Tag Archives: innovation

Beyond language with stochastic parrots

Decorative image of a summer flowerSome months ago, I wrote in unflattering terms about artificial intelligence applications (AI apps) and large language models (LLMs), (see ‘Ancient models and stochastic parrots‘ on October 1st, 2025).  My view is changing, probably as AI apps develop and my user skills improve.  I have started using a couple of different free AI apps as research assistants in three ways.  First, when I am writing administrative documents, such as a job description for a Coordinator of AI in Education, for which a job title was sufficient for the app to generate a first draft that only required light editing and tailoring to the specific context.  Second, using a different AI app, to answer questions about phenomena which have allowed me to construct explanations for observations made of new and, or, complex systems – I could have delved into textbooks and monographs or searched research articles but AI does this much more quickly.  The third way I have used AI apps is to identify gaps in knowledge that could be fruitful topics for research.  This is a more difficult task because AI apps only know about stuff they can find on the internet in the form of language or text.  Hence, I have to ask questions with answers that reveal something unknown or not understood.  This is not straightforward because LLMs are fundamentally constrained by language.  In ‘The Years’, Annie Ernaux wrote that ‘language will continue to put the world into words’.  Yann LeCun, Meta’s former chief scientist, has suggested that to understand how the world works, a model would need to learn from videos and spatial data, not just language, and that without this type of learning human-level intelligence is impossible.  He has set up a new company, Advanced Machine Intelligence Labs, to do just that.  Language is used by people to describe the world from their perspective which might be inaccurate, incomplete or distorted and that can mislead LLMs.  However, using AI apps we can also ‘distort’ videos of the world, so that machine intelligence will have to be based on direct observation of the real-world, which after all is the approach that science attempts to use.

Source:

Yann LeCun, Intelligence is really about learning. FT Weekend, 3-4 January 2026

Annie Ernaux, The Years, Fitzcarraldo Editions, London, 2018.

Extra! Extra!

Extra! Extra! Read all about it!  As newspaper vendors used to shout.  The Pint of Science Festival is happening across the UK in the week beginning Monday 18th May for three evenings in venues in 43 locations.  I am talking on the first evening, May 18th, in Lime Street Social (51 Lime Street, L1 1JQ) on ‘I Sell Here, Sir, What all World Desires to have – POWER’.  My title is a quote from Matthew Boulton, who with James Watt, set up a factory in Birmingham to produce steam engines in the 18th century.  I am going to talk about producing nuclear power units in a factory (see ‘Commoditization of civil nuclear power’ on June 5th, 2024).  If you would like to come to the event and hear three other speakers besides me and have a pint or two then please register at https://pintofscience.co.uk/events/liverpool/.

Experiencing success vicariously

Decorative image of a graduation ceremonyThe final PhD student for whom I will act as lead supervisor is scheduled to finish this month.  I have graduated forty PhD students since I was appointed a lecturer in 1985.  I am still in touch with many of them – they are divided between industry and universities with a bias towards industry (about 60%).  For the first twenty years, I was a sole academic supervisor often with an industrial supervisor providing support.  Then I moved to the US where a PhD committee provides supervisory guidance to the student and supervisor.  By the time I returned to the UK, about fifteen years ago, it had become accepted practice to appoint a second supervisor for each PhD student.  So, although I decided a couple of years ago not to accept any new PhD students as lead supervisor, I am acting as second supervisor for five students.  This is a great role since you have less responsibility, but you are engaged with the exciting research.  The topics vary from understanding the nanoscale mechanics of particles interacting with cells (see, for example, ‘Label-free real-time tracking of individual bacterium‘ on January 25, 2023 through to ‘Structural damage assessment using infrared detectors in fusion environments‘ on March 15, 2023), and just starting this year, innovative methods for communicating confidence in computational models.  Although the research is exciting, at a training session for supervisors during the CDT Winter School that I attended in January (see ‘Experiencing success vicariously‘ on January 7, 2026), we discussed our roles as supervisors and in particular that the research project is not the principal outcome of the PhD.  Instead, the development of the PhD student is the principal outcome.  It’s all about nurturing and mentoring people and the reward is experiencing their success vicariously.

Image: still from a video of a graduation ceremony at the University of Liverpool on December 9, 2025.  As Dean of the School of Engineering, I am at the lectern presenting PhD graduates, but I am hidden behind the Vice-Chancellor who has his back to the camera on the extreme left of the image.  You can watch the video at https://www.liverpool.ac.uk/graduation/the-ceremony/watch-graduation/catch-up/school-of-engineering/9-december-2025-10am/ .

Webs of expertise and knowledge

I am writing this post while I am in the middle of leading a breakout activity for more than a hundred PhD students from our Centres for Doctoral Training in nuclear science and engineering, GREEN and SATURN.  We have asked them to construct a knowledge network for a start-up company commissioned to build either a fusion energy power station or a power station based on small modular reactors (SMRs).  A knowledge network is a web of expertise and information whose value comes from the connections and interactions within and outside an organization.  Our aim is to encourage students to think beyond science and engineering and consider the interactions required to deliver safe, economic nuclear power.

We have brought the students together in York from six universities located in the North of England (Lancaster, Leeds, Liverpool, Manchester & Sheffield) and Scotland (Strathclyde).  This is an annual event usually held in the first working week of the New Year (see ‘Nuclear Winter School’ on January 23rd 2019).

The breakout activity has three one-hour time-slots on three consecutive days.  In today’s time-slot, we have divided the students into twenty groups of seven and given them paper, pencils, and a circle stencil plus an eraser with which to draw knowledge networks.  We are hoping for creativity, lively discussions, and some fun.  In yesterday’s one-hour slot, they had briefings from the Chief Manufacturing Engineer for a company building SMRs and the Deputy Chief Engineer of a company developing a fusion power station, as well as from a Digital Knowledge Management Consultant whose PhD led to a paper on knowledge networks, which we shared with the students last month (see ‘Evolutionary model of knowledge management’ on March 6th, 2024).  Tomorrow, one person from each group will have two minutes to present their knowledge network, via a portable visualiser, to an audience of several hundred.  What can go wrong?  Twenty two-minute presentations in one hour with one minute for questions and changeover.

GREEN (Growing skills for Reliable Economic Energy from Nuclear) is co-funded by a consortia of industrial organisations and the UK EPSRC (grant no. EP/S022295/1).

SATURN (Skills And Training Underpinning a Renaissance in Nuclear) succeeded GREEN and is also co-funded by a consortia of industrial organisations and the UK EPSRC (grant no. EP/Y034856/1).

Image shows thumbnail of figure from shared paper with knowledge networks for an engineering consultancy company and an electricity generator, follow this link for full size image.