All this is happening on the assumption that AI will revolutionise how humans work, earn, and live. Yet it is impossible to think of a time when the promise of a technology, as outlined by its adherents, was as disconnected from the reality of how people experience it. We constantly hear that AI is close to transforming every aspect of human life and yet it remains, for most people, a collection of narrow conveniences and hidden algorithms.
In some specific and bounded fields, the potential of the technology is expressing itself more fully. Unsurprisingly, given it was developed by Silicon Valley, the innards of the tech world seem to be one such place. Basic coding tasks can now be fully automated. Apparently you no longer need to know a programming language to, say, turn out a basic app all on your own. But, frankly, how many humans alive need to produce an app?
Most users actually treat AI as either a replacement for increasingly unusable search engines — or, for those most embedded in the technology, as a virtual companion. It is true that the latter has already taken on troubling dimensions for some people, but it is not yet a disease affecting all of society.
Over the past six months, meanwhile, AI itself has progressed — according to its adherents — by leaps and bounds. The development of “agents” — who can move through the virtual world or your own desktop, supposedly as a replacement for you, doing more complex and interconnected tasks — is one such advancement. Computers using computers is a big step forward for the field, but it has barely changed the way most of us use AI, which has barely changed since 2003. A model might get significantly more powerful, using far more semiconductors, energy, water, and other resources, and yet the user experience for most people does not shift discernibly.
Agentic AI, for example, was supposed to have made personalised digital assistants possible. It should in theory be able to organise your inbox and your life, follow up on your tasks, and run your digital errands. But, even in corporations that pay for the tool, it has become clear that very few of their employees use these most advanced features. According to the last global workforce survey by PriceWaterhouseCoopers, only 14 per cent were using generative AI daily, and only 6 per cent were using the agentic version.
In multiple user-facing fields, AI isn’t quite living up to its supposed potential yet. Self-driving cars are nowhere near ubiquitous. In medicine, a recent survey by Elsevier noted that its use for any particular clinical activity remained quite low. It has not yet taken over classrooms or even remote teaching.
In some of these fields, of course, it is working far better behind the scenes than in front. In medicine, it has already worked to massively improve drug discovery. At the highest level of academia, it has begun solving mathematics problems that have puzzled the world’s greatest minds for years. YouTube and Spotify are inundated with poor-quality AI-generated content, threatening genuine creators.
But if early technologies under-promised and over-delivered — electricity and the smartphone’s valuations and expectations rose alongside their use — AI is over-promising and has under-arrived.
Its promoters might still be right about its transformative effect in the long run. But we are not there yet. Should we be entirely surprised at this? Is it not entirely predictable that a technology developed by and for an insular world of coders has transformed their world, but is struggling to be adopted by the one in which we all live? It is clear that it is not developing the algorithms, which is the really hard task. What will make the crucial difference is how well they are deployed and disseminated — and, for that, I don’t think Big Tech has the answers or approach.
 "Artificial intelligence, AI")