Suggestions
Cameron Neylon
Professor of Research Communications at Curtin University
Cameron Neylon is a renowned freelance researcher, consultant, and a strong advocate for open research, focusing on open access, open data, and open source. With a PhD in Chemistry from the Australian National University and a BSc (Hons) in Biochemistry from the University of Western Australia, Cameron has a vast experience spanning across various organizations. Some of his notable positions include being a Professor of Research Communications at Curtin University, former Advocacy Director at Public Library of Science, co-author of the Altmetrics manifesto, and being named a SPARC Innovator in July 2010 for his work on the Panton Principles for Open Data. Cameron Neylon writes and speaks regularly on open data and technology development, advocating for the usage of the internet in research to bring about transparency and collaborations.
Highlights
I've had a couple of fascinating experiences with AI over the past week. All of them have reinforced my belief that while AI systems are getting more plausible, somewhat more accurate, and becoming more dangerous in terms of how they play into human lack of attention and bandwidth.
In the specific case I'm thinking of, we ran a workshop. As is becoming common, someone sent a bot along. It created a perfectly useable transcript, and a serviceable summary.
The summary was fascinating. It was entirely accurate. At any given point it provided a succinct and precise description of who said what, what the general topic was, and what we being discussed. Some fairly important mistakes but ones that someone familiar with the topic would quickly pick up.
It also completely missed the main points and central threads that came out of the workshop.
There is no question that these systems will get better at these kinds of problems. Part of the issue here is the number of tokens the system needs to hold to track a thread through a whole conversation. Just a question of how many countries you want to burn down for the energy to support that.
But as they get more plausible, the stakes in play for what they miss will get higher, and as we forget how to manage critical things (like maybe rolling back a computer update, or reverting to physical media for an OS to pick an example with currency) or simply become too busy to engage with those meetings that really shouldn't be an email we're going to start seeing failures that will seem disconnected.
Not so long ago I gave up opening a bank account because my application hit a dead end in an email processing system in an account no-one was checking. After three weeks of fruitless phone calls ("you'll get an email with next steps in the next ten minutes") I just changed banks to one with a functional application system.
With core systems creaking under the strain across society, the temptation to replace them with plausible automated systems will only rise. From systems where noone understand the whole of it, to ones where noone understands, can correct, or sense-check even a part of it, it's going to get a whole lot worse before it gets better...
In a few minutes Invest in Open Infrastructure is hosting the second in a series of community conversations focusing on the State of Open Infrastructure report. Alongside Gail Steinhart (IOI), my colleague Karl Huang (COKI), and John Mohr (MacArthur Foundation) I'll be discussing what we found on funding, the quality of information and the limits of what we can tell, as well as what is needed for the future (more research! better metadata!).
I think you can still register for the event on July 17, 2024, at 1 pm UTC.
