Generative AI is having a big impact on the way academics conduct research and related activities. In this section, we will explore how researchers can use genAI through three different lenses: the utilitarian lens, the critical lens, and the reflexive lens.
The Utilitarian Lens
If you view genAI simply as a set of tools to use to help you in the pursuit of your day-to-day work, then the question that this section tries to answer is:
What are the things I need to know to make an informed choice?
One of the first things to understand is the wide variety of genAI models available to you and the different costs associated with them. There are now hundreds (if not thousands) of different genAI products available, but generally, in the background, they will be only a few genAI models powering these products.
When it comes to the latest cutting-edge models, there are four genAIs that dominate the market:
All four of these companies offer a tiered structure for access.
Typically, there is a free tier that allows limited access to a slightly older, less powerful model. A subscription of around £20 per month will buy you access to more powerful, advanced models with a higher use limit than the free tier, and then sometimes there is a super expensive tier that grants access to the most advanced model and larger usage limits.
The key thing to be aware of is that the difference between the genAI model available at the free tier and subscription tier is significant. For many research-based activities, you would notice a difference in the quality of response you receive from these different models. For any task that involves data analysis, coding, or idea formation, you will likely see much better results from advanced models such as ChatGPT 5.2 thinking, Google Gemini 3 Pro, or Claude 4.5 Opus.
A major consideration, however, with any of these models is what is happening with the data you upload. Broadly, if you sign up to the free tier of any of these models, you will have agreed to allow these technology companies to use your data to train future models. While it is sometimes possible to turn this off, it is quite often buried in the settings. If you have a paid subscription, the picture is more complicated, and it will depend on the exact type of subscription you have. For example, here at Durham, we provide all staff and students with Microsoft Copilot (which is powered by various OpenAI ChatGPT models), but this comes with an enterprise licence, which means Microsoft and OpenAI do not use your data for training.
An alternative to the big four models would be to use an open-weight model such as:
The developers of these models have all publicly released the parameters of their models. These parameters or weights are the internal values that determine how a model responds to an individual input. Generally, with an open-weight model, you can download it, run it locally (if your machine has sufficient memory and processing power), and fine-tune it for your own applications.
This additional freedom to modify your models and keep your data private can make these open-weight models very appealing to researchers, although there is a higher technical barrier to entry.
To run a model on your own laptop you can use a piece of software such as LM Studio (available to all Durham staff and students on AppsAnywhere). This allows you to find and download models and run them locally (offline if necessary) on your own computer.
The Critical Lens
How we critically engage with genAI can take many different approaches, from ethical concerns over how AI has been trained and developed to environmental concerns over its impact, to how we evaluate the responses that genAI produces. The continued rollout and the expanse of genAI into everyday life do mean that researchers will need to adopt considered positions about their use of genAI. In this section, we will address some of these concerns and the ongoing debates.
How GenAIs are trained
GenAI models are trained on data, but we don’t know what data exactly since no tech company releases this information. Most likely, it is some snapshot of the internet around a set of data and a lot of uploaded material from books, journals, and music tracks. Some tech companies have used pirated materials to train their genAI, which may have violated an author’s intellectual property rights (this law in this area is not clear, and there are ongoing cases in both the USA and the UK); however, this does give some researchers pause when it comes to using genAI because it is difficult to say if a genAI has been created in a moral and ethical way.
The data that a genAI has been trained upon can also lead to biases in the answers that it provides, and this should be held in mind when checking genAI responses.
The environmental cost
The exact environmental impact of GenAI is difficult to quantify because the major AI companies do not release this information on a regular basis, although it is probably safe to say that cumulative use of genAI does consume a significant amount of resources (electricity and water). There are several individual actions that you can undertake to reduce the impact of your own AI use:
1. Consider your usage
As the line between genAI and search engines merges, we have seen an increase in queries put to genAI that could be answered or completed in different non AI ways. For example, you could ask a genAI to convert between a reference style in Harvard to APA, or as an alternative, you could use an online site such as citethemright or, if you use a reference management software, you could do it there.
2. Switch to a lightweight model
As mentioned previously, there are many different types of models available for you to use, but generally, the “faster” (ChatGPT 5 instant, Gemini 3 Fast, Claude Sonnet) use less resources such as energy and water, and while for some research tasks you will want to use a more advanced model, for a lot of everyday tasks these fast models will perform an adequate job.
3. Use a standalone GenAI model on your computer
If you use something like LM Studio, then you can run a genAI off your own machine, and while these models will have used a lot of resources to train, the ongoing resource cost to use them is simply the energy it takes to run your own machine.
The Reflexive Lens
The impact of genAI on the skills necessary for successful and effective researchers is a matter of some debate, but there is no doubt that genAI is changing the way that research is conducted. A challenge all researchers will face in the near future is reflecting upon their own skill set and asking themselves what personal attributes they need to develop to be successful in their own field of interest.
This is not just confined to roles in academia but also non-academic jobs; for example, in both of these fields, we are now seeing changes in the recruitment process due to genAI. The traditional recruitment model has relied on an application sift prior to an interview; however, nowadays, the number of high-quality applications has risen due to the use of genAI in their preparation. At the other end, many recruitment campaigns would have found a single interview enough to decide on a candidate, but increasingly, organisations are expanding this process to include multiple interviews and presentations so that they can get a better sense of a candidate in an environment where they are confident that AI has not been used. As a result, being able to perform in these types of situations will be an even more valuable skill than it has previously been.
This is just one example where genAI is changing the workplace by increasing the value we associate with the ability to defend our ideas verbally, but at the other end of the scale, the ability to use genAI well in your role will also likely increase in value. So the question for the researcher is how do I develop these skills?