A report was recently issued by Stanford, the “AI Index 2021 Annual Report”[1] that, in its own words, aims to be the most credible and authoritative source for data and insights about AI in the world.  I was struck by point 6 of the top 9 takeaways: “The majority of the US AI PhD grads are from abroad—and they’re staying in the US”, mostly absorbed by industry.  According to the report, the UK offers almost as many AI-specialised master’s programs, and more specialised bachelors programs, as the entire EU27, and it seems likely that there is a brain drain from the UK to US industry.  As the report notes, the US has also suffered an unprecedented brain drain to industry over the past decade or two, driven by the resources and salaries available, and this has a chilling effect on AI start-ups and their success [2].

A transfer of knowledge to industry is a good thing:  One purpose of a university is to benefit society by disseminating knowledge, and one way this can be done is by commercialising research.  Universities and research institutions are a dynamo for entrepreneurship, and for innovation in large companies.  This is especially true of machine learning where core new techniques are continually being developed – for example the amazing capabilities of large language models such as BERT and GPT3 have their genesis in a paper on transformers published in 2017 [3]; and similar techniques are now being applied to image processing and other domains.  Not uncommonly some of the best people manage to combine academia with positions in industry, and the UK punches above its weight in AI, but to avoid a brain drain universities need to retain and train the next generation of world class researchers.

One way to retain the best people is to pursue inspirational goals.  Artificial General Intelligence (AGI) is one; and if it is difficult that makes its pursuit all the more worthwhile.  Interestingly some of the biggest names in the machine learning field, for example Geoffrey Hinton, Yann LeCun, Yoshua Bengio, and others, appear motivated by a desire to understand how the human brain works, and to emulate it.  Universities and research institutes are well-positioned to take a cross-disciplinary approach, collaborating with neuroscientists for cross-fertilization of both fields [4].  Breakthroughs could emerge from more collaboration, for example providing conceptual insights and perhaps even new ways to think about the “easy problems” of consciousness, and possibly even the hard problem of consciousness.  On the more applied side developing new techniques is crucial, and collaboration between universities and industry might be a way to obtain more of the compute resource that machine learning needs.

I read a lot of machine learning papers as part of my job as a patent attorney, and my impression is that a relatively small number of institutions contribute to many of the significant advances.  The papers I read are about new techniques with potential real world applications, and the various groups that the work is from are a source of people with the skills and experience to practically implement advanced new models.  Such people help to solve some of the world’s greatest challenges, including in medicine, sustainable technology, and agriculture, and also seed the entrepreneurial community by founding spin-offs.  Perhaps one way universities can help – as some are already doing – is by pursuing inspirational, cross-disciplinary research, for example towards AGI, to educate and retain the best people for the future.

[1] arXiv:2103.06312 Zhang et al.

[2] Gofman and Jin, “Artificial Intelligence, Education, and Entrepreneurship” October 26, 2020, https://ssrn.com/abstract=3449440

[3] arXiv:1706.03762 Vaswani et al.

[4] e.g. “Neuroscience-Inspired Artificial Intelligence”, Hassabis et al.  http://dx.doi.org/10.1016/j.neuron.2017.06.011