DeepSeek inspires the 'deep seek' for humanity – Democratization of AI

Editor's note: With the launch of DeepSeek's open-source AI model, the barriers to entry and the costs of AI development have significantly decreased. It is foreseeable that AI's role in human society will continue to expand, both in its impact on the labor market and its integration into daily life. As the annual Two Sessions commences, CGTN's SCI-Tech Section focuses on key issues surrounding artificial intelligence and launches the "DeepSeek inspires the 'deep seek' for humanity" series. Authored by Li Mingling, an assistant researcher at the Institute of Public Policy at South China University of Technology, these articles explore some critical questions, including how we can ensure that technological progress – while promoting fairness, justice, and human welfare – does not diminish human value. The article reflects the author's opinions, and not necessarily those of CGTN.

The domestic Large Language Model (LLM) DeepSeek has recently taken the internet by storm, making many in China realize that the era of AI has truly arrived.

DeepSeek's ability to achieve high performance comparable to advanced LLMs like ChatGPT-4o at a significantly lower cost has sparked widespread discussions in the AI community and beyond. Before DeepSeek, China had already launched dozens of well-known Large Language Models. However, despite their technical sophistication, these models struggled to achieve widespread adoption, and AI was previously a topic largely confined to tech enthusiasts, younger generations, or highly educated individuals. DeepSeek's rise has now resonated with a much broader audience, including those with less formal education and older generations, such as people aged 50-60. To some extent, DeepSeek has played a pivotal role in bringing the topic of AI into the mainstream consciousness of the general public in China.

Many articles have explored the international competition, innovation mechanisms, technological landscape, and corporate cultures driving AI development. However, there is another important question worth considering – one that we all must confront in the years ahead: how should we navigate and adapt to the AI era?

This question is not only tied to AI research and development, but also encompasses the direction of AI's evolution, humanity's place in the AI era, the societal changes AI will bring, and more – issues that are deeply connected to the future of humanity.

Before DeepSeek's emergence, a clear digital divide existed, with structural gaps in access, usage, and benefits from AI across different groups.

However, DeepSeek has significantly narrowed this gap. Beyond raising public awareness of LLMs, it has engaged more developers in AI innovation and made AI more accessible and affordable to a broader audience. AI is poised to become a public good that transcends class, education, and national boundaries. Today, DeepSeek has accelerated the arrival of that moment.

In this regard, alongside its technological innovation, the R&D team's understanding of AI as a public good is essential. DeepSeek achieves digital equality in AI through two key aspects: development and usage.

Firstly, through open-source initiatives, DeepSeek shifts the development model from being dominated by a few leading companies to one that encourages widespread participation from numerous developers.

Open-source means releasing the source code of a product to the public through relevant platforms, allowing others to access, modify, distribute and contribute to it. On December 26, 2024, DeepSeek launched the V3 Model and simultaneously released a 53-page technical report, making it the world's first fully open-source Model of Mixed Expertise (MoE). On January 20, 2025, DeepSeek launched the R1 Model and simultaneously released its source code on Hugging Face, the open-source community platform.

Prior to that, major players like OpenAI and Google focused largely on closed-source models, especially with their most advanced AI systems, creating algorithmic "black boxes" where most external researchers and developers had limited or restricted access to the inner workings of these large models. DeepSeek's open-source approach can be seen as the democratization of AI. It not only allows users to run the model locally with their own databases, but also empowers developers to make specific modifications and engage in secondary development, thereby contributing to the model's optimization process.

Compared to leading tech companies, most researchers could be considered grassroots developers, but that doesn't mean they lack the potential to drive breakthrough innovations. Yann LeCun, chief AI scientist at Meta, said that DeepSeek's success represented a victory for open-source AI models, not necessarily a win for China over the U.S., according to a recently published CNBC report.

Secondly, DeepSeek transforms the usage model of AI from being exclusive to a select group of members to being widely accessible to the general public, through a strategy of low or even free pricing.

DeepSeek-R1's Application Programming Interface (API) service is priced 30 times lower than ChatGPT-o1. Moreover, ordinary users can not only receive answers to various questions on DeepSeek, but also observe the process of deep thinking and reasoning behind those answers. In contrast, to access this advanced feature on ChatGPT-o1, users must first obtain a membership, which can cost up to $200 per month.

DeepSeek has directly disrupted the existing profit models of some tech companies and rapidly accelerated the widespread adoption of inclusive AI applications. Recently, influenced by DeepSeek, OpenAI announced the open-source release and low-pricing strategy for its o3-mini Model. Baidu's ERNIE (also known as Wenxin Yiyan) also announced that it would be free starting April 1. From a technology diffusion perspective, the democratization of AI is gaining momentum.

Algorithmic advancements, computing power efficiency, and open-source initiatives have become the technological foundation for AI to reach the general public. The democratization of AI will also usher in a new era of more meaningful technological competition. As Jeffrey Ding, Assistant Professor at George Washington University, notes in his book Technology and the Rise of Great Powers, great power competition hinges not on who leads in innovation, but on who can effectively adopt and diffuse innovations across industries.

Throughout the history of industrial revolutions, from the steam engine and electricity to information networks, true progress has belonged to those who most effectively promoted the diffusion of technology.

(Cover: VCG)