Page 36 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

o Some AI systems, like Carnegie Mellon’s cobot robots, exhibit limitations (e.g. inability
to press elevator buttons).
o To overcome such constraints, these systems ask humans for help.
o This symbiosis between humans and AI highlights the importance of collaboration,
where humans assist AI in completing tasks.

• Learning from human Interaction:
o Collaborative AI systems learn from human assistance. For example, a robot learns the
location of objects or tasks based on human input, improving over time.
o This concept applies to digital agents as well. For instance, a "Docubot" creates
PowerPoint slides based on user requests and learns from user feedback to refine its
outputs.
• Digital agents:

o Digital agents integrate perception, cognition, and action in a virtual context.
o They perform tasks such as reading documents, reasoning with data, and generating
actions (e.g. creating reports or sending emails).
o These agents ask for help when they encounter ambiguous instructions, learn from
human responses, and adapt to future tasks.

• Generative AI as a mediator:

o Generative AI plays a critical role in translating between human language and AI
systems’ internal logic.
o Generative AI enables seamless communication between humans and AI agents,
fostering collaboration and improving task execution.

• AI development philosophy:

o AI systems should acknowledge their limitations and ask for help when uncertain,
rather than attempting to solve everything autonomously and risking errors (e.g.
hallucinations in ChatGPT).
o Continuous learning and adaptation are crucial for building robust AI systems. AI can
improve by observing how humans solve problems and retaining this knowledge for
future use.

• The future of human-AI collaboration:
o The vision for the future includes a symbiotic ecosystem of humans and multiple AI
agents.
o AI agents will interact with each other (e.g. weather agents and social media agents)
and with humans to solve problems collaboratively.
o Humans will play a key role in guiding and teaching AI systems while leveraging their
capabilities to automate tasks, solve complex problems, and scale solutions.
• The ultimate goal is to build AI agents that can learn, adapt, and collaborate with humans,
creating a harmonious coexistence. This requires embracing the fact that AI does not know
everything upfront but improves through continuous interaction and feedback.
• AI and humans will increasingly coexist in a mutually beneficial relationship, with AI systems
learning from humans and humans leveraging AI’s capabilities to enhance productivity
and innovation.

24