Step aside Gen-Z, Gen-AI is here! 

From our digital edition, our regular column CompSci Cat discussed the ethical challenges of AI and Generative AI in research.

One of the many hats I wear at work is the ethics hat. I sit on the ethics committee for our faculty, which means I review students’ and colleagues’ ethics applications for their research. Since I am within the Faculty of Engineering and Physical Sciences, I rarely interact with the more complex applications that may arise in medicine or the biosciences. Most of the ones that cross my desk involve surveys, focus groups, and testing new software. 

For the most part, aside from checking that nothing strange or immoral is happening and that any risks have been carefully considered, my assessments typically focus on data. I check that people have thought carefully about what data they are collecting (do you really need their inside leg measurement to understand how they feel about your new piece of software?), how they are storing it (no, a memory stick on your desk labelled “juicy research results” isn’t appropriate!), whether they are anonymising data appropriately (pretty sure I can tell who S. Kearman-Panza, who works in Chemistry, is…), and that there is no obvious case for subverting their research with ill intent. 

However, with the rise of Artificial Intelligence (AI), Machine Learning (ML), and the way that Large Language Models (LLMs) and Generative AI tools seem to have taken over our lives, there have been discussions about forming a new Ethics for AI Group to handle the inevitable influx of research applications involving these technologies. 

This is, of course, a very sensible approach, and it got me thinking about how much has changed technologically, even in a relatively short space of time. We now have a whole new host of things to consider. 

Ethics for AI isn’t a new concept. During my time running the AI4SD Network, we ran many workshops in this area, and I led a working group on Ethics for AI in Food. Our research considered all the standard areas of Ethics for AI: data bias, transparency, explainable AI, responsible AI, considering subversion with ill intent, ethical tradeoffs, etc. These are still very valid overarching concerns. However, I think the biggest change with LLMs and Generative AI is that it is now becoming increasingly common for content of all kinds not to be human-generated. 

A wise wizard once said, “Never trust anything that can think for itself if you can’t see where it keeps its brain.” Yet, if we aren’t careful, we may do just that on a meta level. 

Cast your minds back to 2007, when Apple released its first iPhone, and Steve Jobs introduced a device that was not only a phone but also an iPod and a gateway to the World Wide Web. (Incidentally, a first-gen model of this iPhone is now on display in the Science Museum in Milan—this made me feel old.) Three years later, they introduced Siri, the voice-activated virtual assistant who was meant to help us with all our questions. This caused a flurry of excitement, though it was somewhat short-lived. Ultimately, Siri was a fun toy but frequently misunderstood commands and wasn’t perceived as very “intelligent.” Fast forward to today, and we have tools that will write essays for us, answer complex questions, retain context, write code, and draw all manner of things. They can be applied to many tasks. 

From a tech nerd perspective, this is very exciting. However, for someone who spends a lot of time thinking about the consequences of technology and the subversion of its usage (whether intentionally or not), this is actually somewhat terrifying. On the one hand, no, I’m still not worried that the AI overlords are going to come knocking, but these technologies have improved in leaps and bounds, and several things do concern me. 

Firstly, there is a concern that we may be losing certain skills, creativity, and the ability to think for ourselves by relying too heavily on these tools. After all, if your code IDE points out your mistakes and predicts what you want to do next, are you really learning how to code and troubleshoot? If you have a tool that provides paper summaries and essay suggestions, will you lose the ability to critically search for and evaluate literature? Will you lose the capability to write comprehensively yourself? Is this a wonderful time-saving tool that frees us up for greater things, or something that will ultimately make us too reliant on technology, leaving us unable to do things without help? 

Secondly, there is an educational concern. How are teachers and lecturers meant to evaluate your work when potentially everyone may be using these technologies to do most of it? 

But a bigger concern to me is that if we all use these tools too much, then anything they produce becomes just another source to feed back in. For example, if journal papers are written by AI tools and reviewers use these same tools to provide their reviews, where is the human intervention? These papers may then get read by AI tools and sent back as part of summaries produced for new papers. 

More worryingly, what if AI tools are used to write machine learning code? This code could then end up underpinning systems that make decisions for humans. The idea of using AI to interpret medical results or drive cars already presents many ethical conundrums around trust, transparency, and the lack of human intervention. Surely, this would be even more concerning if there were less human involvement in creating the code underpinning them. 

I am not trying to scaremonger or suggest that these tools aren’t fabulous advancements. I have had great fun playing with some of the drawing tools to create marvellous cat images, and Co-Pilot was very useful when I used it to search for the perfect cheese pairings for a cheese and wine night I was putting on. However, I do think we need to exercise caution about how and where we use them. They started as another Siri—a fun toy for people to play with—but even in a relatively short space of time, they are becoming very commonplace at university, at work, in coding applications, etc. This message is more important than ever. 

We must keep humans in the loop! By all means, enjoy the benefits of these tools, but don’t let them completely replace human intervention. Instead, use them to complement it. Our best future is when human and machine can work together, not when machines eclipse humanity. 

All ethics application examples have been entirely made up for hyperbole purposes.