AI and IP: A shift worth thinking about.
The relationship between intellectual property and technology has never been entirely comfortable. Each wave of technological change tends to stress-test the existing frameworks. Sometimes stretching them, sometimes breaking them, usually producing something new in the process.
The move from hardware to software was one such moment. For a long time, IP law wrestled with a deceptively simple question: if something that used to require dedicated circuitry can now be expressed as instructions on a general-purpose platform, does that change what it is, and how it should be protected? Different jurisdictions took quite different approaches, courts disagreed, and the debate is arguably still running. It was, in retrospect, a lateral shift: a function moving from one layer to another, but broadly the same function.
AI feels like something different to me. By “AI” here, I mean model-based systems whose performance depends not only on code, but also on training data, compute and deployment context.
What strikes me is that AI isn’t really “more software.” Many AI capabilities emerge from the interaction between several layers: the model architecture, the training data, the optimization process, the compute environment, and the way the system is deployed.
That matters for IP because the valuable capability may not sit neatly in one place. It may not exist only in the code, only in the model, or only in the dataset. In many cases, it is the combination that creates the result.
Take a vision AI system used in a technical workflow. The underlying architecture may be well understood. The training data may be proprietary. Performance may depend on specialized hardware. The commercial value may come from how those pieces are brought together into a product or standard. If the result is genuinely novel, where exactly does the invention sit?
That, to me, is the real shift. AI doesn’t simply move an existing function from hardware into software. It distributes the inventive contribution across data, models, compute, and system design.
That old IP questions don’t disappear. They simply become more complicated.
The old hardware versus software questions don’t go away. They remain, but a new set of questions layers on top:
- “What is the invention when value emerges from the combination of model, data, and compute?”
- “When the same architecture produces very different outputs depending on training data, where does the IP sit?”
- “What does inventorship mean when part of the creative or inventive process was contributed by a system that has no legal standing?”
I don’t have tidy answers to any of those. I’m not sure anyone does yet, and I’m wary of anyone who claims otherwise.
What I do think is that this is a topic worth exploring now, not from a purely legal perspective, but from the perspective of how technology actually gets built and how businesses create value from it. The IP questions are real, but they are also downstream of the commercial and technological decisions that are being made today.
Over my next few posts, I’d like to do exactly that. I’ll share some thoughts on a few specific angles, starting with how AI is intersecting with technology standardization, before looking at what it might mean for licensing models and for how companies think about IP as a strategic asset. Not as a set of answers, but as a way of thinking through the questions together.
If you’ve been wrestling with any of this, whether from a legal, commercial, or technical angle, I’d be curious to hear how you’re seeing it.