Not [even remotely] There Yet

It's been a strangely philosophical day today, with AI featuring front and centre above [and between] several coincidental threads that have been thrust my way, so to speak. In chronological order, I picked up a thread first thing this morning over my first cup of tea, that pointed to an academic paper emanating obliquely from Apple Inc., indicating that some reticence to the wholesale adoption of AI across the broad spectrum of knowledge applications is probably in order. The paper outlines research carried out regarding the performance of various AI LLMs with regard to simple arithmetical logic problems of the order of those used to gauge the performance of year eight pupils in standard testing. We've all done these at school at one time or another: 'John has forty apples in a barrel, three tangerines in his pocket and fifty kumquats in his knapsack. How many orange-coloured fruit does John have?' etc., etc. When the AI is presented with straightforward arithmetical calculation, it arrives at a correct answer.

However, when the researchers threw in non-functional but apparently - from a language point of view - relevant data: 'John has forty apples in a barrel, three tangerines in his pocket and fifty kumquats in his knapsack, but fifteen of the kumquats are starting to go rotten and can't be trusted for eating'; the AI will calculate the number of orange-coloured fruit as 53 - 15 = 38 orange-coloured fruit; missing entirely the linguistic blind of the spoiled fruit issue that any human would spot immediately and reject from the sum as irrelevant to the question posed: the question being purely arithmetical and involving only the number of orange-coloured fruit, and having nothing to do with their edibleness or otherwise. It has extrapolated wrongly that the spoiled fruit, as they are allegedly inedible are ergo to be automatically taken out of the arithmetical sum at the heart of the question posed. It is an apparently trivial test that actually speaks legions.

Similarly, although tangentially, I was listening a couple of AI-focussed radio discussions on BBC Radio Four during the day; the first featuring an interview with the 'Godfather of AI', Geoffrey Hinton, who now harbours a personal disquiet over its future usage and potential abuses. The second was a rather less philosophical discussion of the employment of AI in the creation of novel food recipes. This might sound trite, but actually raised the deepest philosophical point to be made out of all of today's input. The example used was of the humble mince pie. The point made was that the first mince pie of the festive season sparks a unique Proustian evocation from each and every individual - myself not included as I don't eat the little swines - casting the mind back to some, usually childhood past. As the guy being interviewed said, each of us has a [food] memory of such things that is utterly unique to each of us alone and shared by no-one else. AI does not possess this uniqueness of experience and memory. Sobering and comforting in equal measure, methinks. Later discussions with the family turned to the nature of consciousness on the back of these thoughts, but that's a whole other discussion to be had, and I've been to the pub and now need some food to soak up the beer! More later...