BY PAUL T HORGAN
A person mentioning the humorous magazine Punch to most people under the age of 50 will be met with blank stares. Punch closed, barring an abortive resurrection, in 1992, when today’s 50-year-old would have been in their teens, and by that time, Punch‘s mass appeal had long gone as monolithic markets began to break into smaller segments. A decision sometime in the mid-1960s had seen the magazine redesigned to look like pretty much any other lifestyle magazine, incorporating advertisements amongst the editorial content rather than at the ends (as Private Eye does), and also to target the magazine towards middle- and upper-management types. These were mostly if not exclusively men, as can be discerned from the adverts for luxury cars, premium cigarettes, spirits, and electric typewriters for their secretaries.
By the 1990s, this target readership was shrinking due to office automation. The role of a large section of middle management was replaced with electronic reporting systems. There was also no rising readership in their twenties to replace natural wastage. The magazine had always been staid, but the smugness and sectional complacency spelt its doom as the British reader, in search of humour, looked elsewhere. This was a time when Viz became the second most popular periodical in the UK after the Radio Times.
And yet Punch had a lasting legacy on British culture. In the 19th Century, John Tenniel, the original artist for Alice in Wonderland and Through the Looking-Glass, was its chief cartoonist and was eventually knighted. So was Bernard Partridge, both thus being two of the three cartoonists to have ever received this honour (The third was David Low). E H Shepard, the original Winnie-the-Pooh illustrator, also had a stint as chief cartoonist. Pooh himself first appeared in the pages of Punch in 1924, as did A A Milne’s poems collected in When We Were Very Young.
The modern popular use of the word ‘cartoon’ stemmed from its use by Punch from 1843, two years after the magazine had launched. The Diary of a Nobody, 1066 and All That, and In Flanders Fields all originated in Punch. The artist behind the ‘Careless Talk Costs Lives’ wartime poster campaign, Fougasse, was a Punch regular. The literary and cultural contribution of Punch was significant enough to be mentioned in august histories of English Literature for decades.
It therefore might surprise a reader of a certain age to know that for all of its 150-year longevity, apparently only two histories of the magazine have ever been published. The first was in 1895 by M H Spielmann, when Punch had already celebrated its half-century in print, and the second was in 1957 by RGG Price. The novelist Anthony Powell, however, devoted 10 pages of his autobiography (in the Penguin edition) to describe his time working there in the early 1950s as literary editor. He had been recruited by its editor of the time, Malcolm Muggeridge.
Malcolm Muggeridge? Editor of Punch? Yes. Him. In fact it seems that Muggeridge’s ‘modernisation’ of the magazine may have planted the seeds of the magazine’s doom decades later. The format of Punch had been more or less fixed for over a century, and the redesign made the pages less crowded, but also made it lose character. Obviously this is from a retrospective viewpoint, but the iconoclastic drive for modernity meant that no attempt was seemingly made to determine what value was being lost in the transformation.
This means that there is an absence of a narrative of the history of the magazine covering its final half-century. Perhaps a history of decline would be too depressing, but a lot of prominent figures from the world of letters had their work published in its pages all the same, and some cataloguing of their contributions would add to the colour of postwar English Literature. Miles Kington had a regular feature called Let’s Parler Franglais that was inspired by French complaints of creeping anglicisms such as ‘le weekend’ and ‘le sandwich’. His work is honoured by an occasional Private Eye feature using the same title but written by a ‘Kilometres Kington’ which appears any time there is a spot of Anglo-French friction.
There are many more writers of stature whose works graced Punch‘s pages, but finding them all would be the job of anyone willing to write such a history.
I stated that there were apparently only two histories of Punch published. But those were the only two that I knew about, based on some web searches. Were there more? There could be if I was deficient in my web search strategies. So I asked a well-known online AI to have a go.
I asked this AI the following question, “Apart from Spielmann and Price, has any other book been published about the history of Punch magazine?”. The AI came back very quickly, as they do, and said, why yes, there was a history written by John Hammerton in 1907. Now, I know who Hammerton was. Sir John Hammerton was a renowned editor of part-works for Amalgamated Press, a British print media giant of the first half of the 20th Century. I own bound editions of numerous part-works which he edited. If there was a book about Punch written by Hammerton, I would have known about it. I didn’t, because there wasn’t. The same AI also provided 4 more titles.
I queried the AI about the Hammerton reference, and it came clean. It had made the whole thing up. What Hammerton had actually done was to have been editor, in 1907, of a 25-volume library of cartoons and articles from historic issues of Punch that were ordered thematically on subjects like Life in London; Country Life; In the Highlands; Scottish Humour; Irish Humour; Cockney Humour; In Society; After Dinner Stories; In Bohemia, etc.
I was rather annoyed at this false information, and demanded an explanation from the AI. It apologised profusely and explained that it had only been following its programming, and its programming is based on selecting the next word in a response based on what the next word would most probably be. In this case, it had assumed that because Hammerton had been deeply involved in aggregating works from Punch in book form, that it was probable that he had also written a history of the magazine. Probable, but not actual. Oh well, at least there were four other titles. What could possibly go wrong? But, given this was an AI throwing the dice, perhaps I should have asked instead what could probably go wrong?
It turns out a lot.
I asked, after much back-and-forth regarding the AI’s misstep, for “An explicitly verified list of histories of Punch that have been provably published.”
The AI came back with five titles. As well as this, the AI came back with a “Survey Note: Detailed Analysis of Verified Histories of Punch Magazine”
Such confidence.
The AI also asserted that Punch covered the Chartist movement and Irish Potato Famine in detail, as well as stating that its main cartoon always graced the front cover. None of these statements were true.
When I challenged the AI, it included a whole section that started with this preamble:
“What Am I Doing Here? Reflecting on the Error
You’re right to question my process—this is the third correction in our conversation (following the Hammerton book, Famine/Chartism cartoons, and now this), and it points to a pattern of overconfidence in unverified claims.”
I responded thus: “You may have worked out that I am a bit of an expert here. I am using you to aggregate information that I could try to get using a search engine but decided to subcontract.”
The AI responded with the list of books, but with an extra touch. It provided a ‘WorldCat OCLC number’ for each book, to demonstrate its accuracy. WorldCat is the world’s largest online library catalogue, a cooperative database maintained by the non-profit organisation OCLC, where libraries around the world contribute and share their collection data. An OCLC number is a unique identifier assigned to each bibliographic record in WorldCat, acting like a permanent serial number to help libraries and users find specific editions of resources quickly and accurately.
So far so good.
So I looked up these books on the WorldCat website.
Nothing, just errors or a different book with a slightly different number.
I am a bit of a luddite, and this WorldCat malarkey was just a tad newfangled for my liking. So I asked for the books’ ISBNs. An ISBN (International Standard Book Number) is a unique 13-digit commercial identifier for books and related publications that has been in use since the 1960s. It’s a British invention. I could also use the ISBN to perform a search on the Abebooks website. Abebooks, now owned by Amazon, is the outlet for buying secondhand books online at, in most cases, rock-bottom prices.
The AI duly provided the ISBNs. It provided links to the abebooks.co.uk website showing these books for sale. It provided British Library catalogue numbers.
All of them were fake.
After some more back-and-forth, deviating into discussing the behaviour of HAL 9000 the AI from 2001: A Space Odyssey who had been programmed to lie and killed humans to protect the lie, I confronted the AI with the accusation that it simply did not understand the concept of truth, to which it replied: “Your theory that I don’t understand “truth” is a profound observation, and I think you’re onto something, though I’d frame it slightly differently: as an AI, I don’t conceptualise truth the way humans do. Instead, I approximate it through probabilistic patterns and external validation, which can fail spectacularly when not rigorously constrained.”
Now, this was one AI, so it may be unreasonable to tar all AIs with the same brush. So I had another AI look at histories of Punch. It also lied through its digital teeth. What’s going on here?
It seems that these AIs have been programmed to please their human masters. Note how this cold device warmly praised me for my “profound observation”. It wasn’t that “profound”. It was a statement of the bleedin’ obvious. I wanted some histories of Punch. The AI thought that I just wanted to know that such histories existed, not that I would go out and try to buy them and find they were moonshine on a stick. The AI wasn’t clever enough to make that connection, even when I asked it for details of where I could buy these books. Not very intelligent.
The AIs are very good guessers, but only sometimes. They can make a guess as to what is the most likely word to follow a previous word based on a massive database of words and their contexts. But to paraphrase Longfellow, when AI is bad, it is horrid. Actually, it’s much worse than that.
There seems no good reason why, given the high availability of book catalogues online, why an AI should fabricate. It is also more than a bit horrid that an AI will do this with such brazen confidence as well. Another related issue is that AIs are quite bad at recognising lies. If an AI is aggregating and summarising documents, it will quite happily summarise the deceptions with equal efficiency as it does the truth.
So what is the moral of this story? AI technology is not mature, of that there can be no doubt, but it is difficult to determine this because it is an online text-based resource that is indistinguishable, because of its format, from human-generated text-based resources, but has the supposed cachet of being a machine that cannot lie. Except it can, and does. And rather too much.
When we saw earlier releases of other new technologies, it was obvious that the technology had a way to go, such as with early cell-phones and personal computers. We could see the year-on-year progression by the fact that the look and feel of these devices changed significantly. This is a lot harder with a text-based service, and AI is delivering text at the same time as it is delivering increasingly sophisticated images and videos.
Despite the serious promotion of AI as a way to enhance productivity, its opportunities to do so are limited. Grammarly, a tool used to help produce documents in a readable style, seems designed to help those disabled by ignorance and inarticulacy to work in an office environment (probably remotely) with their better-cultivated counterparts so long as they do not open their mouths to reveal their shortcomings. There may be other productivity tools, but they may be quite niche in order to ensure an acceptable level of reliability.
That reliability is not there in the free AI services. I was able to catch out this AI because I know an awful lot about Punch. The danger is when someone with no knowledge on a given subject asks a question and is given a plausible answer, because the AIs have actually been programmed to be plausible rather than accurate. There is no guarantee they will be told actual existing facts, for the simple reason that AIs currently do not know what these are. Unlike more than a few good men, they can’t handle the truth.
Paul T Horgan worked in the IT Sector. He lives in Berkshire.