Source: Financial Times (Published April 1, 2025 )
In a significant strategic shift, Google's renowned artificial intelligence research arm, DeepMind, is reportedly tightening its control over the publication of its scientific research.
According to a recent Financial Times report citing current and former research scientists, the London-based group led by Nobel laureate Sir Demis Hassabis has implemented stricter vetting processes and increased bureaucracy, making it more challenging for its researchers to share their findings publicly.
This marks a departure from DeepMind's long-standing reputation as a bastion of open, groundbreaking AI research – the very culture that produced foundational work like the 2017 "transformers" paper, which underpins much of today's generative AI boom.
Why the Change? The Competitive AI Landscape
The core reason cited for this change is the intensifying race for AI dominance. Sources within the FT article suggest DeepMind is now more hesitant to release research that:
- Reveals innovations competitors could exploit.
- Might portray Google's own AI models, like Gemini, unfavorably compared to rivals such as OpenAI's GPT-4.
This strategic pivot comes after the 2023 merger of Google's DeepMind and Brain AI units, a move partly driven by concerns that Google was losing ground to competitors like OpenAI.
The focus has increasingly shifted towards integrating AI into commercial products and demonstrating tangible progress to investors. As one current researcher noted, releasing something like the original "transformers" paper today seems unlikely under the new approach.
New Processes and Their Impact
The changes reportedly include:
- A tougher internal review process for publications.
- A potential six-month embargo on releasing "strategic" papers related to generative AI.
- Researchers needing to convince multiple stakeholders of a paper's publication merits.
While DeepMind contends these changes help researchers by clarifying early on what work is publishable, preventing wasted effort, and notes it still publishes hundreds of papers annually, the shift has caused friction.
Some current and former staff expressed frustration, feeling the environment has moved away from prioritizing public good through research towards a more product-centric model.
The article highlights concerns that an inability to publish can be detrimental to a researcher's career, contributing to some departures. There's also mention of internal resources, like computing power, being increasingly directed towards projects that enhance Google's commercial AI products, such as the Gemini suite.
Interestingly, the report mentions an instance where DeepMind allegedly blocked a paper critical of Gemini's capabilities compared to GPT-4, but also blocked one revealing vulnerabilities in OpenAI's ChatGPT, potentially to avoid appearing overtly combative. DeepMind stated it follows responsible disclosure policies for security vulnerabilities, allowing companies time to fix flaws before public release.
The Bigger Picture
This evolution at DeepMind reflects a broader tension in the AI field between the traditional academic culture of open sharing and the immense commercial stakes involved. Under Sir Demis Hassabis, who aims to balance Google's commercial goals with the quest for artificial general intelligence (AGI), the message seems clear: DeepMind operates as a company focused on its mission within a competitive landscape, not purely as an academic institution.
Shifting Tides: Google DeepMind Reportedly Slows Research Publications to Sharpen Competitive Edge