While working in the public sector, through advanced data analytics and by combining data from different authorities, we were able to describe phenomena such as social exclusion or identify datasets that illustrated, for example, a city’s attractiveness. By bringing datasets together and connecting them with meaning-making enabled by language, understanding emerged. Based on that understanding, it became possible to consider actions and develop strategies. At the same time, I began to grasp the significance of coding. People who were able to create code could create entire worlds or explain them.
We are living in a moment when the skill of coding is no longer limited to a few. It is in the hands of almost everyone of working age. World-building in this sense has suddenly become more equal. Language is code. It is the medium through which the core features of our identities, cultures and values are transferred and mediated and at the same time the place where they are formed and continuously negotiated. Language differentiates social classes. Mastery of local language is the key to real integration whether it is the language of a sector, everyday language or the way a group of friends speaks to one another. One of the central ways teenagers distinguish themselves from parents or younger children is by inventing concepts used among peers whose meanings others do not understand. Language also involves subtle nuances and embodiment through which the final meaning of words is conveyed. The meaning of beautiful words changes if they are spoken threateningly. Through language, inherited knowledge has been passed from generation to generation for hundreds of years.
The development of language models represents an extremely significant transformation in AI. Language as an interface is deeply human and therefore enables broad groups of people to access and make use of AI. It is an interface that makes it possible to communicate with massive datasets. It is also an interface through which top experts can become even better and through which good professionals can become excellent. In all communication involving text produced by language models, meaning-making remains a human task. In order to create something good, one must know what one wants.
Nobel laureate Bengt Holmström has said that his most important task as a teacher is to help students delimit the problems they are solving. To create a framework within which the problem is addressed. When the whole world is not open at once, it becomes possible to say something. But in order to delimit something, one must already know a great deal. The same applies to language models. If one wants to use language models to produce, for example, a viable workplace well-being plan, one must know a surprising amount about the sector, about what creates strain in that sector, about what human well-being consists of in general and about the role of work within it. One must also be able to provide the language model with sources it can use when creating the plan and know who knows. In other words, one must identify the characteristics of the phenomenon and the related datasets.
The next phase in the development of language models is agentic AI. It can be assumed that in this phase, agents will be integrated into different parts and processes of expert work. This shifts the focus of language model use from the individual professional to the organizational level. It implies a need for strategic planning within organizations. In some respects, this resembles what was happening in factories at the beginning of the twentieth century. The skills of individual craftsmen were mechanized. In order for work to be mechanized, it had to be broken down into parts. A machine could not produce an entire product at once like a carpenter but instead produced parts of the product that were later assembled into a whole.
In this societal process and transformation of work, the question of language is important. Because language reflects values and forms of appreciation, it is not irrelevant which language models and whose language models we use in our organizational processes. I hope that, for example, the AI agents used in the Finnish public sector reflect the Nordic values that we have been shaping for decades through our own distinctive language. I also hope that organizations take seriously the opportunities of the next phase of AI development and reflect deeply on the changes this new phase will bring. I have noticed that identifying dilemmas in new environments is difficult. We must also discuss what the industrial revolution brought about by new technologies means today in modern work.
Human decision-making contains, in addition to its visible layer, much that is hidden in the body. Behind a wise decision there is often, alongside explicit knowledge, embodied, intuitive and experiential knowledge and memory.
The GPT-Lab is also reflecting on this question.
About the author
