With the renewed interest in artificial intelligence (AI) brought about by generative AI (GenAI), there has been a growing focus on the critical but often overlooked issue of biases in AI models. Addressing these biases is increasingly challenging, and it’s crucial to ensure that GenAI doesn’t produce harmful outcomes. In this episode, Dr. Hakim Hacid, Acting Chief Researcher and Executive Director at Technology Innovation Institute (TII), talks about the work the institute is doing to ensure that GenAI is not controlled by a single entity, such as OpenAI, as proprietary technology.

TII aims to make AI open, collaborative, transparent, trustworthy and unbiased. To achieve this, they have created their own open-source AI model called Falcon, which is being adopted by both public and private sectors.

Open source AI research and its importance in the geopolitical landscape

• Hacid discusses TII’s research in GenAI and their involvement in the open-source community.
• Hacid talks about how open source AI is crucial for freedom and independence in a changing geopolitical world.

Open source AI and its challenges

• Hacid explains that Linux has empowered developers and opened opportunities for more openness and freedom.
• Open source AI is more complicated than regular software, with many variables and challenges.
• Hacid argues that the open source AI movement needs to go beyond just sharing model weights and open up training code, data, and other aspects of the pipeline.
• Hacid sees a lot of efforts being made in the open source AI space but believes there’s a need for better coordination of these initiatives.

Open-source AI model Falcon and its adoption in government and private entities in the UAE

• Falcon is widely adopted with 40 million downloads across various versions. It’s used by government and private entities, especially in the UAE, for citizen services, resource management, healthcare, legal sectors, and beyond.
• Hacid highlights UAE government’s proactive approach to AI, including appointing an AI minister in 2017.

Reducing bias in AI models through open-source development

• Hacid discusses Falcon’s multimodal AI model, combining text, image, and video inputs to reduce bias and improve safety.
• Open-source development fosters diverse perspectives, which helps mitigate biases inherent in data. The Falcon Foundation aims to include multiple viewpoints, promoting a more balanced model.
• Hacid also talks about the importance of reducing bias in AI models, acknowledging the risk of censorship.

Guest: Dr. Hakim Hacid (LinkedIn)
Organization: Technology Innovation Institute (TII)