional way companies operate their trust and risk policies needs to change. The original model often involved many security and legal experts writing policies to protect businesses. But Dell believes a more standardized system could remove some cooks from the kitchen.
“If every AI project were a random adventure where you pick your own tools and do your own thing, there would be no structure. You would need a lot of lawyers and security people to make sure you don’t do stupid things,” he said. “But if you have a standardized platform in place, have a standardized operating model, have a good governance structure, and have trained people who know how to do it, you can mitigate some of those factors.”
Rose said Dell is rolling out new principles centered around trust. If trained developers are working on a trusted and approved platform, they will have “more freedom to move faster.”
IBM Fellow Kush Varshney agreed that standardization increases trust and reduces risk.
“Developing for predictability, standardization, and robustness in models and runtimes is developing for trust and risk mitigation,” he said in an email. “These efforts work together to actually accelerate the delivery of high-quality products and services to market.”
Roese shared important guidance on developing such deeply trusted agent AI systems:
- Establish clear rules about agent usage and agent boundaries.
- Issue digital IDs to all agents.
- Work in a safe, predictable, and controlled environment.
He said companies need to start thinking about probabilistic autonomous system protocols rather than policies. By using powerful protocols to ensure that agents can perform enough inference to guide the process, and by complementing them with tools that provide greater predictability, we can shift the focus from comprehensive high-order policies to a more standardized level of detail.
Organizations still require many layers of governance, “both intrinsic governance within the agents and external governance that wraps the agents in a guardrail infrastructure,” Varshney said. He said this governance needs to be systematic and administrative.
Varshney said another major ethical consideration that companies need to take into account is values.
“One of the fundamental assumptions in much of today’s AI development is that you can build a single, highly functional model and make it fit broadly defined human values. But in reality, values are multiple, context-dependent, and evolving,” he said.
He said developers shape these values through choices about data, purpose, and tradeoffs. Like Roese, he believes that one of the most important aspects of agent AI ethics is having trusted, educated individuals managing and working with the agents.
Everett Bishop is a deputy site editor in Informa TechTarget’s AI & Emerging Tech group, covering AI, quantum computing, and other emerging technologies.
