With all the talk in HE of AI and plagiarism, little attention has been given to teaching, learning or research. I have outlined an idea for an AI University, with a relentless focus on AI to optimise teaching and learning, to increase access and reduce costs.
As a corollary, let us look at ‘research’. Can an AI University be a focus to optimise research? It has already happened over the last two decades. The real question is not whether AI can generate research, it is whether it can halt the paper mill it has become. It can, of course, increase productivity but its real role may be in streamlining and producing better research.
Future of research
Reports are tumbling out of institutions telling us that AI simply augments human effort. This is simplistic, so obviously defensive that it is in danger of turning into a pearl clutching meme.
In truth:
AI has been accelerating research for decades
AI will automate many research tasks
AI will eliminate some human research
AI will create new research opportunities
AI will produce big wins, leapfrogging existing research efforts
With cheap intelligence-on-demand, AI has already sped up the process of research across a broad front and promises a lot more.
Wave 1: Existing impact
We have already seen this happen in the first wave, with search, citation search and access to Journals and knowledge at our fingertips, in seconds. The impact of AI on ‘research’ may be much greater than it is on teaching and learning. It has gone under the radar but it the impact has already been astonishing. Google Search and Google Scholar accelerated research, as the whole business of access to Journals, knowledge and ready to use citations, slashed time to completion. I remember waking mile after mile along library shelves to find Journals and books, hour and hours labouring over desk research and citations (the bane of any researcher back then), even the typing of dissertations and theses with zero cut and paste! All have been accelerated or automated by AI. One could argue that every PhD period should have been slashed by several months!
Wave 2: Automating tasks
A second wave is now hitting research, as tasks are being augmented and automated, with a slew of research papers showing that this is not only possible by happening.
Let’s look at tasks the literature has already identified as being possible, productively, by AI:
• Titles
• Abstracts
• Introductions
• Literature reviews
• Research gaps
• Hypotheses
• Scoring/Open text surveys
• Research plans
• Data analysis
• New concepts
• Idea generation
• Full paper outlines
• Full papers
• Faster peer review
Wave 3: Optimise research
A third wave may emerge where research is being eliminated on scale, especially text based speculation, meta studies and data-driven research. There is every reason to believe that AI will create new research opportunities and eliminate entire areas of current research endeavour. One would also hope that is speeds up peer critiques and reviews.
One would hope that, rather than making the current paper mill go faster, we can optimise research by eliminating bad or weak research. The system is in need of a rest on quality and quantity. Far too much research is being published and not used or even read. AI could help solve this problem.
Wave 4: Big wins
A fourth wave will leapfrog research to produce enormous wins. This has already happened and will accelerate. In some specific areas the impact of AI has leapfrogged the existing paradigm.
• Alphafold: 1 billion years of research saved
• MIT: Super-antibiotic Halicin discovered
• Materials discovery of millions of new structures by Deepmind
Demis Hassabis says, one protein was taking 4-5 years to identify its 3D structure. We now know all 200 million, saving 1 billion YEARS of research. Read that again – it is an astonishing figure. This was only the start as we’ve also seen a similar breakthrough by Google on materials science, and a groundbreaking drug discovery by MIT.
Do we think this will stop? I don’t think so.
AI University and research
One radical idea is to break the link between research and teaching, to recognise that this is part of the problem, the old Humboldtian idea that researchers make good teachers and that it is a necessary condition for success. I have argued that breaking this old Prussian bond could be a good thing, on the basis of research showing that researchers as teachers often hinder rather than help teaching in Universities.
What is needed is a University that grabs the bull by the horns and becomes a centre of research itself, or how research can be accelerated to release the bounty that AI offers or engages in research that uses AAI to accelerate progress. Research is, after all, a means to an end, not an end in itself. There is always a purpose or goal, even in blue sky physics, we have a general understanding that understanding the nature of reality has benefits for us as a species and often lays the ground work for new research.
There has to be a recognition that we are returning to a successful model that emerged in the Post-War period, with some notable examples before, (Bell, Eddison, Xerox, Tesla etc) , where large tech companies are doing their own research and leaping ahead of academia. This is not lone wolf activity, as they rely on academic output in terms of talent. However, their speed, ability to take risks, access to data, ability to deliver on scale and competitive mindset means that things happen quicker with more substantial breakthroughs.
Post-war success in the US in particular was engineered by Vannevar bush to co-ordinate academia, Government and business, paying to the strengths of all three. This is what ended the war with the Manhattan project and got them to the moon. All three need an element of humility to co-operate and make human advances to solve big ticket problems, such as climate change and the energy crisis.
It is no accident that almost all the breakthroughs are taking pace in the US, perhaps the Anglosphere of one includes Deepmind. There seems to be something in this idea that the optimal way forward is collaboration between state, academia and commerce, rather than anti-corporate, anti-government or anti-academic sentiment being allowed to hinder progress, as that is clearly sub-optimal.
Conclusion
There is another more serious problem here. Imagine a worldwide system, that trusts institutions to review and produce objective results, that gets billions in public funding, is found to be so corrupt that it has created incentives so bad that it is filling up with fake output, 10,000 fake entities have been found but we know this is the tip of the iceberg, with people buying and selling favours and a major country gaming the system by flooding it with false output. We have already moved towards this position.Bibliography
Aydın Ö., Karaarslan E. OpenAI ChatGPT generated literature review: Digital twin in healthcare (2022)
Chen Y., Eger S. Generating (humourous) titles from scientific abstracts end-to-end (2022)
Wenzlaff K., Spaeth S. Smarter than humans? (2022)
Dowling M ChatGPT for (Finance) research: The Bananarama Conjecture (2023)
Zhai X. ChatGPT user experience: Implications for education (2022)
Adesso G. GPT4: The ultimate brain. Authorea Preprints (2022)
Valavanidis, A., SCIENTIFIC REVIEWS Artificial Intelligence Application withMachine-learning Algorithm Identified a Powerful Broad-Spectrum Antibiotic.
Merchant, A., Batzner, S., Schoenholz, S.S., Aykol, M., Cheon, G. and Cubuk, E.D., 2023. Scaling deep learning for materials discovery. Nature, pp.1-6.
No comments:
Post a Comment