In traditional academia-industry collaboration in research and development, the academy substantially carries out the fundamental and applied research, while industry largely engages in commercialisation.
For artificial intelligence, we’re seeing a notable departure from this state of affairs. AI companies are successfully engaging in both research and commercialisation at a rate that universities cannot compete with.
This is attributed to three main factors which are key to training AI foundational models: AI experts, compute power and data. Universities do not currently hold this power, and are not likely to be able to compete in this area unless there are significant investments.
The huge investment by industry in these three resources has seen many AI faculty members move from universities to the private sector. This was echoed by a UCL AI expert in a recent roundtable discussion, who attested to the increasing difficulty nowadays into hiring and retaining AI-competent people in the university.
As suggested by this article in Science, as of 2004, about 21 per cent of AI experts with PhDs moved into industry – however by 2020 close to 70 per cent were reported to have moved, which is far greater than in other fields.
The upside
However, it is not all good news for those AI experts moving into industry as they lack the freedom of research. Because of the profit-driven nature of private companies, they tend to work on projects that are focused on short- and medium-term gain. Industry research in AI has also faced accusations of being less diverse and less reproducible, when compared to the robustness and diversity embedded in academic research.
This suggests that there are opportunities for universities in artificial intelligence research, despite the lack of financial firepower.
For one thing, as suggested above, universities can take advantage of greater academic freedoms to explore new AI methodologies, with an emphasis on long-term public benefits. This can also encompass innovative or unconventional ideas. In a recent roundtable, an AI technical expert at UCL shared industry leaders’ reflections on what the role could be for the universities in such an industry-dominated sector – the response was that industry does not have the luxury of time to reflect on and pursue long-term AI research.
Interdisciplinarity is another area where universities have an advantage. Being equipped with expertise in a wide range of disciplines is crucial for the sustainable and responsible development of AI. AI itself is inherently interdisciplinary and requires knowledge from computer science, mathematics, ethics, sociology, law, philosophy and several other fields. In teaching as well, universities are well-placed to focus not only on technical AI skills but also address a broader public interest such as the ethical, legal, and social implications of AI.
Universities have the opportunity to lead by example in promoting the ethical development of AI. The adoption of clear ethical guidelines for AI research, and support for transparency, accountability and auditability, can promote the idea that AI technologies must be designed and used in ways that benefit public interest.
There’s also an important policy advocacy role here. Higher education institutions ought to champion responsible AI to inform public policy. Research and continuous engagement with policymakers, stakeholders in the universities and the general public could help universities shape regulations for responsible AI technology development and adoption, away from industrial lobbyist influence. This would enhance public engagement and raise awareness to foster a more citizenry engagement.
This all has a consequence for partnerships as well. Universities need industry resources to upskill their students and update the research endeavours of their faculty according to market demand – while industry requires long-term foresight, public interest considerations and new research methodologies, with all the rigour and reproducibility that universities bring. Higher education institutions should seek to engage in partnerships that complement efforts towards the development of responsible AI.
Beyond competition
Despite the challenges of resourcing, staffing and access to compute power that universities face when “competing” with industry in artificial intelligence research, it is clear that there is space for synergies and collaboration between publicly funded research and private industrial enterprise – to harness the power of AI for public good.
The onus is now on universities to be bold in developing forward-thinking policies and opportunities to redefine their role in this evolving landscape, and leverage their unique position to influence AI’s governance and trajectory. By prioritising ethical development, fostering interdisciplinary collaboration, and engaging with the public and policymakers, universities can ensure that AI technologies benefit society.
There may yet still be light at the end of the tunnel. Apple’s on-device AI models (which use ~256mb of RAM) do not require a server for many of their functions and show that smaller, more localised applications may be more beneficial to society. I’d be amazed if the Raspberry Pi foundation don’t launch their own localised, tinkerable LLM in future for example.
The private sector might have the investment at the minute but they are burning cash rather than making any. No model has proven to be profitable yet and without returns, that investment will dry up. I can see how commercial entities might stump for a $200/month subscription to ChatGPT for advanced work but this isn’t a market of scale. If Apple’s ideas catch on the everyday citizen will question why they need to ever pay for AI when their phone does it for free as just another OS feature.
And if the public won’t pay for it, then they’ll never make any money.