Are you looking at AI yet?

Digital technology has been enabling efficiencies in the rail freight sector for years, and RFG’s own Digital Forum is providing a neutral space for RFG members to come together and discuss the opportunities and challenges from existing and emerging technologies. It is evident that within member companies, and companies more widely, there is great digital technology innovation going on. We should all be used to systems enabling back-office processing and it’s encouraging to see technology being increasingly deployed on wagons to improve reliability, performance monitoring and safety. Some might say about time, as data capture and monitoring on mobile assets is far from a new concept but it’s taken some time for the industry to identify the value of this data and agree who should fund the innovation and pay for its sustained use.

Topical recently has been the opportunity from AI. The UK hosted the global AI safety summit in early November, recognising that the transformation promised by the technology must be seen through the lens of avoiding unintended consequences of its use. RFG’s Autumn Conference included an eye-opening presentation from Daniel Ashville that very clearly demonstrated the power of positive use of social media in our sector, but it’s also not hard to find the pernicious effects of social media on some people with some rather less desirable impacts, especially on young people growing up today. Grappling with AI safety at this early stage is therefore a very sensible thing to try to do.

But what can AI do?

I suspect the term is often adopted as a buzzword for what are in fact normal albeit very clever and powerful technologies that don’t really use much genuine ‘machine learning’. Experimenting and exploring some of the more well-known applications, AI’s potential and current shortcomings are easy to see. As a digital assistant, ChatGPT from OpenAI is incredibly impressive in its ability for something called Natural Language Processing; the ability for you to ask a question and for it to reply as if in conversation. If you don’t understand the answer, further interrogating it with, “But what if I [changing a question parameter]” it will delve deeper or give improved responses as you clarify what you mean.

It’s important to recognise however that it’s not ‘creating’ anything new, but scraping the internet for sources of information, interpreting it to your query with an answer and playing it back to you in a chat. Logic, breadth and depth of access to data and analytical capability is very good but is ultimately limited by the search technologies that underpin it and within its existing programmed limits. But it cannot make up information that isn’t somewhere in the public domain. As with all online resources and technologies, it pays to have some idea of an expected answer and to ask follow-up questions to triangulate and validate what it is telling you. Where it is particularly useful in describing or explaining more ‘factual’ topics setting out, for example, formulas and worked examples.

Another powerful AI tool explored is those that ‘Chat’ with a PDF. Conceptually an odd phrase, these tools allow the user to upload a PDF or Word document, which the technology then ‘reads’ and can then answer questions on. With the relentless stream of policies, strategies, papers and documents produced in the industry and by regional stakeholders for RFG to keep on top of, its time-saving potential is obvious. Within business, its application to distil contracts and tenders is similarly so. ChatPDF is one such tool that I’ve been exploring. As with ChatGPT, it uses Natural Language Processing so it will interpret questions as if you are asking the document questions. As an assistant to pick out and summarise pertinent content, especially in long documents, it works well. But it’s proved prudent to understand it’s answering your written question, not reading your mind. If you haven’t asked the right question, it won’t feed you back what you might be looking for. Such is its speed though, it’s easy to ask several questions or hunt for keywords to pull out the relevant essence of a document.

ChatPDF usefully provides links to the document in its responses where it’s found or summarised content to allow you to read what the document actually says. For the most part, accuracy is impressive, but it won’t pick up inference or nuance and the unwritten commitments or non-commitments documents often include and aren’t a complete substitute for reading the document first-hand. However, used appropriately as an assistive tool, AI drives a lot of time efficiency.

Finally and for fun, you can’t help but admire its potential creativity in language and art. While I doubt it would win any prizes, ChatGPT and its sister product DALL-E came up with the following poetry and art in seconds just based on a few prompt words.

There’s no doubt that we are just at the start of the AI journey.

An Ode to Rail Freight – AI Generated (ChatGPT)

Hail, the mighty trains that carve the track,
Through the heart of Britain, no journey lack.
In the pulse of industry, a rhythmic beat,
The success of rail freight, a tale so sweet.

Across hills and vales, in steel’s embrace,
Cargo journeys with unwavering grace.
In the hum of engines, a symphony grand,
The triumph of rail freight, across the land.

Through challenges faced, through trials untold,
The industry’s success, a story to be told.
In the echo of wheels, and the whistle’s song,
UK rail freight prevails, resilient and strong.

Innovation blooms, efficiency’s gain,
A testament to progress, a thriving train.
So here’s to the tracks, the engines that hum,
In the success of rail freight, the nation’s thrum.


An AI generated image of a freight train.


By Martin Bignell, 
Scottish & Northern Representative