Why AI people don't get us ordinary people
Dear
Quite a few of you have shared some of your existential dread about AI, and I realise that I may have contributed to some of that. I want to try to redress the balance by mixing that editorialising with some more going back to the original traditions of Loose Wire, a quarter century ago, when my mantra was: "tech needn't be scary, if only it was explained better, and with the mindset of ordinary users."
To be sure, things have changed -- there are things to be scared about -- but perhaps I can help by sharing more about what technology can be used for, along with some more specific pointers.
I'm still here on the user's side, but I'll suggest some products and how to use them alongside the more pompous big-picture stuff.
Let me know what you think, like, want or hate. I'm all ears.
In this spirit this week I focus on my boiler, how search has failed, and why Google, Apple, Amazon and others have lost the plot.
Jeremy
Why AI people don't get us ordinary people
Sometimes those working in or promoting the latest generation of AI misunderstand what the ordinary user needs. And in doing so undersells generative AI's best features and oversells its unpolished, or undercooked ones.
Let me share my most recent experience in about as humdrum a scenario as you can wish for: fixing my (and I'm going to get technical here) boiler.
Picture the scene: I'm waiting for the shower water to warm up and even in my bed-addled state I can realise it's taking a lot longer than it should. Eventually I gave up and take a look at the pilot light on the boiler. It's flashing red, or whatever pilot lights do. I press a few buttons and it goes green for about 10 seconds and then goes red again. I remember vaguely our plumber talking about system pressure. But that's about it.
In the circumstances I turn to my trusty wife for help. I'm kidding: it's way too early in the morning for her. So I ask my phone, reading out the brand name, model and the error code. My phone shoots back:
The A04 error code on a Vokera Easy Heat Plus boiler usually means low system pressure. To fix it, you typically need to top up the system pressure by adding water via the filling loop until it’s at the recommended level, usually around 1.2 to 1.5 bar. Check for any leaks in the system or radiators as well, because those can cause pressure drops. If you’re unsure or the problem persists, it’s best to get a qualified engineer to take a look.
Filling loop also sounds vaguely familiar but I don't want to start a fire. My neighbours already have a pretty low view of my technical skills. So I turn to my phone again. What would a filling loop tap look like?
The filling loop tap is usually a small lever or a flat-headed screw valve on the pipes under the boiler. It’s often silver or black, and you turn it to allow water into the system. Once the pressure gauge shows the correct pressure, you close it.
Two minutes later the house blows up.
Actually, no, it doesn't. The pressure gauge goes back up and we're in action. Hot water flows through the pipes and my heart. We're back!
Now, none of this may surprise you. You may be familiar with filling loops (hello, Ian!) But it did me.
All this was done on my iPhone with Perplexity.ai, which I've found increasing useful (with caveats, below). It understood my spoken question, and delivered prompt and helpful responses via both voice and text, suiting my predicament perfectly.
To which I have two questions:
- Why are none of the big players here already (and I mean years ago)?
- Why do AI founders and evangelists think this is a misuse of AI?
As evidence I'd like to present a recent podcast from Siliconsciousness, an interview with Founder and CEO of AutogenAI Sean Williams. It's worth a listen, though I found a lot to disagree with. One is pertinent here. Sean at one point says:
People are still using large language models as search engines, which is a terrible use case.
Which is odd. And made me realise the two above questions are key to understanding where we might go in the next few years.
The big dogs that didn't bark
Why is Siri, or Alexa, not doing what Perplexity is doing? Partly because both Apple and Amazon screwed up. Craig Federighi, SVP of Software Engineering, gave a wooly answer to a direct question from from WSJ's Joanna Stern which suggested they really dropped the ball. Amazon, I can only assume, is just not interested in Alexa answering any questions except for "Can I order some more toilet paper?"
But the big loser here is Google. Google is getting there, but it has clearly not only dropped the ball, but got completely obsessed with monetising search, since the enshitification of its search engine product -- the thing upon which all profits depend -- predates the rise of LLMs.
Theirs is a deeper sin, however. They had plenty of time to figure out what exactly a search engine is. But they've stuck with their 1998 mission statement, “To organize the world’s information and make it universally accessible and useful”. Which should apply here, except it doesn't. Since 2015 they've called themselves an AI first company, but that too seems to have gotten lost. If I was cynical it's because they've focused so much on monetising the user they forgot to continuously hone in what that user might want.
To me it's obvious, and I'm sure it is for you too. A search engine is not a search engine; it's a solution engine. It's: *I have a problem, and I need the answer. The name of a hotel in Kabul. The kind of weird bird I can see in my garden. The boiler that doesn't work. *
The road out of town
To get the right answer I would find a way of phrasing the question, of phrasing the search term, so that the result that came back was the closest one. Then I would go through it again, asking what it looked like, and I would probably end up with a YouTube video. I'm counting on the search engine to do a faster job than if I called up customer service (and after waiting forever would tell me or send somebody out to fix it.)
Perplexity.ai does this in a flash. With voice. So, in other words, Perplexity is a solution engine, if you like. This is where AI can do so much better than existing search engines and the old paradigm of learning how to frame a search term in order to get the result you want, and then having to make a decision of choosing which of the answers to come back with.
Perplexity has solved that problem. And, I would say, put Google on the road out of business. And, with due respect to Sean Williams , highlighted the power and usefulness of LLMs for most of us ordinary people.
The usual caveats apply: It's imperfect, and it should not be relied on for fact checking, writing essays, or getting references. You always need to assume you'll only get 63% of things right. So don't use it for life and death stuff like CPR or fixing a boiler (er, isn't that your whole point? - Ed).
And if you're happy with that, then go for it. But you should be careful, because it can ruin a reputation in a second.
So, credit to Perplexity. And Google, you need to watch out. And AI developers, thinkers—stop overthinking what AI can do and start thinking about what people are frustrated by. You might find a channel to monetization that could make your long-term dreams better funded, and have a better chance of fruition.
And the second question? Why do AI luminaries think a search engine is a terrible use case? I think it's because they haven't taken time to figure out what a search engine actually is, and to understand that users are not all productivity-at-all-costs people, and that we just need a helping hand to get through the day. Sure in the future AI may supersede us; but right now we want to have a shower, get breakfast, go on a school-run, pick up milk from the shop and gently wake our spouse at some point before midday.
In the meantime, my advice: download Perplexity and play with it for a bit, if you haven't already. I would probably say it's the most accessible AI app I've played with which actually does something useful.
Tools used:
(No advertising, free copies or payment of any kind involved)
- Solution engine/research: Perplexity (Free tier allows unlimited quick searches, 5 pro searches a day on its standard AI model)
- Writing and editing: Ulysses MacOS and iOS markdown editor, part of Setapp or $6 per month
- Dictation and transcription: Voicenotes web or mobile, $15 per month, $90 per year
- Transcription: MacWhisper MacOS, iOS transcription app, free version, Pro $60 one time
- Transcription tools: Bearly Free version, Pro $20 per month
As usual, please do share your thoughts: How are you using AI, and how has it worked out?