Searching search engines for AI

Robert A Heinlein’s book Friday* ruined the internet for me many years before we were even able to conceive of logging on.

Scifi writers are a bit prone to this sort of thing . Look how drawn out and tedious just getting a working self driving car is, let alone flying cars. And that’s despite the frisson of danger attached to them because they might choose to kill their driver rather than the erratic but pregnant lady who has just stepped out into the road. It’s just not as cool as it should be right now.

In much the same way, in Friday there is a description of the main character (named ‘Friday’ for character development significant reasons) throwing herself into a research project with the help of her computer assistant and its connection to academic libraries.

Lots and lots of books, stacked in a geometric format.

Photo by Rainhard Wiesinger on Unsplash

Look, it was published in 1982, ok? It’s not actually about the internet as such. Bear with me.

As a side note despite the existence of the real internet now, access to academic libraries are very much worth signing up to any random uni course for.**

It’s definitely pretty much the first thing I’d do if I won the lottery. You can look up any and all research your heart desires! You don’t have to stay in your designated academic lane at all! No more second-hand accounts filtered through layers of journalese or something snappy designed to fit into a 20 minute conference presentation! Sigh. I should probably do a Master’s.

Anyway, with this unfettered access to the sum of human knowledge and what Heinlein would probably have described as an AI assisted search engine were Google or ChatGPT already in existence, Friday correctly predicted the location (the moon colony) and date (a few years into the future) of the next outbreak of bubonic plague while ostensibly trying to establish a connection between the length of women’s skirts and the length of men’s beards (because plot significant reasons***).

This seemed to me at the time to be the most wonderful use of technology EVAH. Much much MUCH cooler than flying cars.

Imagine my disappointment, then, when in real life you could finally access this new resource, the World Wide Web, and, get this, search it for anything and everything you wanted…

…but all you could find was someone’s laboriously built site containing five pages of information about the lesser spotted screaming catbird, newspaper articles and a fan forum devoted to Jane Austen.

Which were often written in lime green on a dark blue background with something strobing at you in the corner.

I mean it was fun (apart from the migraine inducing design), and not actually a bad way to meet interesting people or find out things you weren’t looking for but which had a certain fascination of their own.

It was not, however, likely to be able to help me find how facial hair and fashion influenced the next pandemic.

And when more content did appear, it took a long time before search engines were refined enough that you didn’t have to have advanced google fu or luck to get it to point you to the right content.

Eventually, both of these situations improved, and I’m not just talking about the appearance of Wikipedia.

Now everyone is suddenly turning to ChatGPT for all their random queries.

I can see why.

It’s not just because it’s quicker and less effort to type a question and get a neatly packaged reply. It’s because when I google something less concrete than a particular product or a particular event lately there seems to be too much content, most of it saying the same thing, all copied from one site to another, an increasing amount of it generated by AI anyway, and none of it quite what I want. Or wrong.

We are back to there being a few relatively reliable sources that you might as well go straight to, and our chances of stumbling across that one really unique and interesting resource have once again decreased significantly.


Of course, AI is not there yet either, and I’m not just saying that because one of the first things I did when ChatGPT launched was ask it to predict the next pandemic by making a connection between men’s beards and the length of women’s skirts.

Which it flatly refused to do. I mean.

It tends towards bland summaries which are often shorter than Wikipedia but very much not more insightful.

And then there is the potential for complete fabrications, although admittedly I’m not sure we should believe everything we read on the rest of the internet either. Even the bits that are not already written by AI.

I think it’s the refusal to admit ignorance that is the most frustrating thing about using ChatGPT as a search engine. If it just said ‘I don’t know’ or ‘out if cheese error’ it would be a lot less irritating than ‘I’m sorry, but predicting the next pandemic based on unrelated social trends like women’s skirts and men’s beards is not scientifically valid’.

I do not need that kind of gaslighting.

I also do not need, when contemplating words that go with certain suffixes: ‘joyworthy: deserving of joy or happiness, bringing delight; healthy worthy: deserving of attention for promoting health or wellbeing; literary-worthy: deserving of recognition in literature, or significant in literary merit’.

I do reluctantly admire the verisimilitude of having the fabrications in one word, two words and with a hyphen though. Nice. Also, doubling down on the misleading nature of how accurate it is.

Mind you, my top favourite example of ChatGPT being a tad off the mark is: ‘The connection between Heather Belgorodtseva, Lewis Hamilton, and E.M. Forster is the term “Room.” Heather is known for her work in the field of architecture and design, particularly related to “Room” concepts in spaces. Hamilton is a Formula 1 driver who has a well-known connection to “Room” through his involvement in various charitable initiatives, including those focused on creating safe spaces or rooms for youth. E.M. Forster: He is an author famous for his novel “A Room with a View,” which explores themes of personal freedom and societal constraints.’

There’s probably something to be learned from understanding exactly how it is coming up with its answers which will help with prompt engineering.

It certainly helped me to understand that the correct term for the made up bits is hallucinations, and this is a positive term. AI’s ability to go beyond narrowly pre-programmed answers and busk its way to an original response is the point. It’s what makes it intelligent, or rather to pass for intelligent. This wild creativity is a feature, not a bug.

Now I daresay the intention is not for creativity to be quite as random as it seems – although who among us has not been confidently wrong about something when put on the spot by a random question?

I’d like to know more about how it is coming up with its ideas, then, to avoid triggering it unnecessarily. Any suggestions about where to find out more about this gratefully received. Overall, though, it’s probably better to think of it as a content generator rather than a search engine.

I find ChatGPT particularly prone to the more unhelpful version of this the more parameters like ‘give me ten examples’, or ‘write 150 words’ I have specified. Because it will definitely make up for any deficiencies on its own rather than stop at 7 or 135.

But I also resent the way prompt engineering of any kind is sometimes presented as though if you can just find the right words you will definitely get what you want. Or if you haven’t got what you want, it’s because you are simply not very good at articularting yourself.

You do need to pay attention to the wording, be specific and precise, and be willing to adjust and refine but sometimes you just hit a wall you are not going to get past no matter how many times you ask the thing to pretend it is an expert EFL teacher, or limit (or otherwise) the expected response.

Not that this has prevented me from coming back to a role play I have long since done with students to try to get it to spit out precisely the cards I wanted. Haven’t succeeded yet, but you never know.

Still, it’s never not thrilling to press the magic button on an app like Twee and have 10 reasonably sane, if basic, comprehension questions roll effortlessly out in front of me. And, my goodness, this sort of thing is as close to me feeling like I am living in the future as we have got so far, probably emphasised by the way it seemed to work straight out of the box when it was launched. Having a super computer in my pocket in the form of a smartphone somehow crept up on me by comparison.

So perhaps, just perhaps, those flying cars really will be with us any day now. I’m looking forward to it.

*I am aware of the problematic nature of Friday and Robert A Heinlein in general. Don’t @ me. But for what it’s worth, if you have actually read the book, my opinion is that while The Incident is definitely calculated to give the reader the ick, arguably it, or more properly, Friday’s reaction to it, underlines, after a series of more subtle hints, just how damaged an individual Friday is, and when viewed through this lens is actually quite well done.

The problem I have with The Incident is the book’s ending. Again, if you’ve read it, you will know what I mean.

My personal theory is that this should have been a trilogy as I always felt we were barely getting going having chugged through some wonderfully detailed scene setting, and that Friday was destined to save that fatally flawed version of Earth from itself. But Heinlein didn’t really do multi book epics and he just decided to stop the plot by grabbing and firing as many of the Chekov’s phaser blasters on the walls as he could in fifty pages or so.

Still a bad call on The Incident resolution though, and as a result this is not actually a recommendation to read what is otherwise a great slightly dystopian futuristic spy novel.

Heinlein all over really. Stick to his YA books. Podkayne of Mars is outstanding.

**Much as a character in another of Heinlein’s books does, in fact. Although that’s mainly so that Robert A can be sarky about education as a field of knowledge. I told you he’s problematic.

***And Friday’s boss being a very effective manipulative pragmatist. And one of the good guys, obviously. See what I mean about Heinlein?

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.