Skip to main content
ZICQ
中 Log in / Sign up

Wiki Models & Products

Google DeepMind

Models & Products
Aliases： DeepMind Google DeepMind Alphabet DeepMind ·2026-09-19

Google DeepMind

Google DeepMind is Google's AI research lab, descended from DeepMind Technologies, founded in London in 2010 and acquired by Google in 2014 for roughly $500M. In 2023 Google merged DeepMind with Google Brain into the modern Google DeepMind, led by Demis Hassabis alongside co-founders Mustafa Suleyman (later at Inflection / Microsoft AI) and Shane Legg.

Historical milestones

  • 2013: first demonstrated deep RL generality on Atari games
  • 2014: acquired by Google
  • 2016: AlphaGo defeated Lee Sedol 4:1, putting deep RL on the map
  • 2017: AlphaGo Zero surpassed humans via pure self-play; AlphaZero extended to chess and shogi
  • 2018: AlphaFold predicted 3D protein structures
  • 2020: AlphaFold 2 approached experimental-grade accuracy at CASP14
  • 2021: AlphaFold database open-sourced, covering 200M+ structures
  • 2022: AlphaTensor discovered new matrix-multiplication algorithms
  • 2023: AlphaFold Server, AlphaProof (IMO silver-equivalent math)
  • 2024: post-merger Gemini 1.0 / 1.5 series

Active research lines

  • Gemini family: see gemini.md, natively multimodal
  • AlphaFold: biology / drug-discovery pipeline
  • AlphaProof / AlphaGeometry: formal mathematical reasoning
  • Project Astra: general vision + voice agent
  • SynthID: watermarking and tracing for AI-generated content
  • Gemini Robotics: foundation models for robotics

Influence

  • AlphaGo / AlphaZero turned deep RL into a mainstream AI technique
  • AlphaFold cut protein structure discovery from months of lab work to minutes of prediction
    • 2024 Nobel Prize in Chemistry went to Hassabis / John Jumper (DeepMind)
      • David Baker, with AlphaFold as the core contribution
  • Took over Gemini to put Google back in the top tier of the LLM era
  • Long-time commitment to AI safety and alignment — the most-cited "serious AI lab" by both academia and regulators

Limitations

  • Slower commercialization than OpenAI / Anthropic; the Gemini API came late
  • Bound to Google's wider strategy (Search, Ads, Android), decisions are constrained by the parent
  • Multiple teams (Gemini, Workspace, Cloud AI) complicate coordination, product cadence is less consistent than OpenAI's