The productivity paradox of AI

Labor productivity has seen a significant decline over the past decade, even as artificial intelligence has helped replace repetitive tasks in daily work.

Does AI really increase work productivity? Photo: LinkedIn .

Amid rising concerns about job losses due to AI, optimists argue that it's simply a tool to increase productivity, benefiting both workers and the economy . Microsoft CEO Satya Nadella suggests that users only need to state their goals, while automated AI agents will plan, execute, and learn on their own across all systems.

However, AI is creating a "productivity trap," encouraging more and more people to use, and even become dependent on, it. This will lead to a decline in self-reflection and problem-solving abilities, and more seriously, will affect creativity and breakthroughs in life.

Prioritizing quantity over quality.

According to the FT, AI tools would be ideal when performance alone is enough to solve the productivity problem. The newspaper points out that in the past half-century, many computers have been developed that are supposedly faster than ever before, but the rate of labor productivity growth in developed economies has slowed, from about 2% per year in the 1990s to only about 0.8% now.

With the advent of computers, the internet, and global talent connectivity, breakthroughs should have exploded. Yet, research productivity has declined. A scientist today generates fewer groundbreaking ideas per dollar invested than their predecessor in the 1960s.

Economist Gary Becker once pointed out that parents face a choice between “quality and quantity.” For example, the more children they have, the less likely they are to invest in each child individually. The same may be happening with innovation.

Too many projects at once can negatively impact creativity. Photo: Adobe Stock.

Large-scale studies of patent output confirm that the number of projects undertaken is inversely proportional to the likelihood of breakthroughs. In recent decades, scientific papers and patents have increasingly become incremental additions rather than major breakthroughs.

Meanwhile, great minds throughout history understood this. Isaac Newton once said that he always “keeps a problem before him… until the first sparks of light appear, little by little, and finally burst forth into a clear and complete light.” “Creativity is saying no to a thousand things,” Steve Jobs agreed.

"The AI's Medium-Capability Trap"

Mr. Ho Quoc Tuan, Director of the Master's Program in Finance & Accounting at the University of Bristol, mentioned the concept of the "AI's average ability trap." Jobs that frequently require the abilities of an average person often involve many repetitive tasks, follow clear and quantifiable processes. However, he argues that this is precisely AI's outstanding strength.

Large-scale language models (LLMs) tend to cling to what statistics consider to be general consensus. If you had a chatbot read 19th-century text, it would "prove" that humans couldn't fly, until the Wright brothers did.

A review published in Nature in March 2025 showed that, while LLM can help reduce repetitive scientific tasks, the real leaps in thinking still belong to humans. Mr. Tuan also argued that clinging to what is already known, being reluctant to take risks, and lacking critical thinking are fatal weaknesses in the AI ​​era.

Demis Hassabis, head of the team at Google DeepMind that developed AlphaFold, a model capable of predicting protein shapes, is considered one of the most outstanding scientific achievements in AI to date. But even he acknowledges that achieving truly general artificial intelligence will still require “much more innovation.”

AlphaFold, the Nobel Prize-winning scientific work, also needs "more innovation." Photo: Google Deepmind.

In the near future, AI will primarily help increase efficiency rather than foster innovation. A survey published on Arxiv of over 7,000 knowledge workers showed that those who use AI to a high degree of productivity reduced their email processing time by an average of 3.6 hours per week (equivalent to 31%), while collaborative tasks remained virtually unchanged.

However, if everyone delegates email replies to ChatGPT, the number of emails in inboxes could increase, undermining the initial efficiency. According to the FT , experience from the US productivity recovery in the 1990s shows that the benefits of a new tool will quickly fade if not accompanied by genuine innovative breakthroughs.

Source: https://znews.vn/nghich-ly-nang-suat-cua-ai-post1561451.html