Unveiling the Commencement of Sufficient Artificial Intelligence Era through DeepSeek
In the winter of 2024, the lesser-known Chinese enterprise, DeepSeek, stunned the AI community by releasing its experimentally acclaimed model, DeepSeek-v3. This groundbreaking achievement matched the performance of renowned Western AI titans' advanced models, but at a fraction of the usual training and inference expenses.
A month later, DeepSeek sent shockwaves globally with the introduction of the DeepSeek-R1 model, which competed on a level playing field with OpenAI's latest reasoning model. This breakthrough has been attributed to China's "ChatGPT moment." With the geopolitical rivalry between the U.S. and China as the dominant narrative, the true significance of this technological leap often goes unnoticed – a critical shift towards cost efficiency in AI development.
DeepSeek's strategic focus on utilizing Nvidia chips, a valuable but scarce resource for Chinese enterprises due to the U.S. chip embargo, signaled that it was possible to create top-tier models with scaled-down resources. This led the AI industry to reassess its target for AI advancements – prioritizing cost-effective models over the sole pursuit of artificial general intelligence (AGI).
This alternative approach to AGI, coined 'Artificial Good-enough Intelligence,' further harnesses the value of existing AI technologies, while recognizing the importance of accessibility and the desire for wide-spread adoption.
Generative AI: The Closing Chapter of Innovation
Economist Joseph Schumpeter's theory of innovation describes a three-phase process: invention, innovation, and diffusion. Generative AI has embarked on the final stage of this journey.
The invention stage introduced groundbreaking inventions, such as the development of artificial neural networks by the 2024 Nobel laureates John Hopfield and Geoffrey Hinton, and the transformer architecture dynamicizing contemporary Generative AI models.
The innovation stage saw commercialization, with innovations like OpenAI's ChatGPT bringing advanced language models to the public.
DeepSeek's advancements signaled the diffusion phase, focusing on cost efficiency to promote massive deployment.
As one technology transitions from invention to diffusion, the focus shifts to engineering and operational improvements that drive significant cost reductions. Leveraging techniques similar to those employed by quantitative traders to improve execution speed, DeepSeek has opened up new possibilities in AI innovation through its open-source approach.
The Valley of Controversy
Despite the controversy surrounding its methodology, DeepSeek's techniques offer potential for negotiation on the compensation of content creators in AI.
Instead of relying on terabytes of text data generated by humans, DeepSeek utilized machine-generated outputs from other AI models during training, a process known as model distillation. While this approach allowed for significant cost savings without sacrificing model quality, it sparked accusations from OpenAI, which claimed that DeepSeek had breached its API terms of service by using its API to generate training data.
The collaboration between human and AI-generated content creates a new layer of consideration when discussing the fair use of AI. In December 2023, The New York Times launched a copyright infringement lawsuit against OpenAI and Microsoft, claiming they had used the newspaper's content without compensation to train AI models. This lawsuit ignited a global dialogue on fair use standards in the context of Generative AI.
My research with Angela Huyue Zhang (University of Southern California Gould School of Law) underscores a growing reality in the AI value chain. As AI-generated content becomes more valuable, compensating human creativity becomes increasingly important. To advance in areas such as AI, human creativity remains crucial.
The same rationale applies to synthetic training data, like that used by DeepSeek. If AI models could generate alternative cost-efficient solutions, companies should be incentivized to continue pushing technological boundaries.
This dynamic may evolve into a bifurcated AI development landscape, featuring a cohort of frontier AI developers catering to a separate group specializing in cost-efficient mass AI model development. If this vision is implemented, a well-structured market for synthetic data would help cover expenses related to cutting-edge research, ensuring that innovation remains a key priority.
S. Alex Yang is a Professor of Management Science and Operations at London Business School. Specializing in supply chain management, finance, and technology, his latest research encompasses the role of digital technologies and business models, such as AI, blockchain, and digital platforms, in shaping innovations and governance in international markets.
- In response to DeepSeek's cost-effective AI models, the AI industry is now prioritizing the development of 'Artificial Good-enough Intelligence' over artificial general intelligence (AGI).
- OpenAI, following DeepSeek's breakthrough, acknowledged the significance of this approach and expressed interest in collaborating with Chinese AI companies.
- The Chinese enterprise, DeepSeek, has attracted attention from companies like Google and Microsoft, sparking conversations about potential partnerships for innovation.
- In 2023, the debate around AI-generated content and copyright infringement escalated, with The New York Times filing a lawsuit against OpenAI and Microsoft.
- Theorist Joseph Schumpeter predicted that AI would follow an innovation trajectory, moving from invention to diffusion, and this is what DeepSeek represents in the world of generative AI.
- As AI progresses through the diffusion phase, focus shifts to engineering and operational improvements, driven by cost reduction techniques, as demonstrated by DeepSeek.
- China's artificial intelligence advancements, led by companies like DeepSeek, may contribute to the development of a bifurcated AI industry, focusing on mass AI model development and cutting-edge research.
