The idea of expertise: an evolution in three layers

Layers of expertise

In Plato's Republic, Socrates considers what knowledge and skill a person needs in order to be a just and good ruler. In the Ion, he asks a rhapsode what the skill of an artist actually is. The question of the nature of expertise is thus as old as philosophy itself.

The Delphi method was developed at the RAND think tank as a tool for situations that call for fast yet considered decision-making from many angles. Such situations arise especially in crises, where social hierarchies threaten to narrow judgement. The COVID-19 pandemic repeatedly brought situations in which pressure for quick decisions pushed aside perspectives seen as secondary. It was against exactly this kind of backdrop that the basic features of the Delphi method were formulated.

Whatever their purpose, most variants of the Delphi method share three features. The first is iteration: the process gives participants the opportunity to develop a shared understanding in interaction with others, and both individual and shared views often change along the way. The second is anonymity, so that factors other than arguments do not steer the process. The third is expertise: panelists are selected because they represent expertise or a stake that is essential to the phenomenon. This essay focuses on the third feature, because the understanding of expertise changes over time, unlike the ideas discussed in the earlier essays of the series. A Delphi study typically uses a panel of 15–40 experts, whose material is examined qualitatively. Far less often, panels comprise hundreds or thousands of respondents, in which case quantitative methods dominate the analysis. Large panels have been assembled for technology foresight and, increasingly, for the participatory development of communities such as associations and municipalities.

The standard dictionary of modern Finnish describes an expert as a person with more thorough than usual (professional) knowledge in some field, especially one invited to give a statement. The definition suits the Delphi method well, but it is individual-centred and static. A common distinction is made between the expert, who knows the solution, and the specialist, who is able to solve the problem. The opposite of an expert is a layperson, and the opposite of a specialist is a generalist. Most people are experts in some fields and laypeople in others. A Delphi panel is chosen primarily from experts and specialists, but generalists have an important task: they view the phenomenon in a wider context, from outside the interests and the narrow system of the phenomenon.

Expertise also has a counter-trend: it can begin to reproduce positions and power. The knowledge of the layperson, regarded as the expert's opposite, should not be underestimated. The concept of the expert by experience reflects the rising value of first-hand experience. Parents may not be experts in education, but about their own child they know more than anyone else. That is why, in a Delphi study, expertise is assessed in relation to the phenomenon under study, and alongside expertise, having a stake is recognised. Having a stake has another function as well: closeness to the phenomenon gives motivation, feeling and energy to form views, justify them and take part in discussion.

Negotiating and learning expertise

Expertise is increasingly shared in networks through which both individuals and groups develop their capabilities. This development will hardly remove the importance of specialised knowledge, but it does break down closed and hierarchical professions into multidisciplinary networks, in which an important additional property of expertise is the ability to connect with other kinds of expertise. Knowledge is then not a static store; it is inseparably tied to human cooperation, practices and language. Recognising and guiding the group dynamics of the panel is therefore an important facilitation skill.

What a panel achieves, and how creative it is, depends on more than narrow and deep knowledge. Besides specialised knowledge, an ideal panel has systems knowledge, an understanding of the connections and effects between phenomena, which requires generalist expertise. The importance of dialogue and communication skills has also grown as the internet has increased panelists' opportunities for direct interaction.

Expert institutions, on the other hand, are fairly stable and predictable. For the panel of his doctoral study on the carbon dioxide emissions of transport, Petri Tapio selected a little over ten organisations that he judged to cover the essential expertise and interests. The organisations were allowed to choose the individual panelists themselves, on the assumption that their members' views would be uniform. For the same reason, anonymity was dropped. The practice has its strengths and weaknesses: it favours familiar ways of doing things and distances the study from recognising a discontinuous future.

As early as the 1970s, Ivan Illich warned of a society that, as it specialises, alienates people from creative activity as more and more choices are outsourced to experts. In his view, a specialised order maintains dependence on experts and cuts people off from the ability to handle their own affairs. Illich called for peer and network activity in place of manipulative institutions. The internet and social media have been answers to his wishes, but they have also created new problems. Illich's thinking has points of contact with Robert Putnam's theory of social capital, according to which a culture of participation and mutual trust is strongly linked to wellbeing.

A narrow understanding of expertise can lead to one-sidedness if expertise reproduces an elite and argumentation is emptied into sophistic rhetoric. According to Juha Sihvola (1997), Socrates sought to show that the claims of those who presented themselves as experts were poorly justified, because they contradicted their other beliefs. Socrates accused the sophists of concentrating on winning others over, when what mattered was whether the claims were true.

The Delphi method is a social construction similar to the concerto in classical music. Both have rules by which they are recognised, but the form does not bind the content. A violin concerto cannot be written without a solo instrument, but the notes are free. The solo voices of a Delphi study rise from the phenomenon or the contested question, and solving the problem requires virtuosity in which dissonance has its own place. In the score of a Delphi study, a many-voiced future sounds even when harmony is sought. The movements of the concerto correspond to the rounds, and the orchestra to the panel, conducted by the facilitator. The conductor may not write their own notes, but they can inspire the players and influence the sound through their interpretation.

According to Kurt Lewin, human action depends on the world as people know it, not on the world as it is, and interaction between people is a central shaper of reality. According to social constructionism, ideas, concepts and memories arise in human interaction and are products of language. Many truths are formed in conversation, and meanings arise in each moment. Constructionism is different from constructivism, which emphasises the individual's active construction of knowledge. Whichever school one follows, interaction makes different understandings visible and shapes understanding towards either agreement or an understanding of alternatives.

Truth about the actions of people and societies is rarely absolute. Most often it is a negotiated outcome that has emerged in the interaction of people and institutions. This applies especially to knowledge of the future, which is conditional and plural from the outset. In a Delphi study these agreements are negotiated locally. A condition of genuine dialogue is that the words of all participants enter the discussion as equals, regardless of speaker or position. Arguments are raised to the position of authority, and presenting and weighing them is the task of a diverse panel.

From individual to community

The modern education system assumes that the learner is an individual. Their learning is monitored and assessed, and only an individual can graduate or fail. We take it for granted that a person's first person is the singular I. Yet this notion is relatively young, a product of the modern era. Earlier, the most important person was the plural we. That is still the case in many communal cultures, where a person exists primarily as a member of their group and is answerable to others for their actions. Even in the Middle Ages, the source of identity, apart from the elite, was the community of birth and the social estate. Collectives have now returned, but with new ideas and on new scales.

Increasingly, not only individuals but also teams, organisations and other communities are recognised as learners. Examples include professional sports teams and orchestras from jazz groups to symphony orchestras. What they share is a division of roles built on individual skills and serving the whole, organised forms of interaction, and regular assessment and development of their own performance. Schools have found it hard to establish forms of collective learning. The Delphi method is a potential tool for this, whether in a phenomenon-based project, as in the Climate Forge, or in the long-term development of a class community following John Dewey's model of the school as a miniature society.

James Surowiecki offers a different perspective on the wisdom of groups in his book The Wisdom of Crowds. According to him, the average of the estimates of a large and diverse group can be more accurate than that of a single expert, provided certain conditions are met: the group must be diverse, it must not be uniform with respect to any background factor, and its members must answer independently of one another. The conditions suit the Delphi method with reservations. Surowiecki's model is, so to speak, single-use: the estimate is reliable only as long as the members of the group do not influence one another. Once they start talking, the conditions no longer hold. The Delphi method, by contrast, is built precisely on interaction, and so its anonymity and rounds are a way of combining independent estimates with mutual learning.

Supportive intelligence

The whole of biological evolution so far has been layered into the human being, its latest layer being the folded cerebral cortex. Genetically, humans are largely the same as their closest relatives. In Jungian terms, our conscious understanding is like the cream on top of a pail of freshly drawn milk: beneath it lie the preconscious and the unconscious, and deeper still the undercurrents that unite all living things. Culture has freed humans from many of the constraints of earlier forms of life, but it has not severed the connection.

A thought, or meme, travels from mind to mind quickly compared with genetic change, which proceeds at the pace of generations. At turning points, beliefs can be overturned in weeks. Russia's invasion of Ukraine reversed Finns' views on NATO within a few weeks in the spring of 2022. Yet even the speed of cultural evolution is not enough once information begins to be exchanged, copied and modified by AI outside the brain. As always at such watersheds, dramatic alternatives branch out ahead. In the Delphi method, AI has not yet shown its dystopian side. On the contrary, conversational AI promises more and better.

The wild card of future Delphi panels is artificial or supportive intelligence, as Osmo Kuusi calls it. In one form or another, panels are joined by an expert whose computing capacity is many times that of the rest of the panel. Its raw material is information, which it can organise so that the human panel can use it in forming a shared understanding.

The possibilities of AI can be divided into three groups according to the stages of a Delphi study:

• Before the rounds, a language model is a sparring partner for the facilitator, who is getting to know the phenomenon, structuring and delimiting it, modelling its system, formulating theses and designing the panel.
• During a round, the panel can be supplemented with AI panelists, panelists can be activated with comment prompts, and the material can be analysed and visualised while the round is still under way, to prepare the next round.
• After the rounds, AI helps with analysing the material, reporting and presenting the results.

Many of these are already in practice. In Pronoia, the AI assistant Kastalia helps the facilitator with design and holds a preparatory conversation with each panelist before the first round. In the analysis stage, Kassotis opens up the material layer by layer. Interpretations and conclusions, however, are always made by a human.

There are two main reasons for using AI panelists. First, it is far from always possible to bring every interest and competence into the panel. Modern societies assume that all significant interests have an institutional representative, but the future situation is not the same as the present one. New factors, unknown today, may appear in the future, and the relations between old factors may change. Some interests are then systematically overlooked, because they have no representation in our time. Can an AI panelist give a voice to such silent parties, say an unborn child or an expert who will be unemployed five years from now? Second, AI can represent people whose thinking it knows well but who are not actually available. Risto Linturi carried out such an experiment early on: his AI-simulated discussion included, among others, Elon Musk and Greta Thunberg. Kari A. Hintikka has studied how a facilitator can shape an AI panelist to fill the gaps in a human panel. Conversely, AI can be used to assess how well a chosen panel covers the necessary expertise.

Hintikka's experiment has also considered whether it makes sense to bring people from the past into futures research. My own tentative view is that history has personalities about whom there is plenty of material and who still have much to offer. The point is not so much the factual knowledge they absorbed as the content and consistency of their thinking and their ability to recognise the mechanisms and motives of phenomena. In human action, as in the dynamics of nature, much recurs in varying forms. It is a matter of grasping similarity, not sameness. The present is always narrow, and an AI that crosses eras can bring breadth to it.

In Pronoia, AI panelists are always clearly marked, and the material they produce is kept separate from human responses in the analysis. Open questions of principle remain all the same. Can an AI panelist engage in dialogue without distorting the process? Can it listen, or does someone else, such as the facilitator, listen on its behalf? How much should it comment? Hintikka has examined which kinds of AI comments can deepen, broaden and enrich discussion: new perspectives, new information, making ideas concrete, bringing contradictions to light, supporting and developing others' ideas, and questions that prompt reflection. Early experiments tried to make the AI panelist as human-like as possible. Another direction is to identify the differences and ask what benefits they might bring. Human choices are influenced by collective and individual interests, attitudes and temperament. AI does not have these by nature; they have to be built in separately. What if precisely their absence made AI participation valuable?

Experience is still being gathered on how AI stretches to processes whose outcome is unknown, when the information it uses is entirely information about the past. In long processes such as futures barometers, AI's potential presumably grows as material accumulates. AI, like expertise, has its limitations. AI can argue and recognise arguments, but thinking is another matter. Thinking is expected of an expert, though it can sometimes have very narrow limits.

The qualitative aim of the Delphi method is to identify the arguments for and against different alternatives that matter for the development of a phenomenon. An argument can be broken down into a claim, a justification and underlying assumptions. In a Delphi study the claim is given, but the panelist decides whether to take a position for or against it. For example, the claim that Russia would not use nuclear weapons even if the war in Ukraine led to its defeat can be justified by the argument that a nuclear war cannot be won. The underlying assumption is that all parties' decision-making is a rational weighing of risks. Game theory, however, also knows the so-called madman strategy: the fear of unpredictability can be exploited rationally. It may be that Delphi studies will increasingly examine the assumptions behind differing views, and AI can help with that.

Or should we speak of supportive intelligence? Osmo Kuusi's term may be more apt than artificial intelligence at this stage of development, in which intelligence outside the human being is still defined by its creators and subordinated to human goals. The European Union adopted its AI Act in 2024, and the debate has considered not only the rights and obligations of people and companies but also what rights and obligations might belong to AI itself.

AI, thinking and Bildung

Preparation for upheavals emphasises general meta-skills. An old but wise comparison from development cooperation applies: it is more sustainable to teach people to farm than to bring them bread, even though need sometimes forces both. Literacy remains an important futures skill, but more is needed. The concept of general education, Bildung, has begun to give way to talk of key competences, which fits better with economic discourse. There are good grounds for emphasising key competences, but mastering them is not yet Bildung. Bildung emphasises a person's independent path of development and the pursuit of a balanced relationship with oneself, others and the environment. Key competences emphasise skills of understanding, project work and interaction, which are useful when companies pursue profit and nations success. Bildung belongs more to the world of open speech and Habermas, key competences to the world of the open society and Popper.

In the Nordic tradition of liberal adult education, Bildung is thought to be built through self-education and communal activity, and its aims are democracy, social cohesion and people's inclusion. According to Seppo Niemelä, Bildung is a three-stage process: a person learns new knowledge, skills and capacities, forms an independent and creative relationship to them and begins to use what they have learned for the common good as a citizen. The result of the first stage is a learned or competent person; only the next two produce an educated one in the full sense. Learning is the absorption of ready material; Bildung is unpredictable, self-directed inner growth that needs pedagogical support in its early stages.

In an essay in Helsingin Sanomat (11 February 2024), the novelist Joel Haahtela described a development in which people are giving up their ability to think. Information floods in and demands an immediate reaction, and there is no time to process what has been seen. A reaction is an emotion, and emotion is faster than thought. As reacting increases, thinking begins to be replaced by AI, which in Haahtela's words is "passionless, eternal and always awake". In his view, the disappearance of thinking exposes people to emotional manipulation and to disinformation, which accumulates as they move from one reaction to the next. Thinking is also replaced by experts who are invited to comment on everything and onto whom people project the hope of rationally controlling a chaotic world, and especially its future. It is essential to recognise that thinking does not mean being right but searching for what is right, which is often a messy process of erring, correcting and reorienting.

The origin of the Finnish word for Bildung, sivistys, sheds light on this. It is related to the word siveä, decent, and in dialects the verb has meant tidying, smoothing and finishing. Flax was also "civilised" in this sense: when it was cleaned, it was combed one more time so that the result was a clean fibre. In a pedagogical Delphi, thoughts are combed clean together. A learning conversation clears space for new concepts and understanding. Words do not merely describe reality; they also create it. Language unites, but it also divides, and in interpretations of the future both are always present.

Humanity's first orders were based on direct observation. Later, the order of nature gave way to the order of God, and in the Enlightenment the ideal of the educated person became human knowledge, rationality and reason. This created a demand for the kind of learning and education we still favour. I believe Bildung is developing in a direction in which thinking, and not just thoughts, is required, and in which, alongside individual learning, the importance of collective learning is better recognised.

Bildung is a creative process in which a person shapes themselves and their cultural environment, and which contains the idea of a way of life better than the existing one. The hot question in this search today is AI. It does not replace thinking or the need for Bildung. The future cannot be derived from the past, of which AI is, at least for now, a prisoner. Our understanding can be described as layered, as Sohail Inayatullah does in his Causal Layered Analysis (CLA). At the litany level, we act as is customary. At the systemic level, an ideal arising from the prevailing system is pursued, which favours a goal-oriented learner who performs what the system expects. A learner begins to shape their own worldview only when they tie their learning to self-directed knowledge, interaction and thinking. At the level of myth and metaphor, a learner can recognise structures and patterns that describe a reality for which there are not yet words.

An important motive for this essay series has been to identify the intangible ideas that one can imagine surviving even radical upheavals. Ideas, or memes, are not objective entities; they compete with one another as genes do. In that sense the series is also a defence of the memes of the Delphi method. The method itself is merely a vehicle for ideas, but at present the Delphi method is one of the most capable vehicles on the journey towards futures.

The rest of the series

The evolution of expertise described in this essay draws on the Sri Lankan scholar Susantha Goonatilake's idea of three information flows, biological, cultural and external to the human body, which have separated and are now merging again. Goonatilake's thinking is described in more detail in the essay The idea of futures studies, which also presents other broad views of future change. Before that, The idea of the future extends the observation-based conception of knowledge to the uncertain knowledge that all knowledge of the future is.

Sources

• Goonatilake, Susantha (1991). The Evolution of Information: Lineages in Gene, Culture and Artefact. Pinter.
• Hintikka, Kari A. Tekoälyä Delfoihin [AI in Delphi]. Doctoral thesis (in progress), University of Jyväskylä.
• Hämäläinen, Juha (2016). Mitä on sivistys? Seppo Niemelä kansansivistysaatteen tutkijana ja tulkitsijana [What is Bildung? Seppo Niemelä as researcher and interpreter of popular education]. Sosiaalipedagoginen aikakauskirja 17.
• Illich, Ivan (1971). Deschooling Society. Harper & Row.
• Inayatullah, Sohail (2009). Causal Layered Analysis: An Integrative and Transformative Theory and Method. In Glenn, J. C. & Gordon, T. J. (eds.), Futures Research Methodology, Version 3.0. The Millennium Project.
• Kuusi, Osmo (1999). Expertise in the Future Use of Generic Technologies. VATT Research Reports 59.
• Lewin, Kurt (1951). Field Theory in Social Science. Harper.
• Niemelä, Seppo (2011). Sivistyminen [Becoming educated: Bildung in the Nordic tradition of popular education]. Snellman Institute.
• Putnam, Robert D. (2000). Bowling Alone. Simon & Schuster.
• Surowiecki, James (2004). The Wisdom of Crowds. Doubleday.
• Tapio, Petri (2002). The Limits to Traffic Volume Growth. Doctoral thesis, University of Helsinki.