How is AI evolving?

Key Concepts

• Models Knowing Things: AI systems primarily focused on information retrieval, understanding, and generation.
• Models Doing Things: AI systems (agents) capable of performing actions, making decisions, and interacting with specific environments.
• Environments (for AI Agents): Complex systems or applications where AI agents operate and perform tasks.
• Agent Decision-Making: The process by which an AI agent chooses and executes actions within an environment to achieve a goal.
• Change Management Exercise: The process of guiding individuals, teams, and organizations through transitions, particularly those involving new technologies.
• Adoption Curve: The pattern of how new technologies or innovations are accepted and used by a population over time.
• Policy Issue: Challenges or decisions that require governmental or organizational rules, regulations, and frameworks.
• Technological Issue: Problems or limitations inherent in the technology itself, requiring further research and development.

Paradigm Shift in AI Capabilities: From Knowing to Doing

The fundamental trend in artificial intelligence is a significant shift from models that primarily "knowing things" to models that are capable of "doing things." This evolution moves beyond mere information processing to active task execution and decision-making. The immediate implication of this shift is the emergence of "environments" where these AI agents can operate.

The Role of Environments and Agent Decision-Making

As AI models transition to "doing things," they must interact with and navigate various "environments." These environments represent the specific contexts or systems where tasks are performed.

• Examples: Navigating a complex "healthcare system" to assist a user, or interacting with a "weather app on your phone" to provide personalized information or actions.
• Agent Function: The core purpose of these advanced models is for the AI "agent [to] make decisions for you" within these environments, effectively acting on behalf of the user.

Current Stage of Development and Speculation

The speaker emphasizes that this transformative phase is "just getting into the beginning of that." Due to its nascent stage, there is "wide variance" in speculation regarding the speed and trajectory of its development and adoption. Different perspectives exist on how quickly and effectively these capabilities will improve.

The Aggressive Trajectory: Policy Over Technology

One perspective, termed the "most aggressive trajectory," posits that training AI models to "do things" within environments will become "quite easy." If this holds true, the primary challenges will shift dramatically:

• Shift in Focus: The issue will no longer be a "technological issue" but rather a "change management exercise in the economy."
• Adoption Curve: The "adoption curve" for these technologies will then become predominantly a "human and policy issue," requiring societal and regulatory adaptation rather than further technological breakthroughs.

Future Outlook and Policy Implications

Despite the potential for rapid advancement, the speaker clarifies that "We're not there from the technology standpoint" currently. However, a specific prediction is made:

• Timeline: "In the next two to three years," the technology is projected to reach a point where it will "push the change management and policy makers to say like what do we do with this because it's getting pretty close."
• Implication: This suggests that within this timeframe, technological capabilities will advance sufficiently to necessitate serious consideration and action from policymakers regarding the integration and governance of these "doing things" AI agents.

Synthesis and Conclusion

The core takeaway is the imminent and profound shift in AI from passive knowledge to active agency within complex environments. This transition, currently in its early stages, is expected to mature rapidly within the next two to three years. While technological hurdles remain, the speaker anticipates that future challenges will increasingly revolve around human adaptation, societal change management, and the development of appropriate policies, rather than purely technological limitations. The advancement of AI to "do things" will compel policymakers to address its implications proactively.