The Man Who Taught AI to Beat Go Just Raised $1.1 Billion to Unlearn Everything

David Silver spent over a decade at DeepMind proving that artificial intelligence could conquer domains once thought to be the exclusive province of human intuition. He helped build AlphaGo, which stunned the world by defeating a Go champion using moves no human would ever consider. He built AlphaZero, which taught itself to master chess, shogi, and Go from nothing but the rules. And then there was AlphaFold, arguably the most consequential AI project in history, which cracked a 50-year-old problem in biology by predicting protein structures with breathtaking accuracy.

Now, Silver is walking away from all of that. Not from AI itself, but from the very paradigm that made those breakthroughs possible.

The former DeepMind principal scientist has launched Ineffable Intelligence, a new AI lab that has already secured $1.1 billion in funding at a $5.1 billion valuation [1]. The mission statement is as audacious as it is technically fraught: build AI systems that can learn and reason without relying on human-annotated data. No millions of labeled images. No armies of contractors tagging sentences. No curated datasets of known protein structures. Just raw, unlabeled data and algorithms smart enough to extract meaning from it on their own.

If Silver succeeds, he won't just have built another AI company. He will have fundamentally rewritten the playbook for how intelligence is engineered.

The Data Dependency That Haunts Modern AI

To understand why this matters, you have to appreciate just how deeply the current AI industry is addicted to human labor. Every major breakthrough of the last decade—from GPT-4 to DALL-E to Stable Diffusion—has been built on the backs of massive, meticulously curated datasets. These datasets are expensive to create, time-consuming to maintain, and increasingly controversial.

Consider the economics. For most enterprise AI projects, data acquisition and labeling consume anywhere from 60% to 80% of the total budget [4]. A single computer vision project might require hundreds of thousands of bounding boxes drawn by hand. A natural language model needs millions of sentences annotated for sentiment, intent, or entity recognition. The infrastructure required to manage this pipeline—data lakes, version control systems, quality assurance workflows—has become an industry unto itself.

This is the bottleneck that Silver intends to break. Ineffable Intelligence's focus on data-agnostic learning represents a deliberate and radical departure from the data-intensive paradigm that defined his own greatest achievements [1]. AlphaFold, for all its brilliance, required substantial datasets of known protein structures to train on. AlphaZero, while more autonomous, still needed the explicit rules of its games to bootstrap learning. Silver's new venture aims to go further: build systems that can discover structure and patterns in raw data without any human annotation whatsoever.

The technical architecture remains undisclosed, but the stated goals point toward a convergence of several cutting-edge approaches [4]. Contrastive learning, which extracts meaningful features by comparing similar and dissimilar data points without explicit labels, is almost certainly part of the stack. Generative adversarial networks (GANs) , typically associated with image generation, can also enable unsupervised feature learning. And then there are world models—a nascent but rapidly evolving research area focused on building internal representations of environments that allow AI systems to plan and reason without constant interaction with the real world [4].

The challenge, of course, is making these techniques work at scale. Unsupervised learning has long been the holy grail of AI research, but it has historically struggled to match the performance of supervised methods on specific tasks. The computational requirements can be enormous, and the tuning required to get these systems to converge is often more art than science [4]. Ineffable Intelligence's success will depend not just on securing funding, but on demonstrating that these approaches can deliver tangible, production-ready results.

From DeepMind to Ineffable: A Technical and Strategic Pivot

Silver's departure from DeepMind is not just a personnel change; it is a signal about the direction of the entire field. His contributions to the Alphabet subsidiary were foundational. AlphaGo's 2016 victory over Lee Sedol demonstrated the power of reinforcement learning, where AI learns through trial and error without explicit human instruction [2]. AlphaZero expanded this paradigm by mastering Go, chess, and shogi from scratch using only the rules of each game [2]. These were not just technical achievements; they were philosophical statements about the nature of intelligence itself.

But DeepMind, now a subsidiary of Alphabet Inc., operates within the constraints of a large corporation [2]. Its research priorities are influenced by the commercial interests of its parent company. Silver's move to found an independent lab suggests a desire for greater autonomy in pursuing a research agenda that may be too radical—or too risky—for a corporate R&D division.

The timing is also significant. The broader AI industry is experiencing a moment of reckoning. Large language models like GPT-4 have demonstrated impressive capabilities, but their reliance on massive datasets and computational resources raises serious questions about sustainability and accessibility [1]. The cost of training a single frontier model now runs into the hundreds of millions of dollars. The energy consumption is staggering. And the data itself is increasingly subject to legal and ethical scrutiny, with lawsuits over copyright infringement and privacy violations becoming routine.

Ineffable Intelligence's approach could offer a path toward more efficient and equitable AI systems [1]. If models can learn effectively from unlabeled data, the barriers to entry for AI development drop significantly. Smaller teams and startups could build sophisticated models without the massive data infrastructure that currently favors deep-pocketed incumbents. Industries like manufacturing, healthcare, and finance, which often face data scarcity or privacy concerns, could unlock new applications [1]. And the reduced reliance on human-labeled data lowers the risk of bias embedded in those datasets—a growing ethical concern that has plagued everything from facial recognition to hiring algorithms [1].

The Enterprise Implications of Data-Agnostic AI

For developers and engineering teams, the promise of data-agnostic AI is almost intoxicating. Data preparation and cleaning are consistently cited as the most time-consuming and frustrating parts of any machine learning project. The technical friction from these tasks—figuring out which features matter, handling missing values, ensuring consistency across sources—is a major contributor to project delays and failures [4].

If Silver's approach works, much of that friction disappears. Teams could feed raw data directly into models and let the algorithms figure out the structure on their own. This would democratize AI development, enabling smaller teams to build sophisticated models without massive datasets [1]. It would also create demand for a new generation of engineers specializing in unsupervised learning techniques, rather than the data wrangling skills that currently dominate job descriptions [4].

But the transition won't be seamless. Organizations that have invested heavily in data annotation infrastructure—the labeling platforms, the quality assurance workflows, the armies of contractors—may find their investments devalued [4]. The cost of acquiring and labeling data currently dominates many AI budgets, and a shift to data-agnostic learning could diminish the value of those investments significantly. Companies that have built competitive advantages around their proprietary labeled datasets may need to rethink their strategies entirely.

For enterprises and startups, the implications are profound. Industries that have struggled to adopt AI due to data scarcity—specialized manufacturing, rare disease research, niche financial instruments—could suddenly become viable targets for machine learning [1]. Privacy-sensitive sectors like healthcare and finance could train models on raw data without the risk of exposing personally identifiable information through annotations. And the reduced reliance on human-labeled data could accelerate the pace of AI deployment across the board.

The Competitive Landscape and the Race for Unsupervised Intelligence

Ineffable Intelligence is not entering an empty field. DeepMind, now competing with Silver's venture, has the resources to explore both data-driven and data-agnostic approaches simultaneously [2]. The company continues to advance reinforcement learning and self-supervised learning, and its track record suggests it will remain a formidable competitor.

Other players are also exploring alternatives to the data-centric paradigm. Some companies are focusing on federated learning, which trains models on decentralized data without direct access to raw information [4]. Others are investing in synthetic data generation, using AI to create training data that doesn't require human annotation. And the open-source community is making rapid progress on techniques like self-supervised learning, with models like Meta's DINO and Google's BYOL demonstrating that meaningful representations can be learned without labels.

The race to develop efficient, adaptable AI is intensifying, and Ineffable Intelligence's success could accelerate this trend significantly [1]. Over the next 12 to 18 months, we can expect to see increased investment in unsupervised learning, greater emphasis on data infrastructure that supports these approaches, and deeper ethical debates about the implications of data-driven AI [1]. Startups specializing in unsupervised learning and data infrastructure solutions are particularly well-positioned to benefit [4]. Slow adopters or those overly reliant on traditional data-intensive methods may struggle to keep pace [1].

The Unanswered Questions and the Path Forward

For all the excitement surrounding Ineffable Intelligence's launch, significant questions remain unanswered. The company has not disclosed its team size, its initial research focus, or the specific technical approach it plans to pursue [1]. The $1.1 billion funding round, led by undisclosed investors, signals strong commercial confidence, but it does not guarantee technical success [1].

The fundamental challenge is whether truly "ineffable" intelligence—AI that learns without any human knowledge or guidance—is even achievable. Unsupervised learning has made remarkable progress in recent years, but it still requires significant computational resources and careful tuning to match the performance of supervised methods on specific tasks [4]. The world models that Silver's approach likely relies on are still in their infancy, and scaling them to handle the complexity of real-world environments remains an open research problem.

There is also the question of what gets lost when we remove human data from the training process. Human-annotated datasets are imperfect—they contain biases, errors, and cultural assumptions—but they also encode valuable knowledge that has been accumulated over generations. An AI system that learns entirely from raw data might discover patterns that humans have missed, but it might also miss patterns that humans consider essential. The trade-offs are not yet well understood.

Mainstream media coverage of Ineffable Intelligence has focused on the eye-popping funding numbers and Silver's departure from DeepMind [1]. But the deeper significance lies in the potential paradigm shift toward data-agnostic AI [1]. If Silver succeeds, he will have demonstrated that the most powerful form of intelligence is not the one that learns from human teachers, but the one that learns from the world itself. If he fails, he will have provided the field with invaluable lessons about the limits of unsupervised learning.

Either way, the bet is worth watching. The $1.1 billion that investors have placed on Ineffable Intelligence is not just a bet on a single company; it is a bet on a vision of AI that is more autonomous, more efficient, and less dependent on the messy, expensive, and increasingly contentious process of human annotation. Whether that vision proves to be a breakthrough or a mirage will shape the direction of AI research for years to come.

For developers, entrepreneurs, and technologists watching from the sidelines, the message is clear: the era of data-centric AI is not over, but its dominance is no longer guaranteed. The tools and techniques that emerge from Ineffable Intelligence—and from the competitors it will inevitably inspire—could fundamentally change how we build, deploy, and think about artificial intelligence. The only certainty is that the next chapter of this story will be written without the crutch of human-labeled data.

---

References

[1] Editorial_board — Original article — https://techcrunch.com/2026/04/27/deepminds-david-silver-just-raised-1-1b-to-build-an-ai-that-learns-without-human-data/

[2] Wired — AI-Designed Drugs by a DeepMind Spinoff Are Headed to Human Trials — https://www.wired.com/story/wired-health-2026-how-ai-is-powering-drug-discovery-max-jaderberg/

[3] The Verge — The plan to quietly kill Coyote v. Acme blew up in David Zaslav’s face — https://www.theverge.com/column/918079/david-zaslav-warner-bros-discovery-coyote-v-acme-batgirl-tax-writeoffs

[4] MIT Tech Review — Rebuilding the data stack for AI — https://www.technologyreview.com/2026/04/27/1136322/rebuilding-the-data-stack-for-ai/