Can A.I. be inventors?
The news that an AI system named Dabus has been refused the right to be named as the inventor on a patent application took me back to 1918. In his final book, published that year, Georg Simmel mused on the various transformations and continuities of modern life. One thing that struck him was how a series of technologies were radically altering perception, with some significant consequences. In particular, the advancement of the telescope and microscope had, he reflected, changed how the world was seen. The boundaries of the senses had been breached.
Not only did these technological appendages of the human senses change what was visible, Simmel also wondered if they would impact on what he described as the ‘structure of our cognition’. The changes in perception brought about by technology would lead us not just to perceive things differently but also, Simmel speculated, to think differently. Such technological shifts in ‘total organisation’, as Simmel puts it, required an adaptation of human sense-making. With the increases in the levels of spending illustrated in the above chart of world AI spending over recent years - which is likely to offer only part of the picture - in more recent years that restructuring of cognition has moved quickly and has mutated in software code. The investment in AI software in particular might tell us something of the wide-ranging attempts to codify cognition1.
As the Dabus case indicates, the move toward AI inventors is the next step in this long restructuring of cognition. The big difference is that such a step would not only append human senses and restructure cognition it would also seek to bypass human thinking in the process, leading to a more profound total organisation, to use Simmel’s phrase. This is not simply to say that the human is ‘out of the loop’, as it is often put. This is not just about algorithms making decisions outside of human discretion, although that would be part of it. It is that the actual creation of ideas themselves would be automated. Or at least that would be the aim. There is a step beyond this too. Having AI creating, inventing and even designing other AI technologies would mean that instead of these systems playing by an existing set of rules they would also have a role in organising or even creating those rules. The ambition and logic behind this is important to understand even if some of these aims might never actually be achieved.
With algorithmic systems, machine learning and artificial intelligence, what Katherine Hayles has referred to as the ‘cognitive assemblage’ continues to expand. Reflecting on these broader developments, Luciana Parisi has observed that:
‘what is at stake here is the automation of automation: the automated generation of new algorithmic rules based on the granular analysis and multimodal logical synthesis of increasing volumes of data’.
The phrase automation of automation emphasises the meta and organisational step that is underway. This is where AI would not just invent products, it would also oversee the development of AI. This is not simply automation, it is the attempt being made to automate that very automation. The patent in the Dabus case is not taking such a step, it is not about automating automation, this system was said to have invented a food container, yet it is suggestive of this type of step and of the ambition within this industry to give AI more creativity. Of course, we should temper this a little. The power that is ascribed to technological developments is to be questioned, especially as part of the appeal is to get there first and to hype the potential powers of these systems.
Another way of understanding this is to see AI as moving beyond responding to patterns and instead taking on the job of arranging the pieces. In a recent LRB article on AI futures, Paul Taylor wrote that the:
‘hope is that data-driven machine learning will be able to move beyond simple pattern-recognition and start to develop the organising theories about the world that seem to be an essential component of intelligence. AI software has, so far, only managed this in very constrained environments, with games such as Go or chess.’
The turn to AI developing ‘organising theories’ is, for Taylor, at the centre of the formulation of new types of intelligence. The reference point here is the ability of AI to master participation within actual games, yet it is in the ability to theorise and organise rather than just play the game that is where the attention is now being placed.
In the case of Dabus, the system was not used to invent other AI systems but to develop products, yet it would appear to represent a step in a particular direction. There is an active pursuit of the organisational properties of AI, the creation of AI that set rules and, crucially, an investment in what Parisi called the automation of automation. Even if AI are not granted patents, and are therefore denied the official label of being an inventor which is reserved, the judgement revealed, for humans only, it doesn’t mean that they won’t invent. It also doesn’t mean that the automation of automation won’t also arrive. This will then create all sorts of new questions about the future of discretion, judgment and creativity.
Erik Hoel, in a Substack provocation on the automation of art, points out that these types of developments in AI, which will be owned and controlled largely by large tech companies, will concentrate the ownership of creativity into a few hands (whilst also impacting upon how creativity is understood and realised). Perhaps the automation of automation will also raise further questions over who will develop and own such AI and therefore who will have most influence in deciding how the world is ordered and organised. It won’t just be about the named ownership of a patent, it will be a question of who owns the patents on the AI systems that are themselves generating new patents.
The above chart, which shows the ownership of active families of patents in the field of AI, gives an impression of where the ownership of AI more broadly is located. The energetic activities amongst these large tech companies - including Alphabet, Tencent and IBM - is central to understanding AI developments and is also then likely to be important in understanding what will happen with creativity and invention. If it comes about in the form that is intended, then the potential development of AI that offer organisational theories and automate themselves will be performed within this economy of patent production and within the power structures of larch technology companies. The structuring of cognition observed by Simmel continues, with AI it takes a more direct and purposeful form and is deeply embedded in emerging forms of capitalism and in the politics of intellectual property ownership.
There are no doubt some definitional issues at stake in the reporting of these AI revenues in the charts included here. What is included as AI spend will vary depending on how the classifications systems are set up. This in itself will be interesting to reflect upon, not least as the various terminology around AI and what constitutes it will be reconfigured over time. The impact here will be that estimates around financial spend will vary, but the use of a wider discourse and classification of AI will also be important for how such systems are developed, funded and implemented.