Interview with Dr Álvaro Tejero Cantero
This is part of a series of interviews with academics. Find out more about Woolf and the academics who are driving research forward at woolf.university.
Dr Álvaro Tejero Cantero is currently a Senior Research Associate in Computational Neuroengineering at the Technical University of Munich, Germany.
Álvaro, you have a background in quantum physics. You also completed a PhD in computational neuroscience in 2012, and you are presently working on machine learning. Would you like to say a few words about your academic path and your current interests?
I studied theoretical physics in Madrid and Paris, and then came to Germany for a Masters degree in fundamental quantum mechanics. I enjoyed the beauty of the theory, but I was more and more interested in the applied mathematics that is used to model sociotechnical systems. It is one of the reasons why, for my PhD, I moved to computational neuroscience. Since the brain contains dazzlingly complex networks of neurons in continuous dynamical interaction, we need to combine techniques from signal processing, network science and dynamical systems theory to approach it. I worked during my PhD on theoretical models that sketch how the hippocampus can replay memories of extended events, and compared the results to the oscillations recorded from real hippocampal tissue. More data analysis at Oxford and the fertile intellectual exchange in the Computer Science and Statistics Departments there progressively drove me to Bayesian machine learning as a structured paradigm to learn about complex systems from data.
I currently enjoy the privilege to do research in this area in the Computational Neuroengineering group of the Technical University of Munich, after stints in industry and academia working on topics ranging from urban mobility through health monitoring, recommendation systems or knowledge graphs. A highlight of doing applied machine learning is indeed the variety of disciplines where one can get involved, and, in passing, I think education represents a particularly fertile ground!
You have been recently involved in projects on Industrial Artificial Intelligence. Could you tell us more about machine learning and how it may relate to the wider challenges in higher education?
Machine learning is a discipline at the intersection of statistics and computer science. It has become useful through a conjunction of advances: in calculation, with substantial speedups harnessing hardware originally designed for graphics; in learning algorithms, most visibly with a concomitant explosion of neural network architectures and reliable ways to train them; in data availability, thanks to the commoditization of sensors, storage and the cloud.
As a natural scientist I am personally most drawn to Bayesian machine learning because it enables us to distill the subtle regularities of the world from observations, and combine this newfound knowledge with our own preexisting knowledge and intuitions, expressed in a quantitative way. This is machine learning for understanding; however most of us will experience it as the result of an engineering process, embedded in products and services.
As an engineering discipline machine learning systems are operating a wide-ranging transformation of society, most prominently by finding structure in images, sounds, and generally, sense data, which can then be fed into normal computer programs — think voice assistants, robotic driving or automatic document reading. Having computers ‘understand’ speech or a satellite image suddenly makes the world accessible to automated processing without us having to excruciatingly type in the data. Computers thus gain eyes and ears, so to speak.
This expansion of the domain of practicable automation will impact education as well as everything else. Think, for example, cognitive enhancements. Wikipedia on mobile is an example of an artificial memory augmentation device without machine intelligence, but one which already affects our relationship to learning and memorization. We can expect further advances in cognitive prosthetics, such as assistants to help discover ideas and facts that are relevant for one’s interests and understanding, and accompanying visualisations that make the complexity digestible. Moving into a more speculative territory, brain implants trained with machine learning could relay our thoughts in a pre-verbal stage to computers, if we surmount the bioengineering challenges.
Attention management and cognitive control will become the central battleground between commercial interests and our own sense of self. It is possible that machine learning systems will support us in this, but it is probably advisable to develop, rather than outsource, our own agency.
More down to earth, an application close to my heart is inferring dependencies between concepts from extant sources, i.e. finding out in which order ideas can best be presented. Contemplating such networks of dependencies one can better understand how the puzzle of a discipline comes together, revise the material in a structured fashion, and avoid the confusion of confronting a topic with a fuzzy notion of the prerequisites. Ultimately, this should bring us truly personalised materials and curricula.
Yet if we concentrate only on the static problem of how to learn what we want, we will be missing the larger point of education which is, of course: why learn? Topics become relevant to us as part of a dialectic, and emotional, process of discovery. I believe that the main role of a teacher is here to foster, guide and nurture the motivation which makes learning possible. This maieutic process is necessarily challenging, critical and empathic, and its reflective dimension (humans thinking humans think) is far beyond current machine capabilities. Perhaps empathising needs commensurateness of experience, a common becoming, the experience of ignorance undone. This is the central reason why I believe in Woolf’s potential — close, intense discussion as the ideal vehicle for transmitting high standards of thought and scholarship; a learning that is performative, selective, and critical.
This critical element is, I think, crucial in the age of artificial intelligence: the transition to ubiquitous AI will have to be politically managed. If we want to stay clear of high-tech dystopia we may find ourselves having to collectively reflect on the wants of an affluent society rather than fight over the individualistic pursuits of a subsistence society. This ethico-political reflection will necessarily draw from classical humanities thinking modes, as it involves fuzziness, historical experience, action-knowledge loops, social-emotional agents and teleological questions. Engaging into criticism of the technical utopia as sublimated by AI, where thought has been, at least initially, mechanised and dispossessed of its idle and poetic qualities, will become more relevant as the activities we used to learn for are progressively taken over by automation.
What opportunities from the Woolf platform most excite you, and why?
While an undergraduate in Physics, long before lecture videos were routinely posted to the web, I started the Alqua project to create and share lecture notes with a copyleft, so that students could free energies from note taking to invest them in understanding. I was therefore delighted when I first learnt about tutorial teaching, which is all about the latter. I believe tutorial teaching to be specifically therapeutic in math: a good tutor can spot knowledge gaps and early frustrations and ensure that the student is back on track before she gives up in despair. Thus, facilitating a critical, personal tuition of the highest caliber to as many students as possible is a very motivating prospect. Woolf can also bring about innovation in teaching: the emergence of scholarly communities, at the same time decentralized and democratic, organised into topical colleges means opportunities to rethink more than just the curriculum.
We can also expect Woolf to impact research: currently, a phenomenal amount of time (and quite some integrity capital) is spent just to acquire funding and in all the attendant administrative and marketing activities. A predictable income for researchers can support original, farther-sighted investigations with a higher failure rate, which is a hallmark of healthy science.
There are some challenges, too, for example: how to provide open access to research literature, how to fund infrastructure for basic research, and more importantly, how to make quality education accessible to many. These are stimulating questions that I look forward to addressing with the ingenuity of the Woolf community.
Learn more about the Woolf project and the people involved on the Woolf website.
