DeepMind’s David Silver just raised $1.1B to build an AI that learns without human data

The News

David Silver, formerly a leading researcher at DeepMind, has launched Ineffable Intelligence, a new AI lab, and secured $1.1 billion in funding, valuing the company at $5.1 billion [1]. This marks a significant departure from his previous role at Google DeepMind, a subsidiary of Alphabet Inc. [2], and signals a potential shift in AI research toward reducing reliance on large, human-labeled datasets [1]. The funding round, led by undisclosed investors, aims to develop AI models capable of learning and reasoning without extensive human-annotated data, a major technical challenge [1]. The formation of Ineffable Intelligence and the scale of the funding highlight growing interest in self-supervised or unsupervised learning within the AI community [1]. Specific details about team size and initial research focus remain undisclosed [1].

The Context

Silver’s departure from DeepMind and founding of Ineffable Intelligence reflect a blend of technical ambition and strategic business decisions [1]. His contributions to DeepMind include pivotal roles in AlphaGo, AlphaZero, and AlphaFold [2]. AlphaGo’s 2016 victory over a world champion Go player demonstrated reinforcement learning’s power, where AI learns through trial and error without explicit human instruction [2]. AlphaZero expanded this by mastering Go, chess, and shogi from scratch using only game rules [2]. AlphaFold, arguably his most impactful project, revolutionized protein structure prediction, a decades-long challenge for biologists [2]. This achievement relied on deep learning and geometric reasoning but still required substantial datasets of known protein structures [2]. Ineffable Intelligence’s focus on data-agnostic learning represents a deliberate shift away from this data-intensive paradigm [1].

The technical architecture of Ineffable Intelligence remains undisclosed [1]. However, its stated goal—AI that learns without human data—suggests a focus on contrastive learning, generative adversarial networks (GANs), and world models [4]. Contrastive learning extracts meaningful features by comparing similar and dissimilar data points without explicit labels [4]. GANs, while typically used for generative tasks, can also enable unsupervised feature learning [4]. World models, a nascent research area, aim to build internal environment representations for planning and reasoning without constant real-world interaction [4]. The challenge lies in achieving sufficient fidelity and efficiency for practical use [4]. The data stack, a critical bottleneck for enterprise AI adoption, remains a key hurdle [4]. Many organizations struggle with data quality, accessibility, and governance, hindering large-scale AI deployment [4]. Ineffable Intelligence’s focus on data-agnostic learning could address these issues, offering advantages in enterprise applications [4].

The timing of this venture coincides with broader industry trends. Warner Bros. Discovery’s decision to shelve Coyote v. Acme reflects a trend of prioritizing short-term financial gains over long-term creative investment [3]. Conversely, AI-designed drugs developed by a DeepMind spinoff, now in human trials, showcase AI’s potential to unlock value in other sectors [2]. This underscores growing confidence in AI’s problem-solving capabilities [2].

Why It Matters

Ineffable Intelligence’s approach has implications beyond AI research labs [1]. For developers, data-agnostic AI could reduce the burden of data annotation, a costly and time-intensive process [1]. This might democratize AI development, enabling smaller teams to build sophisticated models without massive datasets [1]. Technical friction from data preparation and cleaning—major contributors to project delays—would also be minimized [4]. However, the transition to these techniques may require new skillsets, creating demand for engineers specializing in unsupervised learning [4].

For enterprises and startups, training models on unlabeled data could unlock new applications [1]. Industries like manufacturing, healthcare, and finance, which face data scarcity or privacy concerns, could benefit [1]. Reduced reliance on human-labeled data also lowers the risk of bias embedded in those datasets, a growing ethical concern [1]. The $1.1 billion funding signals strong commercial confidence in this approach, suggesting potential disruption to the AI landscape [1]. Companies reliant on data annotation infrastructure may face challenges [1]. The cost of acquiring and labeling data currently dominates many AI budgets, and a shift to data-agnostic learning could diminish these investments’ value [4].

Adaptability will determine success in this ecosystem [1]. DeepMind, now competing with Silver’s venture, has the resources to explore both data-driven and data-agnostic approaches [2]. Startups specializing in unsupervised learning and data infrastructure solutions are also poised to benefit [4]. Slow adopters or those overly reliant on traditional data-intensive methods may struggle [1].

The Bigger Picture

Ineffable Intelligence’s emergence and funding signal a broader AI industry trend: growing recognition of current data-centric approaches’ limitations [1]. While large language models like GPT-4 show impressive capabilities, their reliance on massive datasets and computational resources raises sustainability and accessibility concerns [1]. Data-agnostic AI could offer a path toward more efficient, equitable, and robust systems [1]. This aligns with the push for foundation models adaptable to diverse tasks with minimal fine-tuning [1].

Competitors are also exploring alternatives. DeepMind continues advancing reinforcement and self-supervised learning [2]. Other companies focus on federated learning, which trains models on decentralized data without direct access to raw data [4]. The race to develop efficient, adaptable AI is intensifying, and Ineffable Intelligence’s success could accelerate this trend [1]. Over the next 12–18 months, expect increased investment in unsupervised learning, greater emphasis on data infrastructure, and deeper ethical debates about data-driven AI [1].

Daily Neural Digest Analysis

Mainstream media coverage of Ineffable Intelligence has emphasized the $1.1 billion funding and Silver’s departure from DeepMind [1]. However, the deeper significance lies in the potential paradigm shift toward data-agnostic AI [1]. Sources do not specify the technical challenges of building AI without human data or its limitations [1]. While reducing data dependency is promising, unsupervised learning often requires significant computational resources and careful tuning to match supervised methods’ performance [4]. Overhyping this approach risks neglecting its inherent complexities [1]. Ineffable Intelligence’s success will depend on securing funding and demonstrating tangible results [1]. The question remains: can truly “ineffable” intelligence be achieved without human knowledge, or will we simply trade one set of limitations for another?

---

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/