Reflections Before The New Year
University of Oxford
2022–2025 marked an unprecedented moment in human history. During these years, we witnessed the birth of contemporary AI in its modern form. And while AI has existed for decades, it has never existed like this: with the sophistication afforded by transformer-based architectures, the naturalness of human-like communication, and a realism that renders it a transformative technology rivalled only by the arrival of the internet.
AI has not yet reached terminal velocity; it seems to improve ceaselessly. Yet “better” here should be taken in a neutral sense: better as in technically more advanced, not necessarily better in a moral or social one. Better for whom? Better in what sense? Better toward what ends? These are questions that demand reflection, disagreement, and debate.
Take deepfakes, for instance. I have long argued that their emergence is socially significant. The challenge they pose is unmistakably socio-technical, not merely technical. The realism of modern deepfakes demonstrates the point I am making: although deepfakes have existed for some time, only now has their fidelity become almost epistemically indistinguishable from real content.
Their social salience is undeniable—offering both creative possibilities and profound risks. Yet the solutions being proposed remain largely technical. Will watermarking or labeling deepfakes as “unreal” truly suffice to ensure users understand and believe that what they are seeing is fabricated? Or is something more robust required?
I digress.
2026 marks the next step in this unfolding AI revolution. We will now need, more urgently than ever, to grapple with the technical demands and accelerating progress of this technology before it yields serious societal harm. But at the same time, we want to enjoy it, to marvel at its creativity, its utility, its almost magical capacities.
Is it possible to achieve both safety and wonder? Or must something ultimately give way?
I am, at present, unsure.