The other day I was driving my daughter to swimming. She was playing Pokemon Go on my phone and duly informed me that she had ‘swiped away’ a message from her mum. I tapped on the messages icon on my car’s display and tapped on my wife’s name. Siri told me I had two messages from my wife and asked if I wanted to hear them. I said yes. It read them out and asked if I wanted to reply. Again I said yes, then when prompted I dictated a message. It read the message back to me and asked if I wanted to send it. Once I agreed, the message was sent. This entire interaction was conducted by my phone whilst my daughter continued to play Pokemon Go on it. Granted my interactions were via CarPlay, but that is simply a secondary portal to my phone, all actions and interactions were managed by my phone whilst my daughter continued to use it for an entirely different purpose.
I relate this fairly mundane tale because in many ways it is awesome. Smartphones are incredibly powerful computers that sit in our pockets or our hands and let us have freedom and control of our lives in ways that were barely imaginable when I was my daughter’s age. For me this is exactly the kind of thing technology should be doing: allowing me to drive, stay in touch with my wife and entertain my kid all at the same time. The other reason I relate this is because I’m not sure what difference adding generative AI into the process would make. I guess it could offer suggested content for my message to my wife, but how much time is that likely to save me? I’d still have to select one of the options from the prompts it gave me, saving no time at all vs dictating my message. It certainly couldn’t drive the car for me whilst I text my wife back. Maybe I’m being unfair, or maybe I’m just unimaginative, but this is exactly the sort of area I would think that people who say generative AI is the future would suggest it will make a difference. The problem is it feels to me like in so many areas that difference has already been made.
Take ordering dinner. We’re constantly told that generative AI will make this easier. It will look in our fridge and tell us what we can make for dinner. Or ask us what we want, order the ingredients for us and give us a recipe when the ingredients show up. Or even somehow know what we usually like on a day like this and order it for us. Personally only the first two of those sound vaguely desirable to me and actually some combination of both would be preferable. Although I like going to buy my own ingredients and I’m not sure how much I’d be willing to pay for the recipe service when I can just plonk the contents of my fridge into Google and get recipes for free.
I recently bought a new phone for the first time in five years. The difference between this phone and my last one is notable, but not radical, probably mainly because they both ran the same operating system. Over the five years that I owned my last phone there were probably significant, if incremental improvements in the capabilities of that operating system every year. Elements of what used to be called AI will have been introduced and enhanced in things like the phone’s text prediction, voice recognition and image synthesiser (or camera as it’s more commonly known). Adding generative AI to these things can only make a further incremental improvement to them, equivalent at best to a full version upgrade of iOS*. I’d be interested to understand how much Apple spend on IOS development a year and ditto Google, Samsung, et al on Android. I suspect it is significantly less than is currently being poured into generative AI. To be clear that investment is estimated to be around $400 billion this year. This is justified by saying that generative AI specifically will revolutionise every aspect of our lives, yet we are presented with very few concrete examples of how. If there was an iPhone-like platform on which people could build apps, I could see how an ecosystem could form, but the financial models don’t work because that ‘platform’ (such as it is) doesn’t make any profit. In fact it makes billions of dollars of losses. People will point to Amazon or Uber and say that they lost money for years before they made a profit and that is true, but those companies’ costs decreased with scale, which is not the case with generative AI. OpenAI’s latest model was their most expensive to ‘train’ and costs the company more money per query than the previous model. The infrastructure costs that are being ploughed into AI data centres are not even one off, building the buildings is a fixed cost, but AI data centres run ‘hot’ pretty much all the time, almost literally burning through server hardware constantly. If an AI server lasts three years, it has done really well. All of this capital and infrastructure is being poured into something that is a marginal improvement in UX at best. It’s equivalent to all the biggest companies in America investing all their spare cash into Clippy or Clippy clones. If someone can tell me a functional, chargeable use for generative AI that isn’t either a marginal improvement of UX or an artificially cheap enshittification of a previously human delivered service they’re taking an awfully long time to do so. There was an article in the paper the other day where one of the UN climate directors said that AI was a potential problem, but also that it helped with climate modelling and solutions. The problem is he’s not talking about the same AI. There are specialised machine learning systems trained to model the climate or develop new kinds of medicines or myriad other useful applications, but none of these are the LLMs such as ChatGPT, that are designed as general purpose AI applications and are where the vast majority of the investment is going. In effect generative AI is living off other AI’s glory, almost like it’s some massive con or something. Nothing about generative AI justifies the obscene amounts of money that are currently being poured into it. It is not going to have an iPhone moment because the iPhone already had that. No amount of (extremely vague) input from Jonny Ive is going to change that.
In five years time I’ll probably buy another phone because the technology will have moved on sufficiently by then to warrant an upgrade. Some of those advances will be in what is broadly defined under the umbrella of AI and machine learning, but they will almost certainly have nothing to do with the LLM based generative AI that is where all the money is currently being poured. Honestly if the new features on my phone rely on some vast datacentre somewhere ‘thinking’ about an answer it would probably make me think twice about buying a new phone. I’m still close enough to the EU that I was able to turn off the ‘dumb rainbow’ (Apple Intelligence) in my current phone and I will fight long and hard to keep things that way. This isn’t an anti technology or anti progress stance, it is very much the opposite, I want technological progress that actually improves my life. Enshitification is everywhere in the technology that we are presented wiith, gradually making it harder for us to make technology work for us and easier for technology companies to make us work for them. Generative AI is merely the latest incarnation of this, where there are very few consumer value use cases, but the product has the potential to privatise even more aspects of our lives and then sell them back to us in a shitty format, at a premium. The only problem I foresee for generative AI is that in order to properly enshittify, a technology needs to be appealing and useful in the first place. If it starts off shit there really isn’t anywhere to go. Yet the tech giants continue to pour billions of dollars into building out the infrastructure that only really works for this specific technology (AI data centres are not fit for general purpose computing applications). How does this end well? It doesn’t, but even before the catastrophe unfolds we can start to think about what the alternatives could and should be.
I propose a new approach to technology. No more ‘democratising’, no more ‘disrupting’ and no more ‘paradigm shifts’. Technology should be a thing that improves gradually with the needs of its users. If necessity is going to continue to be the mother of invention, it should be the user’s necessity to do something useful, not the necessity of a massive corporation to make ever more profit (or in the case of the AI companies, attract ever greater investment). We need to find a way to identify technology that is genuinely useful and genuinely consumer focussed. Not as easy as it sounds. Earlier today I was reading a thread about how Corey Doctorow posts long threads on Mastodon. When someone pointed out that if you want to read Doctorow’s long threads as a single ‘article’ you can just add a filter, I thought “that’s a bit of a faff, not sure I can be bothered.” If I can’t be bothered to take steps to make technology work for me, what is the likelihood that someone less engaged is likely to? Maybe we get the technology we deserve and if we can’t be bothered to make a little effort in configuring our technology, then we get configured by it. It’s not like AI actually understands us or does what we want without us learning how to do the prompts ‘right’. The grief of setup and learning a new interface is not removed, it is just presented as an opportunity, rather than a burden. If we understand this, then maybe we can think a bit more about what that setup and interaction results in, what we expect from a technology and if we need it at all. You would think that after crypto, the metaverse and AI, we’d all be wary of technology companies telling us they’ve seen the future and we just don’t get it yet. Let’s hope next time we learn, let’s hope that people start to judge for themselves whether they need a technology or not. Some hope when there’s every chance they’ll have given away their last scrap of agency to generative AI.
*however, given the nature of generative AI, that it is impossible to predict an absolute outcome from it, the chances are that adding it will in some cases make the outcome worse or less reliable.
