Generative AI and its applications for your business

We sat down with two of our artificial intelligence (AI) experts to discuss generative AI and its applications which range from supporting UK businesses to achieving a low-carbon future through fusion technologies. 

Generative AI is a type of machine learning is a tool that processes data in a way that is inspired by the human brain, to generate new content or environments. It uses a number of techniques like AI foundation models, which are trained on datasets to develop an AI model that can generate text, images or simulations based on word prompts.  

Rob Firth is one of our AI experts who works with text and image-based generative AI to help support UK businesses.  

Thank you Rob for talking with us, can you explain what is generative AI?

Generative AI for text creation is built using large language models (LLM), some commonly known services for this type of generative AI being ChatGPT, Claude and Bard. LLMs are built on neural networks using transformers, a type of computing architecture. Transformers are trained on large text datasets so they can simulate complex language patterns to generate text. This allows them to seemingly create new pieces of work based off short suggestions. It is like you are given the suggestion to make dinner and you make a salad, it may seem like a new salad, but it is based on all the other salad recipes you have ever read or salads you have ever eaten. Something similar happens with generative AI, the content it produces is based on the dataset it is trained on. This can be really useful as you can develop models which can quickly and accurately summarise or categorise large quantities of data. Currently, the Hartree Centre is working with Collaborative Conveyancing to develop an encoder-only generative AI model to help categorise emails and draft emails responses to reduce the administrative burden and speed up the conveyancing process. 

That sounds very interesting, is it the same sort of process when you use text-to-image AI generation? 

Similarly, to create images with generative AI you can use a text suggestion fed into an AI model, in this case an image diffusion model which uses probability to simulate images. Models like this can be used to augment existing images or to create simulations that can model different environments. Generative AI can be used for a range of applications from enhancing or repairing damaged photos or used to develop more accurate weather prediction models. AI is already being used to increase the accuracy of weather forecasting, a necessity with the growing effects of climate change. At the Hartree Centre, we worked to create a digital twin as a part of the Climate Resilience Demonstrator (CReDo) project, which helped to optimise the speed at which AI weather models could be run. Generative AI could help increase the accuracy of weather simulations by processing large quantities of data.  

It is important to recognise that when using models like generative AI on sensitive data, you need to develop the models with clear consideration of their training data. At the Hartree Centre when we are developing models like LLMs, in most cases we are working in conjunction with a business and their own datasets. Our specialists can then help to process and clean the data to improve the accuracy of the model. As a part of the BridgeAI programme we are helping organisations of any size to explore how different types of AI can support their business needs, such as using image diffusion models to help generate a virtual environment or to support an augmented virtual reality. 

Jony Castagna is one of our AI experts working to generate realistic environments to simulate plasma flow to help develop fusion energy. 

Jony, you work with AI to help simulate environments, can you explain how that works? 

Generative Adversarial Networks (GANs) are made of two neural networks trained in an adversarial way to each other until the outcome becomes so realistic that the computer cannot tell the difference between the model and reality. The simulated environment can then be used to run experiments which are too dangerous, costly or are not yet possible to perform practically. In our ongoing collaboration with the UK Atomic Energy Authority (UKAEA) we are developing virtual fusion energy technologies. We are using GANs to speed up simulations of plasma flow to model turbulence for nuclear fusion to increase modelling accuracy, helping to make fusion energy a reality.  

Jony and Rob, what do you see as the future of Generative AI? 

Rob – Access to generative AI tools is readily available, you can go online and experiment on a range of programmes to see how they could work for you. However, at the moment for them to be fully effective and useful for specific business needs, they need to be trained with the right datasets to tackle the distinct challenges an organisation is facing. They can give your business a competitive edge if they are integrated properly. 

Jony – I agree that as a tool generative AI can be used to help support, simplify and speed up an organisation’s work. This is why models like GANs are so important, they are removing the need for physical prototypes and helping to design a functional nuclear fusion reactor that could produce commercial fusion energy, a renewable clean energy source. 

Generative AI can support your business to grow by harnessing the power of LLMs to drive innovation or removing the need for physical prototypes. In our BridgeAI programme you can explore how you can use generative AI and AI technologies for your business.