Open Source thought rummaging
Microsoft has been tearing through the AI world in the last few months, in part helping it to acquire the coveted spot of being the most valuable company in the world. Having OpenAI, the creator of ChatGPT in its back pocket, investing in Mistral a large Open-Source AI model provider, and luring the team and founders of Inflection AI (1) to a newly created unit just last week, shows that their corporate tentacles are actively at work. And if one thinks that independent academia is safe in its pursuit for independent novelty, look again. One day before Inflection’s cofounders jumped ship, the University of Sydney revealed that they are just as easy of a meal, in a newly released “memorandum of understanding” (2) between the two organizations.
In general, one would laud collaboration. Yet there are two strong cases to be made for an independent, alternative path and not leaping into Microsoft’s arms.
First, the relationship is heavily lopsided, not necessarily in USyd’s favor. While the wording of the memorandum remains vague, with the hope to ‘Improve teaching…, advance knowledge…, and explore ways to leverage AI’. No clear roadmap, investment, or program has been outlined. One could add that they plan to ‘solve AI’ and leave none of us any wiser. In return, Microsoft would most likely gain additional access to important student data and datasets, free graduate manpower (not that they would need any with the spate of recent acquisitions), and most importantly, a view and say over academic research and projects. Fei-Fei Li, an important AI researcher, only this month (3) warned that the large costs of running these new large AI models mean that too many researchers opt to collaborate with large tech corporations to test out their theories; harming the delicate balance of AI inquiry and researchers’ independence. A little over a week later, Usyd and Microsoft announced their partnership
And second, the proliferation of open-source AI foundation models and techniques shows that one does not have to be fully reliant on large companies for novel research. These new techniques and models such as Llama 3, allow for cheaper and easier experimentation for specific, niche projects. Collaboration instead with smaller model makers, Australian startups, and other (Sydney-based) universities, diffuses more of this innovation locally and does not make us overly reliant on one company. It also creates a homegrown and diverse AI environment. Imagine more Canva and Atlassian, and less Microsoft. Diffusion allows for less of a stranglehold over the objectivity and trustworthiness of the research output.
With an annual budget of over 2.6 billion AUD and an endowment of almost double, Usyd has the money, researchers, and capabilities to forge its own path. In the uncertain future of AI, it would be wise for it to spread its bets even if only to show that they seriously care about the future of independent and trustworthy research.
Sources