By David Kwan, AI Literacy Consultant

https://www.linkedin.com/posts/privili-online_davidkwan-ailiteracy-aistrategy-activity-7351073036407640064-VNHn?utm_source=share&utm_medium=member_desktop&rcm=ACoAAALtH10B8cC3dH8qM4PFD7-foX2IG7u3cEQ 

Republished from Ming Pao, Monday, July 14, 2025

Original URL: https://news.mingpao.com/pns/%E8%A7%80%E9%BB%9E/article/20250714/s00012/1752424112412 

 

Over the past year, AI has transformed in handling complex tasks. Previously, AI struggled with deep reasoning, but reinforcement learning within language models changed that.

AI’s Cognitive Leap

Traditional AI training used subjective human judgment, often favoring polished but flawed responses. Reinforcement learning with verifiable rewards—like passing code tests or solving math problems accurately—drives progress. In programming and math, AI surged from high-school to expert-level performance in a year, showing that with enough data and precise feedback, AI can reach professional proficiency.

White-Collar Automation: No Technical Barrier

Researchers see a path to automating tasks like Photoshop editing, travel planning, or tax filings. Hurdles are time and cost, not feasibility. Current frameworks suffice for large-scale automation; the bottleneck is high-quality data—screen recordings, mouse/keyboard sequences, decision paths, emails. With these, reinforcement and imitation learning can outperform humans. Training costs are lower than salaries, making automation inevitable, driven by data and economics.

The Curse of Moravec’s Paradox

AI struggles with long-context processing, cross-document tasks, or ambiguous goals needing trial-and-error. Clear feedback is key. Energy and computational limits are solvable engineering issues. Moravec’s Paradox persists: AI excels in cognitive tasks like chess or coding but lags in sensorimotor skills toddlers master. This may lead to humans as AI’s “hands and feet” via AR, performing physical tasks. This divide may last until robotics catches up, reshaping society.

Author: David Kwan, Founder of an AI company and university AI course instructor

Note: The opinions expressed aim to highlight flaws in systems, policies, or measures to encourage lawful improvement, with no intent to incite hatred, dissatisfaction, or hostility toward any government or community.

#DavidKwan #AIliteracy #AIstrategy #Privili #WhiteCollarAutomation #HandsandFeet