From Academia to Industry Research
Ever since I was little, I have been captivated by the big questions related to how the world works and our purpose within it. I’ve always had a broad range of interests and appreciated a holistic view over a seemingly fragmented world of disciplines. At university, I studied a bit of literature, linguistics, philosophy, psychology, behavioural economics and a lot of maths and computer science. Finally, I discovered the field of Machine Learning or “Artificial Intelligence” (AI), which was just an emerging academic discipline back then. With the power of statistics and computational resources, I felt like there was immense potential for understanding how our minds and societies work, via analysing language and vision in the plethora of textual and visual data humans produce and upload to the internet each day.
I ended up doing a PhD in Multimodal Natural Language Processing. It is an area of Machine Learning, where the goal is to build models which understand language in a similar way as humans do: by grounding meaning in the context of human perceptual information, such as vision and hearing. In other words: teaching the computer to understand words and sentences by associating language to images, videos, sounds etc.
Thanks to some major technological breakthroughs, while I was doing my PhD, the field of AI exploded. This has some momentous consequences.
Firstly, the centre of gravity of AI research shifted from academia towards industry. At first, it was mostly visible in the increasing presence of big tech companies at conferences, then trickled into smaller companies and startups. The recent boom of large foundation models has led to another tectonic shift, causing the gap between academic AI research and research in industry to further widen. This is mostly due to the discrepancy related to financial and computational resources.
Secondly, with AI becoming one of the hottest industries, it wasn’t long before the market saw an influx of disruptive applications. From a quirky academic field, AI quickly turned into a technology with huge immediate and long term impact on our society. Along with it came a variety of questionable applications and unintended societal consequences.
Like many PhD students, I also encountered an existential crisis during the course of the programme. During this period, and considering the aforementioned climate, I realised that the next best direction for me had to involve applying my skills and knowledge in AI to something that positively impacted society and/or could mitigate already existing harms. I was seeking a place where I could make a direct impact on the masses, who use and are affected by the internet and AI on a daily basis. After some searching, I found Unitary. While their mission strongly resonated with my goals, their methods of multimodal video content understanding matched my expertise.
Interestingly, concentrating on the tangible goal of detecting harmful content in videos has a side effect, resonating with my younger self’s interest in analysing meaning and society. By having access to large amounts of social media data, and having the capabilities to run and analyse models on them, we can, and have to learn all sorts of information on what trends surface on social media. In fact, it is crucial for us to discover social phenomena and biases which may lurk in the data and in the models.
Don’t get me wrong, I believe that independent academic research is essential. In the field of AI and at this moment in time it is arguably more crucial than ever (as many have pointed out at the Exploring Foundation Models Symposium this year). In industry, there are all sorts of market incentives, but there are also people with an honest desire to make a positive impact. We also need them more than ever. When it comes to technological change and society, policies and incentive structures matter as much as individuals, if not more. Currently, if we want to understand and influence such structures in AI, we need to be in industry.
As for me, I am trying to paint my own bits of this picture and be part of responsible development even if one can often get lost among all its contradictions.
Please comment and like under the original post on Medum.
