Open-Sourced AI: A Made-in-Canada Approach to Generative AI

This blog was provided by our partners at Meta.

Over the last two years, generative AI and the foundation models that power it have given us a glimpse into the significant economic and social benefits that lie ahead. From scientific discoveries to education to energy efficiency, the use cases seem boundless. 

Canada has long been a leader in AI, but the enormous resources and compute needed to keep up with the cutting edge of generative AI model development makes for a compelling case that a made-in-Canada approach requires strong public-private partnerships based on open-source technologies that are free to use, highly customizable, and safe and secure for deployment. 

What is Open-Source AI?

Unlike proprietary AI models like Open AI’s ChatGPT and Google’s Gemini, where applications can be built on top of APIs but access to the underlying model are restricted, an open-source approach to AI involves making public the underlying foundation AI models, which can be downloaded, inspected, fine-tuned, and tailored for an unlimited amount of use cases and applications, all for free. Meta’s Llama (for Large Language Model Meta AI) is one prominent example of an open-source generative AI model.

Already, hospitals and health professionals are using open-source models like Llama to run health applications on premise, processing patient data locally within the hospital’s secure IT environment rather than sending sensitive data into the cloud for processing. Similarly, non-English speaking countries are taking Llama and training the model to speak different languages, creating highly localized and culturally relevant large language models and applications.

Just last month, during United Nations General Assembly week,  we unveiled alongside UNESCO and Hugging Face a new online translation interface built  on Meta’s No Language Left Behind (NLLB) open-source AI model, supporting high-quality translation in 200 languages, including low-resourced and Indigenous languages like Asturian, Luganda, Maori and Urdu. Through this partnership, open-source AI is helping power the International Decade for Indigenous Languages.

So What are the Benefits of Open-Source AI for Canada?

1. More jobs and economic growth
   According to J.P. Morgan Research, AI could increase global GDP by $10 trillion and accessible open source AI is a quicker path to getting there. Open source models will help spread this general purpose technology, spur innovations and its economic benefits.  As one example of this democratization effect, as of early April, Llama has been downloaded over 180 million times by developers, innovators and entrepreneurs around the world, including in Canada.
   
2. Better privacy, security and control
   Open source facilitates more security and control over data access because open models can be downloaded onto local hardware. There is no need for developers to share any data with external providers – a significant advantage, particularly when much of the innovation from open models results from applications of novel (often proprietary) datasets. Organizations with concerns over data access – for data protection, commercial, or other reasons – can experiment and innovate with open models with greater peace of mind.
   
3. Increased competition will come from an equal playing field
   Open source models , like Llama, democratize access and reduce barriers to entry by providing state of the art AI tools to a wider range of developers and innovators – lowering the cost of innovation and experimentation that would otherwise be prohibitive.  Rather than being deployed and guarded by a  few technology companies with the computing infrastructure to pretrain large models, open source AI can be used by the  broader community of developers and organizations that stand to benefit from these innovations. This puts all of our country’s small and medium sized businesses  on an equal playing field – with a fair shake at success with access to the tools to solve for their own unique business challenges.
   
4. The pace of scientific discovery will be accelerated
   Foundation models are critical for modern scientific research, and broader access to them is necessary for scaling scientific discovery. Housing some of the top researchers in the world, Canada will benefit from turbocharging discoveries using AI.
   
5. A made in Canada approach and better social outcomes
   Open source AI models can be customized and fine-tuned using Canadian data leading to truly made-in Canada large language models (LLMs) and generative AI applications.

We look forward to discovering what open-source can do for Canada .

Read Meta’s NewsRoom poston Open Source AI  here.

---

By Kevin Chan, Global Policy Campaign Strategies Director

Kevin Chan is Global Policy Campaign Strategies Director at Meta Platforms, and helped lead the effort to build the UNESCO Language Translator, powered by Meta and Hugging Face