This evergreen history article uses authoritative archives and official records. Exact dates are used when documented; gradual inventions and rollouts are described as periods rather than being assigned a misleading single birthday.
Quick facts
- Google used machine learning in search, advertising, translation and recommendations long before generative chat became popular.
- Google Brain began in 2011, while DeepMind was founded in London in 2010 and acquired by Google in 2014.
- DeepMind’s AlphaGo defeated Lee Sedol in Seoul in 2016.
- Google researchers introduced the transformer architecture in 2017.
- Gemini was announced in 2023 as a natively multimodal model family developed by Google DeepMind.
Machine learning inside Google’s early products
Google’s first search engine depended on mathematical ranking rather than modern neural networks, but machine learning gradually entered spam detection, advertising, translation, speech recognition and recommendation. The company’s scale supplied enormous datasets and computing infrastructure, turning AI research into a practical product capability.
Google Translate, launched in 2006, initially used statistical methods and later shifted toward neural machine translation. Search ranking added systems such as RankBrain to interpret queries. Photos, Maps, YouTube and Android also became major environments for computer vision, language understanding and personalisation.
Google Brain and DeepMind
Google Brain began in 2011 as a deep-learning research effort associated with researchers including Jeff Dean, Greg Corrado and Andrew Ng. Large distributed systems trained neural networks across many machines. The project helped move deep learning from academic demonstrations into products used by billions of people.
DeepMind was founded in London in 2010 by Demis Hassabis, Shane Legg and Mustafa Suleyman. Google acquired the company in 2014. DeepMind combined reinforcement learning, neuroscience-inspired ideas and large-scale computation. Its systems learned Atari games and later produced AlphaGo, which defeated Go champion Lee Sedol in Seoul in March 2016.
Transformers and foundation models
In 2017 Google researchers published “Attention Is All You Need,” introducing the transformer architecture. Transformers made it easier to train large sequence models in parallel and became the basis of most modern language models. Google also developed BERT, which improved contextual language understanding in search and natural-language processing.
The company continued research in language, vision, protein science and multimodal learning. DeepMind’s AlphaFold demonstrated that AI could contribute to scientific problems, while large language models showed that one pretrained system could support many products and developer workflows.
Bard, the Google Brain–DeepMind merger and Gemini
The public success of generative chat systems intensified competition. Google launched Bard in 2023, initially based on its LaMDA research and later updated with newer models. In the same year Google combined Google Brain and DeepMind into Google DeepMind, bringing major research groups under one organisation.
Gemini was announced in December 2023 as a family designed for text, images, audio, video and code. Google integrated Gemini across consumer products, developer platforms, cloud services and mobile devices. The name increasingly represented both model families and user-facing assistant experiences.
AI as infrastructure and the next challenge
Google’s advantage lies in connecting AI research with search, Android, Chrome, Workspace, Cloud and custom chips such as tensor processing units. That integration also creates responsibility: errors can spread through services used at global scale, and model deployment affects publishers, creators, advertisers and software markets.
The continuing history of Google AI will depend on factual reliability, cost, energy use, competition, regulation and whether multimodal agents can complete useful tasks without undermining security or user control. Gemini is one chapter in a longer story that began with ranking web pages and expanded into general-purpose systems.
Common misconceptions
- Google did not begin its AI work with Gemini; machine learning had supported products for many years.
- DeepMind was founded independently in London before Google acquired it.
- The transformer architecture was introduced by Google researchers, but it became a global research foundation used by many organisations.
- Gemini is not only a chatbot name; it also refers to model families and developer services.
Timeline: key years and locations
| Year | Location | Event | Why it mattered |
|---|---|---|---|
| 2006 | Mountain View, California / global service | Google Translate launches | Became an early large-scale language-technology product. |
| 2010 | London, United Kingdom | DeepMind is founded | Created a specialised laboratory for general-purpose learning systems. |
| 2011 | Google, California | Google Brain begins | Expanded deep learning with distributed computing. |
| 2014 | London and Mountain View | Google acquires DeepMind | Combined DeepMind research with Google resources. |
| 2016 | Seoul, South Korea | AlphaGo defeats Lee Sedol | Demonstrated the power of reinforcement learning and search. |
| 2017 | Google research, United States | Transformer architecture is published | Became the foundation of modern large language models. |
| 2023 | Mountain View and London | Google Brain and DeepMind combine; Bard and Gemini emerge | Unified major AI efforts and entered the multimodal assistant era. |
| 2024 onward | Global products and cloud platforms | Gemini expands across Google services | Turned AI models into a broad product and developer layer. |
Frequently asked questions
Who created Gemini?
Gemini was developed by Google DeepMind, the organisation formed by combining Google Brain and DeepMind teams.
What was Google’s role in transformers?
A team of Google researchers published the 2017 transformer paper that introduced the architecture now used widely in language and multimodal models.
Is Gemini the same as Google Search?
No. Gemini is a family of AI models and experiences; Search is a separate product that may use multiple AI systems.