The era of artificial intelligence is officially upon us. While there are bound to be troughs of disillusionment moving forward as media tries to grapple with what AI is and isn’t, and we wade through the string of spectacular headlines and their moderately interesting accompanying full texts – the fact is that our lives and works are being transformed as we speak.
For the purpose of this article, I will be using the term ‘AI’ as an umbrella term referring to Machine learning, Neural Networks, Deep learning and any other emerging applied research branch fitting into the fluffy ‘AI’ umbrella. However, all use cases discussed in the sections below will pertain to the notion of Narrow AI, i.e. the discussion of Artificial General Intelligence, while deeply fascinating, will be outside the scope of this discussion. We are talking AI based software trained to do one thing and do that thing very well – often far better than a human.
Beyond allowing researchers to work more efficiently with their data sets, whether video and images, text, numbers, or a combination of these, AI technology will also radically change the landscape of scholarly communications. Here are a few clear examples.
Short term: Content management
Managing, navigating and extracting information from a vast corpus is an insurmountable task for a human – and in the age of digital scholarly communication where we count hundreds of millions of research papers, conference proceedings, patents, public data sets, clinical trial report, it is becoming more vital than ever. The silos of research field and to a certain degree geographical location become exuberated in the digital world when we lack tools and ability to venture beyond what we already know. Interdisciplinary research is of massive importance but becomes increasingly difficult.
AI tools already allow us to do amazing things with large collections of documents, and the field of Natural Language Understanding will continue to see major breakthroughs over the next few years. Creating machine-made interdisciplinary indexes and ontologies, going beyond keyword queries to efficiently capture the evolution of a field’s jargon, performing deep corpus analysis to identify relevant information or extract trends, summarizing articles either through concept overview, hypothesis extraction or plain English summaries – all of these core technologies are already here.
The barriers to getting them in your hands is a combination of limited financial incentives of commercial players to enter academia, the inertia embedded in the university system, and the admirable skeptical nature of a lot of researchers – but there are players working hard to bring these tools into your workplace. At iris.ai, we’ve built tools to semi-automate the systematic research landscape mapping, and we’re one of many working to make these tools available to you.
Mid-term: Machine validation
Systems to help us index, navigate and summarize a vast corpus of document will be incredibly helpful for anyone who wants better access to scholarly knowledge, but AI will allow us in a 5-year perspective to go beyond this. Argument mining of a paper is no longer science fiction, which means we will soon be able to automatically distill a paper into its basic arguments, and thus identify all similar or contradictory arguments made by other researchers through time and across fields. Tracking and backtracking these arguments will allow a machine to ‘red flag’ any contradictions, highlight disproven theories and give a probability score to each argument’s validity. Keep in mind this should never be an attempt to label an argument as a binary ‘right’ or ‘wrong’ – and of course considerations such as outliers and paradigm shifting breakthroughs, as well as field specific considerations, needs to be properly accounted for. But such a system, a “knowledge validation engine” of sorts, obviously in combination with human peer reviewers, will allow a much more solid and up to date coherent view of state of the art – based on all the knowledge we have.
Long term convergence
What is important to remember is that AI is not a standalone technology. It is fundamentally based on and enabled by advances in computing power and digital access to data, and we will see beautiful convergence with a variety of other fields such as quantum computing. For scholarly communication though, I would highlight the convergence between AI and technologies such as blockchain. While blockchain as a field of technology and a community has a long way to go, and we will have to see either fundamental technological changes or even full shifts into other approaches, the idea and practical implementation of a decentralized, immutable network holds very interesting promise for scholarly communications.
It can allow us to build a system where everyone who impacts the world of science have their contributions immutably tracked and remunerated (whether it is publishing knowledge, peer reviewing others’ knowledge, sharing their data or building useful tools) and everyone who leverages the common knowledge contributing value back to the ecosystem (disrupting the current subscription models, bringing corporate research and development better into the mix), and this whole body of scientific knowledge validated and managed by an AI system in collaboration with the human contributors – and have this entire ecosystem be co-owned by all contributors with mechanisms for keeping large players at bay using ownership caps. (To see a prototype of this vision, download a white paper from projectaiur.com)
And when a system like this is in place, where contributions are immutably tracked and remunerated, I personally think that the research paper of the future will not in any way be shaped like the research papers of today, but as a collection of collaborative pieces of knowledge, never static but in continuous development as we learn more about the world – as such reflecting the true nature of scholarly knowledge.
The convergence of AI and blockchain could create an entire paradigm shift and revolution in scholarly communication. It is not the technology that is the limitation in a 10-year perspective: it is the current incentive and commercial structure of scholarly publishing and of course how to organize the millions and millions of bright, driven, stubborn and brilliant researchers holding the keys to the solutions of the biggest problems we’re facing as a human species.