With many projects ongoing at MBZUAI, including development of an efficient AI operating system, metaverse technologies, smart grids, clinical diagnostic tools, smart drug design, the University is making strides in supporting the UAE on targets in the National Strategy for Artificial Intelligence 2031. Timothy Baldwin at MBZUAI gives us his perspective.
One of the major areas where research from MBZUAI is already having an impact, including on sustainability, is LLMs
To ensure the widespread availability of LLMs it is also important to ensure they are available in a variety of languages.
Creating LLMs in English is easier than other languages because the quality datasets needed for training the model are available.
AI has the power to bring immense benefits to humanity, which is exactly why MBZUAI is building the ecosystem of skills, talent and research needed to help it fulfil its potential. Few technologies have stirred as much debate as Artificial Intelligence, AI, which is rarely out of the news thanks to its unmatched potential to bring sweeping, and, I would argue, overwhelmingly positive changes to humanity, if done right.
Yet despite the widespread use of AI algorithms and chatbots in popular e-commerce services, and large language models, LLMs such as ChatGPT and the recently-released Arabic LLMs Jais and Jais Climate, AI remains at a nascent stage, and its real impact is yet to be felt across most areas of industry and society.
I am optimistic that, if developed and deployed in the right way, and if we as a society have the skills and knowledge to use it effectively, AI can help humanity overcome many of the world’s most pressing challenges including expanding the reach and quality of healthcare and education, and the complex issues of sustainability and climate change.
One of the major areas where research from MBZUAI is already having an impact, including on sustainability, is LLMs. MBZUAI, in collaboration with researchers at UC Berkeley, CMU, Stanford, and UC San Diego, developed Vicuna, a sustainable LLM that is both vastly cheaper to train and has a miniscule carbon footprint and requires a tiny fraction of cooling water compared to mainstream models such as ChatGPT.
As we are all seeing, the applications of LLMs are myriad, and with their use growing rapidly, it is imperative to consider their environmental footprint and affordability, because transformative technologies should be within the reach of everyone.
To ensure the widespread availability of LLMs it is also important to ensure they are available in a variety of languages. This will expand their reach and ability to positively impact society and prevent a winner takes all effect whereby the most widely spoken languages, most notably English, gain yet more dominance due to the growing influence of LLMs, while other underrepresented languages, including Arabic with around 500 million speakers, risk losing influence among their community.
Creating LLMs in English is easier than other languages because the quality datasets needed for training the model are more readily available. But we are finding new ways of using data to train LLMs in other languages and supporting the decentralization of the creation process by open-sourcing LLMs.
I am particularly optimistic after our work on Jais, the world’s highest quality Arabic LLM that was developed in partnership with Core42, an arm of Abu Dhabi technology company G42 and Cerebras Systems.
MBZUAI and Core42 jointly worked on Jais’ training, fine-tuning, evaluation, and AI safety features, and the model is now being evaluated across a range of public and private sector organizations in the UAE, including in the energy sector, telecoms, and financial services. This effort potentially has immense value to organizations and individuals who communicate in Arabic.
At the University
Mohamed bin Zayed University of Artificial Intelligence, MBZUAI is focusing on developing an AI ecosystem built around skills, knowledge, and research, to create a pipeline of AI talent with the expertise needed to advance AI for societal good, and R&D that is laser-focused on addressing societal challenges.
We have recently added two new departments, computer science and robotics, each with dedicated MSc and PhD. programmes, to complement our existing computer vision, CV, machine learning, ML, and natural language processing, NLP programs. Robotics is a particularly important discipline as it represents one of the key touchpoints for AI to interact physically with people and the world around us.
We are also looking to launch other programmes that will cater to students with backgrounds in disciplines as diverse as mechanical engineering and biochemistry, but who lack the in-depth mathematical and computer science skills required for the existing programs. The focus here will be on developing the AI engineering skills to implement AI solutions in their chosen field.
We have already seen firsthand the power of introducing AI to leaders in all fields of industry and government in the UAE through our MBZUAI Executive Program, which culminates in capstone projects that design AI solutions to address specific challenges in areas including sustainability.
With many research projects ongoing at MBZUAI, including the development of an efficient AI operating system, metaverse technologies, smart grids, clinical diagnostic support tools, and smart drug design, the University is making strides in supporting the UAE on the ambitious targets in the National Strategy for Artificial Intelligence 2031.
Just as importantly, we are also inculcating in our students and indeed in our thinking as an institution, the importance of considering the impact of our research now and in the future.
This is why ethics is infused in our teaching across all of our disciplines, ensuring all our students have the training to understand the wider complexities and societal impacts of AI, and ensure that research is developed for the good of society from its inception.
Leave a Reply