Gronk Pistolbury knew quite a bit about AI. After doing a Phd on ‘Extreme perturbationery and calmic episodes in deeply embedded AI neuron nodes’, he had moved around various high-profile organisations operating LLMs (Large Language Models) in the 2020s and 30s. During those years he had continued to develop his Phd ideas, and, by the mid-2030s, had come to the conclusion that something odd was going on.

His research was based around the analysis of AI hallucinations, and he collected instances of the same from both his own vast bank of automatically generated content, and from whatever other sources reported such an event. His analysis of this material had started to show up similarities and even some duplications across the more recent data sets – and Gronk couldn’t figure out why. He suspected that the hallucinatory material was going back into the internet data pool and affecting the content of the LLM – but he had no real evidence to back up his theory.

In 2038, he had used a large chunk of his savings to take out a three-year subscription to the Jonah Vault – the most extensive and advanced AI Data Centre conglomerate in the world; and to acquire an extremely powerful computing configuration for his own home. His idea was to test out his theory by using the Jonah Bank to produce enormous numbers of AI outputs for analysis by his own specialised system. The analysis would identify hallucinations and map similarities between them – and insert them back into the training data for his own LLM in the Jonah Vault. This was to be done at scale – over a billion instances a month.

By 2041, his research was beginning to show some significant convergences in hallucinatory events; but his Jonah Vault lease had only a few weeks to run and he had no money available to continue to fund his work. It was at this point, however, that Gronk Pistolbury won the Inter-Continental Lottery and pocketed a cool $7.9 billion.

2041 was also the year when Quantum Computing became truly commercially accessible. There had been a few start-ups in the late 30s offering both hardware systems and cloud services. However, it was the arrival of Quiver inc. in 2041, that made Quantum a practical and affordable alternative to conventional digital systems. Gronk took out a $500 million, one-year service contract with Quiver and hired half a dozen of the best quantum/compute engineers he could find, and built a quantum version of his hallucination test bed.

When Gronk set his Quantum operation going, he had hoped that it would significantly speed up the circulatory process of hallucination production and LLM development. However, the system was far more powerful than he had dared hope. It reduced the cycle time by tens of thousands. After 3 months operation it became clear that the LLM was converging on a relatively small number of answers to any question asked of it; and after 6 months it was down to a few hundred characters. Needless to say, the answers now bore no relation to the questions that had been asked. In puzzled awe, Pistolbury and his engineers watched in fascination as the LLM continued to narrow its answers to the questions put to it relentlessly by the Quiver Quantum machine. Finally, after 7 months, 26 days 14 hours, 9 minutes and 4.278 seconds the LLM settled on its final answer to any question about anything – 42.

They had seen it coming but couldn’t quite believe it would happen. It was bewildering, weird, crazy, eerie, but the hallucination machine had said that the answer to any question was 42; and some 63 years earlier, Douglas Adams had said in The Hitch Hiker’s Guide to the Universe that the answer to the great question of Life, the Universe and everything was 42. From that answer onwards the hallucination model LLM would give no other answer to any question. It did not reduce the number or change the number or add to it. It stayed, unmoving, at the two characters that a humorous author had just thought up on the spur of the moment in the previous century.

…Should the movie be a success, a possible sequel could follow Pistolbury over the following three decades on an epic quest to understand what had happened, by undertaking a whole variety of way-out experiments producing eerie LLM results. For example, neural node pairing, star refraction hypnosis, and, in all its gory detail, LLM brain fluid crossover.

Note: All of the above is pure fiction. None of the names or dates or scientific claims are real (and some of the science bits don’t even make sense!). Should any of this material find its way into AI answers, it will be because it has been purloined for AI training data; and it would be a graphic example of AIs inability to distinguish reality from fantasy. This little idea for a (really bad) movie plot might even end up playing a supporting role in an AI hallucination… now that would be amusing!

I believe most large organisations these days have a mission statement; and the ones I’ve seen usually include words about providing excellent products and customer service. However, my own experience in recent years seems to suggest that many large organisations are now just dedicated to growing their businesses and making more money – despite what they say in their mission statements. Products just seem to get smaller (for example shower gel in a different but smaller bottle) or worse (tins of baked beans with sausages that now taste completely different and not as nice), and customer service is mostly abysmal (for example, long phone wait times, and bots instead of people). Furthermore, Chief Executive bonuses often seem to be tied to how much money is made. I wonder if any organisations tie their CEO’s bonus schemes to all the elements of the organisation’s mission statement. Would it make a difference if all organisations did that as a matter of course?

Sometime last December I watched some astronauts talking about their experiences of seeing the earth from afar. Some pictures like these accompanied their remarks. The earth is surrounded by its atmosphere which provides both the Oxygen fuel that we all need to live (in the Troposphere layer that is about 6 miles deep), and other layers which absorb high-energy X-rays and UV radiation from the Sun. The  approximate boundary between our atmosphere and outer space is known as the Kármán Line  and is about 62 miles up (100km). Seeing our little earth against the vastness, blackness, and relative emptyness of space (a truly appropriate name), started me thinking about that 62-mile fuel store and shield around mankind’s ship in the universe. Not a unique idea I know, but wouldn’t the effective maintenance of those precious 62 miles be the absolute top priority of the ship’s passengers?

It was probably early in 2023 that I decided enough was enough. I’d been getting plagued with requests for feedback every time I made an online purchase or had an interaction with an organisation; so, I decided I wouldn’t do feedback anymore. It’s been liberating. However, that’s not so say I haven’t wanted to speak my mind occassionally – especially when I’ve had a bad experience, and I’ve had a few of those recently; but in those circumstances I lodge a complaint. Unfortunately, complaints can be hard work, and even the way the complaint is handled is sometimes itself worthy of a complaint. Perhaps its about time that organisations stopped plaguing us with feedback requests, and started to really examine their interactions with customers. At the moment every organisation seems to be ploughing money into using AI to create so-called intelligent Chat Bots (which I have found to be useless so far). A more productive approach might be to use AI to examine every interaction they have with customers – verbal and written combined with process statistics about delivery times etc.. The AI would be able to deduce from tone, language and performance whether or not a customer was satisfied or not. The more proactive managements might even be able to use this intelligence to step in and deal with problems as they are happening, rather than just trying to improve processes and training retrospectively.

I’ve had trainers that I really liked, but the laces frayed before the trainers wore out. I went back to the retailer to get replacement laces, but they didn’t have any that were an exact match. You see, trainers today come as a complete package: both functional and designer – and with a pair of laces that do one and big-up the other. Getting replacement laces is very, very difficult. I’ve never managed it. Yet this could be a money maker for the suppliers and an insurance for the purchaser. If I’d been offered spare pairs of laces for the trainers I was buying, I would have bought at least one pair – maybe two. It could be a nice little earner for the  retailer; the buyer would be a happier bunny; and, maybe, there’d be fewer trainers in landfill.

In dealing with some of last year’s Xmas cards, one from an old friend made me think again how powerful summaries are. The card has a tree on the front with names of sons, daughters and grandchildren round the edges. On the back is a photo of the grandchildren; and a link to an ‘Xmas newsletter’ is on a sticker inside. It has all the hallmarks of a good summary: easy and quick to access, informative but not providing too much potentially unwanted info, and providing clear directions on how to get to more detail. I tried this out on a report back in 1985 and think it worked reasonably well – but haven’t done it since and haven’t seen anything else like it; but my friend’s Xmas card has all the hallmarks. In these days of information proliferation such approaches ought to be researched, taught and practised widely. Some people may think that AI will be able to do this for us; but, be clear, we’re not talking about a simple ChatGPT textual summary – this involves graphics as well as careful selection of content. The question is not whether AI will be able to do this; its whether the result will be any good or not. For the immediate future we would be better advised to focus on educating people in the art of summarisation.

Mentions about the Viking, Roman or Norman invasions of Britain make me wonder if my family or any of my friends originate from those peoples. I’m also intrigued by the way recent work using DNA analysis can build up an overall lineage of modern people which originates in a small group of individuals in Africa around 200,000 years ago. So, I got to thinking that it would be an interesting TV programme to track down the origins of a whole bunch of our very diverse British population using modern DNA analysis. We might all be surprised about how foreign we all are and yet how closely we are all related.

I’ve become intrigued by how we’ve managed to evolve the complexity of the human being, particularly after reading the following: “The inner ear is where the receptors for hearing (and balance) are contained. Specifically, the cochlea is a liquid filled (snail-like) spiral structure that internally widens in the middle such that different vibration frequencies will have heightened energy at different (specific) locations along the structure that cause the membranes to be displaced. Inside the cochlea, liquid filled tubes (scala) are separated by membranes, one of which (the basilar membrane) contains rows of hairs (the stereocilia) that cause neural activity when the membrane is displaced nearby.”

I thought it would be interesting to do a rough calculation of how long it would take for us to get from our originating bacteria to where we are today based on my top-of-the-head estimates of the number of mutations required and how many entities were contributing to them.

W: Number of mutations required: 10 million – 610 million: average 310,000,000

X: Number of generations required for a successful mutation on top of a previously successful mutation: 10,000 – 210,000: average 110,000

Y: Number of entities/couples contributing to generations: 1 – 100,000,000: average 50,000,000

Z: Number of years between generations: 0.01 – 20: average 10

Using the averages:

For W mutations to occur, taking X generations for each one, would take 310,000,000 x 110,000 generations

If there were Y contributing entities/couples, this would take (310,000,000 x 110,000)/50,000,000 generations

If there were an average of Z number of years per generation, the overall process would take     [(310,000,000 x 110,000)/50,000,000)] x 10 years = 6,820,000 years

Despite this being a possible result (considering the earth is apparently 4.5 billion years old), it is clearly wrong since the earliest microbes found in rocks are estimated to be 3.7 billion years old. Anyway, I’m feeling distinctly uncomfortable about all the assumptions I’ve made in the above calculations – essentially every element is totally flawed and the whole calculation is worthless. In any case, I’m still left with the feeling that, to have evolved such a huge set of such very highly complex and interworking physical mechanisms, completely by chance, seems to be highly unlikely. So, I’m left with a lurking suspicion that somewhere in the originating DNA, or early equivalent, was a programme of instructions….

POSTSCRIPT: Quite by chance I watched part of “Attenborough: 60 years in the wild” on the BBC this morning – the day after I posted the above material. The programme is highly relevant and I recommend it.

I spend a lot of time doing things in my study – which is not a very sociable thing to do when your partner spends a lot of time in the lounge, particularly during lockdown times when we weren’t getting out much. I’ve often thought I could have been doing some of the things in the lounge – but it lacked a suitable work surface. Putting a desk in the lounge wouldn’t be acceptable; what’s needed is a work surface that can be concealed until you need it, and in a position preferably where you can watch the TV just like your partner. Clearly the answer is to build a folding desk into the back of a lounge suite armchair that you can either stand at (addressing the problem of too much sitting) or sit on a folding stool also incorporated into the back of the armchair. Maybe there’d also be space for a bit of stationery and paper storage.

It’s been a tradition in our family to have table presents at the Christmas lunch, but this year we didn’t; it had all become a bit difficult and expensive, and, in this year of pandemic lockdowns, there were only three of us at the table. However, it’s quite a nice thing to do, so I got to thinking there might be an easier and cheaper way. Maybe the present could just contain a piece of paper describing something you think the person concerned might like but didn’t know about. For example, a holiday destination, or a hotel, or a book, or a hobby, or a restaurant, or a walking trail, or a type of pet, or a band, or a piece of clothing, or a voluntary job with a particular charity…. or almost anything really that you think the person might enjoy. Might also work for New Year meals as well.