The intelligence age is upon us

[trilight]

## Navigating Intelligence, Autonomy, and Anonymity


As the field of artificial intelligence grows in popularity, so do the implications of involving machine decision-making in every aspects of society, including global privacy, power repartition, and the nature of autonomy. Mass popularity comes with confusion, lack of clear definition, manipulation, and overall individuals bombarded by information in which the fundamentals. Understanding the potential implications for privacy and anonymity requires understanding the underlying power struggles at play.

## From Information to Intelligence

AI in its current state is only a series of statistical models of varying performance, with the goal of minimizing the measured difference between the output they are generating and the training input they were given. The higher the quality of the training data, the higher the quality of the output results. This stands true for image recognition, one of the earlier form of "AI", but also image generation and large language models. In other words, the power of a model will depend on the ability of its creators to capture the information flow. The logical conclusion is that we are already witnessing an internet scraping war, with large companies wanting to capture and extract as much value as possible from high-quality information flow. The cost of breaching privacy regulations and of capturing more information is overshadowed by the value they create, therefore putting private information at risk due to supply and demand.

This indirect attack on privacy is an absolutely essential fuel to the AI hype and will continue. Content produced innocuously between the early internet is now tremendously worth, and the value of this content will increase as it disappears from the internet, is captured, or restricted. While the quantity of content produced is ever-increasing, the quality is not, and particularly not when human-made content is becoming indiscernible from AI-generated content in appearance, only lacking the depth in substance. Information must be protected, encrypted, and erased whenever possible. Interest is now in the flows and the masses, not so much in the individuals and identification, but the latter remain absolutely possible and worrying.

## Future Governance

At the heart of discussions about AI is the question of who holds the reins of control. The naive assumption that technology can be designed purely to operate in humanity's best interest glosses over the reality that these systems reflect the biases and motivations of their creators. For anyone concerned with freedom and power dynamics, this is a troubling prospect. What choices will individuals have? Can they opt-out of their data being used for training? Perhaps. Can they opt-out of other people's data being used to take decision impacting them? No, our market economy will make sure that the most efficient decision-making takes places even if at the expense of fundamental rights.

The computing cost required for capturing and processing information flow is huge, leading to an accumulation of capital, expertise, and power in the hands of a few early, and arguably existing large, players. Individuals and small entities are limited to fulfilling niche for as long as they can before their outputs are stolen, copied, or bought by one of the large players: consolidation is the inevitable end-game of capitalism. In addition, the high cost and dubious economics behind training large language models will, in the future, make them inaccessible to those without the means to acquire them, leading to further inequalities in power and abilities. This in turn will further the ultra-centralization we are witnessing. Any concentration of power is detrimental to the individual, and to one's privacy and anonymity.

## Sentience and Autonomy

The moment at which artificial intelligences will become worrying, Matrix-style, will be with the apparition of emerging capabilities, or when the whole will be greater than the sum of its parts. A self-improving AI will grow exponentially, within a constrained energy envelope. When this sci-fi breaking point will be reached is only a matter of time, although it could still take decades. In the meantime, AI is just a buzzword for advanced statistics and prediction. Analogically, the training of a model is not so far from what biological beings do: adapt and replicate. The adaptation part is clear, while the replication part is for now handled by us human overseers. Is a bacteria sentient? There is no consensus. A bacteria reacts to inputs and acts (outputs) accordingly. At what point then is an AI model sentient?

What is clear is that artificial intelligence has the mean to enable autonomy, or behaving without active supervision. Autonomy is not something new, decentralized autonomous organizations are one of the latest examples. If the ability of a DAO could be expanded beyond if-then statements, the importance and potential of autonomous entities would increase. The processes behind the creation of a large language model are still open, and there are luckily players who have found interest in publishing their models as open. The use of these open, and often small-size models, could lead to new architectures of DAO and other autonomous models, fighting for privacy and independence. We could imagine an entity watching over data breaches warning users impacted, or a bot capable of relaying information to places where it is censored.