All children - and more adults than like to admit it - have imaginary friends. For this series, I had four of them, sitting around a fireside, having after-dinner chats. It's all very informal but we looked at some serious issues in a way that belied the complexity and depth of what they were talking about. There was a lot of fun as I let four different voices, with different approaches, run riot in my head and onto the page.
The "Four Imaginary Friends" idea came from Monty Python's Four Yorkshiremen" before they were Monty Python. It's here: https://www.youtube.com/watch?v=VKHFZBUTA4k
But it wasn't the only influence: when looking at #artificialintelligence and language, The Italian Language Class in this piece "https://www.youtube.com/watch?v=8YFZowBNk3Q" played its part.
Go on: teach your #LLM models to cope with this :)
When your computers shake themselves to pieces because it understands this silliness, then you can say you've got a viable product.
Keep watching that programme for more things that computers can't cope with.
Then there's this at 1:30. Aside from the silliness, this also relates to another issue raised in the series: accents.
This series has been a huge amount of fun to write but it was time to bring it to a conclusion.
It was laden with twists and turns of language, playful examples of inconsistencies that current "AI"models are unable to handle,
Editing of the series has been characterised by battles over how Americanized (see what I did there?) the four Englishmen could become before losing the essence of their conversation and their speech flow. I think we mostly got a balance. Those battles reflect one of the major issues with algorithmic analysis of language-based data: Alan Turing said that machines would need to learn English. But if that's the case, first we have to know what English is.
Is it 'Strine, Singlish, Manglish or that most corrupted, and corrupting, version: American? One thing is certain, for a huge amount of development which has been commercialised, it's not English.
Should we allow the rampant modifications approved by Oxford University or should we try to encourage Standard English (which is English not, as the Americans insultingly call it "British English")?
Should we, the English, have a government sponsored unit to define English, as the French have with French and the Malaysians have with Malay. And if so who will define it and to what purpose? In Malaysia, traditional Malay words are being replaced with modern alternatives, much to the annoyance of many who appreciate the beauty of traditional Malay. Or the French who seem to have given up their rear-guard action against "le Weekend" and are now beginning to face the demands of those who want to destabilise centuries of certainty by abolishing "le" and "la." The irony is that French is quite a strict language, far less flexible and interpretative than English and it is, therefore, more suited to the demands of algorithmic interpretation than English. The loss of certainty that is rapidly accelerating in English, if followed into French, will result in rapid deterioration of the ability of French Language Models to produce accurate results.
The bottom line is this: you can't have "artificial intelligence" unless it is backed by Actual Intelligence. As the final article says, to machines men are gods but gods that do not have the power to grant free will.
Thanks, The Yuan and the whole team. It has been fun but it's also been challenging, to write techy stuff about tech in a totally accessible fashion for a global audience not only with so many versions of English but so many levels of capability, too. So you should read all six parts. Go on, the fun I was having makes it to the page, more or less intact :)
If we have made some minds enquire, we've done well. And if we've upset a few apple carts, we've done better.
The articles are here: