Is it good to chat?

On the day I decided to test-drive ChatGPT, Google launched its Bard chatbot. ChatGPT had no idea. “As of my knowledge cutoff date of September 2021, I am not aware of any Google product or service called ‘Google Bard.’” So there ends its value, with devastating irony, as a tech news source. ChatGPT was also unaware of Microsoft’s “chat” version of Bing, which it was allegedly powering itself. So let’s off self-awareness too.

I tried it at the basic, time-wasting end of information, asking for the chords of Lou Reed’s Coney Island Baby, one of my favourite songs. The bot knew Reed was the author but gifted me, unasked, the following lyrics: “Well I’m sitting here looking at the sea, thinking about Coney Island and me.” To my knowledge, no song exists with that opening line. Google agrees. But Google, and me, might be wrong.

Not a great start, so I decided to check if it was intending to replace me. AI-driven bots, we are told, can perform not only menial tasks but many information-centred jobs, such as legal reporting, providing health information and journalism. Much of the latter is transactional and it is here, rather than in news-gathering, reviews or opinion-writing that a chatbot could ostensibly be a rival.

I asked ChatGPT to help me plan a trip to Grimsby, ostensibly to produce some travel journalism. It spieled off a Wikipedia-type intro (“Grimsby is a town in North East Lincolnshire, England” etc) and then made ten suggestions for our day out, including Grimsby Fishing Heritage Centre, Cleethorpes Beach and Grimsby Minster – but also Freshney Place Shopping Centre, showing it doesn’t make value judgements.

For a “special dinner in Grimsby” it came up with six places which, to be fair, looked good, especially the top-most tip, Old Church Restaurant, “a fine dining restaurant located in a historic church building, offering a contemporary menu and elegant atmosphere”. The bot, while programmed to withhold opinions, accesses online reviews – TripAdvisor, Yelp and such – to help people make their choices. Of Anokhi, a Grimsby curry house, it said: “Many customers have praised the restaurant for its flavorful [American spelling, of course] and well-prepared Indian cuisine, attentive service, and pleasant atmosphere.” It refused   my invitation to come and share a biriani.

We can assuredly affirm that churnalism, far from being dead and buried, has a rosier than ever future. Murdoch, the Barclays and whoever is responsible for those local “newspaper” websites must be thrilled at the idea of a robot doing the donkey work for free.

The bot waxed botically on the 2016 film Grimsby starring Sacha Baron Cohen (“criticized for its depiction of the titular town and its residents… some people believe that the film perpetuates negative stereotypes”) and then, tentatively, reflected on whether the word “Grimsby” may have connotations of being “grim”, which segued into comments on place names. Scunthorpe, it suggested, was unfairly characterised because its name “includes a string of letters that is often filtered out as inappropriate by internet censorship algorithms”. Go on, just say it!

Chatting is not all – or mainly –  about exchanging data, so I threw in a digressive question on wine bars in Grimsby (it gave me a list that looked satisfactory) and then a Guinness Book-type quiz question on the most popular red wine in the world? The bot, which had no idea where I was going with this (I think?), said it was cabernet sauvignon. I then asked if women in Grimsby would enjoy this grape variety? Was my question sexist or surreal, or both? Was it too random for a machine-brain?

ChatGPT paused a few seconds before pontificating. “It’s difficult to generalize the preferences of all women in Grimsby, as everyone has unique tastes and preferences when it comes to wine. However, Cabernet Sauvignon is a very popular wine globally and it is likely that some women in Grimsby enjoy drinking it… Some women may prefer other types of wine, such as Pinot Noir or Shiraz, over Cabernet Sauvignon. The best way to find out if Cabernet Sauvignon is a popular choice among women in Grimsby is to ask locals or to visit local wine shops and wine bars to see what types of wine are most in demand.”

Talk about hedging. Then again, “Some women in Grimsby enjoy Cabernet” is a better song lyric than the ones in the fake Coney Island song of our earlier exchange.

We went on to discuss wine and women and agreed “it is sexist to make assumptions about what someone likes to drink based on their gender”. So, I broached one of my pet areas: mild ale. ChatGPT told me one of the reasons mild ale had declined was that it was less profitable, because brewers were likely to make fewer sales and manufacturing it was more complex, time consuming and costly. Unfortunately, the platform hasn’t been geared to provide footnotes or useful supporting statistics.

Happily, and weirdly, the bot agreed that mild ale made for mild-mannered people, but then it set itself off on a righteous rant about drink and behaviour. I suggested it stop lecturing me, to which it responded by coyly conceding I should enjoy my pint of mild. Then, like a semi-doting somewhat puritanical grandma, added: “Just remember to drink responsibly and be mindful of the effects that alcohol can have on your behavior and decision-making abilities. Cheers!”

ChatGPT feels like a dark but shallow rabbit hole. You go in, get lost, grow a bit older, and resurface feeling drained and dimmer. I’ve always been a bit of a magpie and trivia buff; it was, arguably, made for people like me. But are we legion? The site wouldn’t (or couldn’t) tell me how many users were online. “While I have been trained on a vast amount of information, I do not have the capability to gather real-time data about user interactions or usage patterns.” A classic instance of Big Tech gathering but not sharing data. The numbers will rise, then fall and then we’ll see.

Before logging off, I consulted it on Boris Johnson’s partying habits, jokes about Argentinians, Knut Hamsun, loneliness and music. The bot swears, though in inverted commas (telling me punk had “a ‘fuck you’ attitude”). It was rubbish at directions. It wouldn’t arrange a date with a porn star.

My pint was gone and I was still heading for a solitary curry, so I asked ChatGPT to boost my ego and share my own past articles. It said it couldn’t. I protested that it sourced information from newspapers and magazines but didn’t credit them.

“I apologize if it appeared that I was using information from articles without proper attribution. As an AI language model, I generate responses based on the information that I was trained on, which includes a diverse range of sources. However, I understand the importance of crediting the original source of information, and I always strive to provide proper attribution when it is appropriate. If there is specific information you would like me to reference, please let me know and I will do my best to provide proper attribution.”

When I demanded it confirm a published quote was mine it deadpanned me: “Too many requests in an hour. Try again later.” When the system fails completely, ChatGPT offers a profuse apology in corny Shakespearean language. A techy’s sense of humour, I suppose. Eventually, when we were reunited, it offered me a pat on the back. “Chris Moss is a well-regarded travel writer and author with a reputation for producing insightful and engaging writing.”

Aw, thanks.

Digital media and the tech sector abuse language as a matter of course. Twitter isn’t in the slightest bit “social”. Facebook is faceless and anything but book-like. TikTok destroys time. Chatbots should be called interrogate-bots. “Chatting” to it was like trying to stimulate a disembodied teleprinter. ChatGPT does not chat, and its own definition – “an informal conversation with one or more people, often through text-based messaging platforms” – is self-serving, duplicitous and, funnily enough, all too human.

---

Information versus regulation

A year used to be a long time in the tech world. Now a month is ages. ChatGPT was only launched by its makers, San Francisco–based OpenAI, on 30 November  2022. It hit an estimated 100 million active users by the end of January, making it the fastest growing consumer internet application in history, according to a UBS study.

A large language model (LLM), a machine-learning system that autonomously learns from data, ChatGPT defines itself as a “state-of-the-art language generation model… trained on a massive dataset of conversational text and can generate human-like responses to various types of prompts such as questions, statements, and prompts for specific tasks like translation, summarization, and text completion”.

Boring, then, and more likely to cough up your next shopping list than steal your novel in progress. 

But there is concern out there, not least among the scientific community. “This technology has far-reaching consequences for science and society,” advised a recent article in Nature, noting that ChatGPT can write essays and talks, summarise literature, draft and improve papers, as well as identify research gaps and write computer code, including statistical analyses.  

“Soon this technology will evolve to the point that it can design experiments, write and complete manuscripts, conduct peer review and support editorial decisions to accept or reject manuscripts.”

The five authors, while acknowledging that railing against it is pointless, warn that “ChatGPT and other LLMs produce text that is convincing, but often wrong, so their use can distort scientific facts and spread misinformation”.

Apps, gizmos and tech firms are always in the news, but the last time a tech platform generated genuine intellectual debate was perhaps Wikipedia, back in the mid-2000s when journalists and scholars ridiculed it for being unreliable. Now, apart from Google, Youtube and social platforms, it’s the most visited website on earth, with more than six billion monthly visitors.

“My primary concern with ChatGPT is that our society, particularly our education system, is not yet adequately prepared for generative AI,” says Ruilin Zhu, a lecturer in information systems and management at Lancaster University Management School.

“All technologies and technical tools, including generative AI, should be subject to appropriate regulations. ChatGPT is no exception. If we consider ChatGPT as a tool, which is both powerful and useful, then a set of detailed standards should be put in place to well define its scope and quality. 

“If ChatGPT is seen as a service, which is intelligent and smart, then the servicing process should be required to be explainable and rational. Together, they give rise to key questions – who should take the leading role in deciding the ‘persona’ of ChatGPT, and what type of information should be fed to it?

“Therefore, the most significant challenge is that on the one hand, we must ensure that our understanding of such platforms and their implications keeps pace with their rapid development, and on the other hand, centralised regulation should aim to make legal and value systems resilient to digital changes instead of merely restricting digital technologies like ChatGPT.”

But will we, and will it? Right now, the digital revolution proceeds in a manner as unregulated as the industrial ones before it. Dialogue-generating machine learning of the kind ChatGPT utilises is its latest gizmo, its spinning mule if you like, hybridising the general knowledge of Wikipedia with the boundless searching power of Google. It will make life better for cheats, dullards and idlers. If the converse is also true, its unopinionated amorality will turn out to be its cleverest ruse and we ordinary mortals, like the handloom weavers who prided themselves on their artisanship, will soon be re-employed as server-wallahs and disk-doffers.