In a Bloomberg interview, journalist Karen Hao, author of the book "Empire of AI," stated that the modern vector of artificial intelligence development is not the only one possible. In her view, the industry can evolve without an excessive concentration of power in the hands of a few corporations, the exploitation of low-paid labor for data labeling, and colossal energy consumption.
What Happened
Journalist Karen Hao, in her recent interview with Bloomberg, criticized the dominant scaling laws paradigm. She pointed out that the current path of AI development relies on the aggressive concentration of computing power and data, which creates serious socio-economic and environmental costs.
Context
The modern AI industry is characterized by an "arms race" of large language models, where success is determined by the scale of resources used. This has led to a model criticized for its reliance on the exploitation of workers in developing countries for data labeling and for its significant environmental footprint due to the high energy consumption of data centers.
Why It Matters for the Industry
For the industry, this signifies a potential shift from infinite parameter scaling toward searching for more efficient and decentralized development methods. This could stimulate the growth of research grants in the field of efficiency, the development of small language models (SLMs), and the implementation of new reporting standards for environmental footprints (ESG metrics) and the ethics of data supply chains.
Why It Matters for Users
For readers and users, this opens a discussion on how technological progress does not have to be accompanied by social inequality or environmental damage. In the long term, this could lead to the emergence of more sustainable and accessible technologies that are not dependent on a narrow circle of tech giants.
What Remains Unknown / Limitations
The discussion is predominantly socio-economic and strategic in nature; therefore, no direct technical disagreements regarding model training methodologies are provided in the presented materials.
Sources
Author
Look at AI, Editorial Staff