The Rise of Ineffable Intelligence
David Silver, the computer scientist formerly responsible for DeepMind’s AlphaGo and AlphaZero breakthroughs, has launched a new venture titled Ineffable Intelligence. The startup has secured $1.1 billion in seed funding at a valuation of approximately $5.1 billion, marking what is described as the largest seed round in European history.
The funding round includes participation from several high-profile investors, including Sequoia, Lightspeed, Nvidia, and Google, alongside capital from the UK government. Despite the significant capital injection, the company currently has no commercial product.
The “Era of Experience” Thesis
Silver’s strategic direction for Ineffable Intelligence departs from the current industry standard of Large Language Models (LLMs). According to reports from a recent FT Lunch discussion, Silver posits that contemporary chatbots have reached a ceiling because they rely on human-generated text and feedback. He characterizes human data as a finite resource, comparable to fossil fuels that are nearing depletion.
Instead, Silver is pursuing a “superlearner” model. This approach seeks to move into what he terms the “era of experience,” where AI systems learn through pure trial and error. This methodology mirrors the development of AlphaZero, which taught itself chess from first principles without human datasets.
Safety Protocols and Simulation
To address the risks associated with developing advanced intelligence, Silver has committed to a “sandboxing” safety strategy. The company intends to develop and study these systems exclusively within simulated environments. These simulations will consist of societies of AI agents designed to interact independently, potentially inventing their own systems of language, mathematics, and currency.
Silver has established a three-year runway to demonstrate either superintelligence within these simulations or significant evidence that such a milestone is imminent.
The proposal has met with skepticism from some observers on social media platforms like Reddit. Some commenters have questioned the efficacy of digital sandboxing, suggesting that containment is difficult to maintain in networked environments, while others argued that mathematical breakthroughs in a sandbox may not translate to the complexities of the physical world.
Ethics and Social Responsibility
Addressing the broader implications of AI development, Silver noted that the current race between international laboratories and nations is “unhealthy,” noting that governments and societies are struggling to adapt to the pace of technological change. He also suggested that certain industry leaders have failed to exercise sufficient responsibility.
In terms of corporate policy, Ineffable Intelligence has stated it will not engage in military work. Silver has also pledged to donate all equity gains from the venture to charity. When questioned on the necessity of developing systems that exceed human intelligence, Silver pointed to humanity’s historical struggles with war, poverty, and preventable disease as the primary justification for the pursuit of such technology.