A Soliloquy on TechTonic prompted by a reader’s email

The CANADA DAY edition where I avoid work and express doubts about recommendation engines

Jun 30, 2025

In Canada, it is a big stat holiday called (surprisingly) Canada Day, which turns into four days off when the actual date is on a Tuesday (or Thursday). So it will be later in the week when my keen critical eye will evaluate how well my predictions for 2025 were.

READERS POTLUCK: An email and a conversation with a 13 year old generates thoughts

I’m sure those free software tools that evaluate your headline are going to give this one zero points. I don’t care.

Yesterday morning I received an email from my best reader contributor. After giving some positive feedback, the bulk of the email contained a concrete example of where AI is doing some good. He is totally right. There are many examples of where AI is providing significant benefit. From what I can tell most of these cases involve special purpose instances, occasionally small language models, and/or hybrid AI implementations.

It would be interesting to read more about these types of cases, but they are hard to find. They are hidden behind the wall of AI hype I take aim at: hyper-scaled general purpose LLMs and their very close sibling LRMs. Everyone is shouting so loudly that really interesting stories cannot get through. This is another niche in the market waiting to be filled: doing in-depth factual analyses about real AI success stories. Just know that you have a monumental uphill marketing battle.

I am hoping we will be reaching peak peak hype about AI soon. No, that is not a typing error. Peak hype is part of the hype stage in the technology / product innovation lifecycle, that I first studied back in the 70s. In my lifetime there’s never been so much hype about anything else, so that’s what the second peak is about.

I started on my path of a technology career in 1970 when I took programming in high school. It was the first time my school had ever offered this course. In my first job, I helped a national bank move from manual to automated transaction systems. I still remember the fear in the eyes of the bank’s customers when we informed them that the manila cards on which we typed their savings account activities and the tiny books where we wrote by hand their updated balances, were being replaced by a large, chest high, benign looking computer, that could’ve doubled as fancy looking file cabinets1.

I then spent another 45 years putting in large technology implementations where my role was primarily strategic, in terms of selecting the appropriate mix of technologies, the implementation approaches, and user adoption strategies. I also managed very large projects and built two consulting firms whose sole purpose was helping business organizations implement technology.

Throughout this time I had a front row seat watching people - from all walks of life and vocation - and their various reactions as a wide range of new technologies was thrust upon them. As an example, I had an important role in the first fully automated stock exchange nearly 40 years ago, which was the initial crucial step in allowing us access to the zero commission online trading of the 2020s.

I started this newsletter because I was concerned with the variety of negative effects that I saw on individuals, and society writ large. I also believed that these negative affects interacted with one another, and compounded, creating a growing entanglement of thorny issues. I purposely chose to write in a satirical, sarcastic and occasionally humorous voice to poke pins into people’s balloons about tech celebrities, flawed technology products, and most importantly, the profound human behavioural impacts.

But I fear that it might be too late for all of us. Let me explain why.

One of my retirement projects is establishing a new FM community radio station on the little island that we live on. Yes I understand, what a stupid idea in 2025. On Saturdays I have the distinct pleasure of working with an enthusiastic 13-year-old volunteer who, in addition to being highly skilled and courteous, is wise beyond his years. He has tech savviness built directly into his DNA. But I find that he has a much better perspective on technology issues than people many times his age.

This week we had an enlightening conversation about the sweep of technology change over the last 50 years. I told him that in my time of consulting, the greatest - truly the greatest – change that I saw in my entire career is that most people have changed from technophobes to technophiles. We have gone from a society that was worried about what would happen to their savings if they weren’t on manila cards to one that will adopt literally anything technological in an absolutely thoughtless fashion.

He nodded and said he understood. He is thoughtful, considered, and slow to reach conclusions. He’s spent a lot of time reading about technology change and even though he probably doesn’t have the slightest clue what manual systems are, he does see the philia all about him. I felt an understanding from him than I don’t from many people who pretend to talk about technology, but without deep understanding or comprehension of its multiple dimensions. Another acute implication of the last 30 years of technology assimilation: unbridled expressionism no matter how far to the left you are on the DK curve.

Because philia2 means love, and I see so many people in love. Right now it is AI, but there are so many other tools being rapidly chosen without very much consideration at all. And when you’re in love you make many mistakes and take inappropriate risks. You don’t realize that your partner might abuse you, might cause others all sorts of psychological harm, might cause you to change your primary values and mental models without you even realizing it. That’s what love can do.

My entire role has gone from encouraging people to consider new technologies in a thoughtful and methodical manner to wishing thinking that we would fall out of love with technology, and use it in its proper role as a tool. A tool where you actually judge the pros and the cons, costs (both personal and societal), long-term impacts, etc. A tool to help you get back to life or improve your life.

Not become your life.

We used to have an expression that a person couldn’t see the forest for the trees. But now so many people walk around staring at their phones, navigationally challenged without their GPS map locators to guide them. So it’s only when the map makes an error that these people accidently wander into a forest. But instead of seeing the forest or even a tree, they look up briefly from their phone and see the little elf door. They open it and go through.

From reality to fantasy.

Because in our collective journey from phobia to philia, my other critical observation is that we’ve gone from mature adults to immature children.

And that is why I fear we are too late.

MENU MISTAKES

Did they show up in your Spotify recommendations?

This important research was spared DOGE’s scientific spending shears.

SPECIAL DISH : Calling BS on recommendation engines as an AI success story

Friday I read two posts that referred to recommendation engines as a success story for AI. I felt uneasy about this conclusions. Just as I had 15 years earlier when a client asked us to put together a recommendation engine as good as that for Amazon books. My partner and I took down his requirements - which were very challenging - and we explained to him what we could do with machine learning. We also chatted afterwards saying, “do you actually find these recommendations very helpful?” Neither one of us did.

Later on Friday I had two interactions today’s evidently AI supercharged recommendation engines that caused me to write this little dispatch. My wife and I always have a good end of week dinner on Fridays and we love to listen to jazz music as we eat and talk. During the cheese and wine course my musical selection had ended, and the AI kicked in. It was raucous, uptempo, unmelodic jazz selection of the type that I’ve always disdained. I have hundreds of jazz albums in my collection along with dozens others in the upmarket streaming service I use called TIDAL. So shouldn’t this great recommendation engine, which is supposed to have state of the art pattern recognition realize that I hated that kind of jazz?

Evidently not. Yay-sayers will say that it is just a bad recommendation engine. (Post script. At our Sunday dinner [roast salmon, greek potatoes and BBQed vegetables since you asked] my wife totally agreed that neither Roon or Tidal [our music technologies] ever recommend songs or artists she wants to hear. Both of our dinner guests said the same thing about Spotify)

I then had cause to use the much touted recommendation engine from Amazon. Surely this would be a great experience because people that believe in its recommendation engine are like monks whispering in sacred gardens about something spiritual. If anything, it was worse than the music selections. It repeatedly could not remember that I was looking for a studio quality Headset - microphone combination. It recommended many things other than that. In the end, I went to Perplexity to help me with my choices.

Got to a short list and back to Amazon for a purchase. I was down to two choices and put them in one by one. The product that I was leaning towards was more expensive but I almost decided to buy it. Then I found it didn’t even come with the cables to connect it to an amp. The Yay-sayers are smiling - the recommendation engine came through in the end. Big no. I had to find this out by reading negative reviews. Presumably the great AI driven recommendation engine not only reads through the specifications for every product in the Amazon catalog, but also should be able to ingest all of the buyer comments. But at the bottom where it says you should also consider buying one of these, as Amazon is always cross and upselling you, no cables suggestions in sight. None.

I’m pretty sure that this information is available in their product database. I’m also pretty sure I could still hack together a COBOL program to read that database and find which headphone / microphone products don’t have cables, in order to recommend them to people.

I then looked at Apple’s book recommendations. Fifteen books you might like, suggestions based on your reading activity. I have hundreds of books for them to determine these mythical patterns that I don’t even see. They batted zero. This enormous engine evidently couldn’t determine that 80% of my books were either non-fiction or essays. Instead current and old pulp fiction are the suggestions. Well maybe I should try, that in my seventh decade the AI has figured out that trite fiction is what I really will like.

On the other hand here is a list of mostly fiction books many which I would seriously consider. Written by an actual human being who has read them, and recommended them previously to others. Key thing to note, no marketing affiliate fees are being paid. Light coming on yet?

Cultural Capital

Absorbing novels to read on holiday

Hello, Welcome to Cultural Capital…

2 months ago · 132 likes · 69 comments · James Marriott

I couldn’t turn to another vaunted recommendation engine, Netflix because I stopped subscribing to this streaming service long ago. They never got what I wanted right. Ever. Despite me watching hundreds of titles. I hate horror movies why are you suggesting them? As I wrote that down I suddenly had a memory of when recommendation engine did work. It was in the early 2000s when I was a lonely divorcee who went to a minimum of four movies a week. At the time I was using Yahoo and they had a fabulous utility, well ahead of its time, where you could rate each of the films that you had seen and interact with similar cinephiles. The best thing was that their recommendation engine for upcoming movies was surprisingly accurate. But isn’t this like over 20 years ago before we had fabulous AI built into these things?

Then it all began to change. Yahoo, which made a series of strategic blunders at this time, started recommending movies I would never be interested in. I did research and found out that they had decided to change their engine to recommend what their advertisers were paying for, not with their users were really interested in. So no recommendation for the latest Spanish movies which we know you love, because Weinstein productions pays us big bucks and you need to go see Kill Bill instead.

I couldn’t give up. Why do people believe so much in recommendation engines. I did some research about the quality of AI in these engines. Here is a small extract I read about the power under the hood of these AI engines:

“Utilize advanced algorithms such as collaborative filtering (including K-nearest neighbors), content-based filtering, matrix factorization, and neural networks, depending on the specific needs and available data”

So are you keeping up with this? Can you define matrix factorization? For it is the real answer to this dilemma:

I think it means forget about what our customer really wants, just make sure they never leave our platform and they buy something. Lots of somethings hopefully!

Consumer vs commercial interests, that is what is at the bottom. I think that there are engines that can probably be great. Yahoo did it in 2003/4. But other factors are in play. Amazon knows you are going to spend the bucks and you can always return the stuff easily. They don’t care about all the effort in the buying / return endless cycle. There are bigger business factors:

Engaging you.
Tracking you.
Selling to you.
Never letting you go.

I guess there are also people who don’t have strong preferences, and who are overwhelmed by the act of choosing. Scott Galloway - whose opinions I value - wrote about having really no preferences for his clothing, his liquor choices, hotels, and so many other things (except they needed to be expensive). Evidently he just doesn’t care. So maybe it is useful for people like this.

But I think the act of choosing is very important. It is where you express your values, beliefs, and preferences. Where the rubber really hits the road. The opposite of a virtue signal; it is virtue engaged. Like us Canadians resolutely not buying from the US as they try to destroy us, even though they are the only source of quality lemons and digital services.

In the end, I feel that it’s just one more subconscious technological implication that we don’t explicitly see. Have we so lost our us-ness in favour of tech-ness, that we now just do what our technology tells us to do?

But I could be totally wrong. What do you think about recommendation engines?

A LITTLE SPICE

And now we have big companies controlling the space, who have commercial incentives to move fast and not think deeply about what their products are doing and what effects they'll have on the world.

Melanie Mitchell