Data and AI literacy for everyone

2.1Review of existing frameworks

Heidrich et al. [3] present a competence matrix based on Ridsdale [4] with five areas of competence, individual competences, and an evaluation according to degree of difficulty. The competence framework on digital competences of the EU Science Hub (DigComp 2.0), on the other hand, places data and information literacy as one of four sub-competences of digital literacy [5]. In this context, the terms data and information literacy are not differentiated from each other, and some skills that are defined as core competencies listed by Heidrich et al. are missing.

In contrast, the ProCivicStat project developed a far more extensive conceptual model [6]. It relates data literacy or statistical literacy to contextual knowledge and differentiates more strongly between data forms that occur in the scientific environment and data forms, with which the citizen comes into contact. This leads to the question of the purpose of education (imparting skills for the responsible educational citizen or for specialist disciplines) that the framework should reflect.

Data literacy skills should enable their critical adopter to address and solve real problems, which means, context is important. In technical environments, where data is used, for example, to ensure the quality of production processes. In contrast, problems with a societal dimension require additional skills and attitudes like data ethics, measurability of phenomena, classification of results in contextual knowledge, or derivation of recommendations for action. The mechanism of collecting or procuring, evaluating, applying and interpreting the data appropriately regarding the relevant usage needs to be internalized.

“To collect, manage, evaluate and apply data in a critical way”, in Ridsdale’s words, defines a process that requires data literacy to master. This process should be represented as a cycle: the application of data often leads to new questions. Specific tasks can be arranged along the different steps of the process cycle. Data literacy can then be seen as a set of competences or, in other words, a cluster of effective behaviours and attitudes towards the fulfilment of a certain task.

Competence frameworks often follow a thematic order; for example, the OECD Competence Framework groups various key competences into competence categories [7]. The European eCompetence Framework (e-CF) [8] arranges competences in fields of competence, defining levels in each case in accordance with the European Qualifications Framework for Lifelong Learning (EQR) [9] and gives examples of knowledge and abilities or skills. In addition to knowledge (theoretical and factual knowledge) and skills (cognitive and practical skills or abilities), the EQR distinguishes between actual competences (responsibility and autonomy). The German Qualifications Framework (DQR), on the other hand, distinguishes between technical competences (knowledge and skills) and personal competences (social competence and independence) [10].

This already reveals the problem of differentiation: responsibility and autonomy in the sense of motivational, volitional, and social skills are represented in the EQR and DQR as a competence category, but in the e-CF they are represented by the description of competence levels. In the field of competence “planning”, for example, performance level 5 is shown in that someone that “uses far-reaching leadership strength to achieve consensus and commitment of the company’s management to the ITK strategy”. Such a level description does not distinguish precisely between ability and willingness, but rather assumes both. The ESCO (European Skills and Competencies of Occupations) project of the European Commission, on the other hand, defines “attitudes” or “attitudes and values” as independent, general competencies [11].

While EQR and DQR propose “knowledge”, “skills” and “values” as different categories of a competence, the KSAVE model [12] proposes them as different dimensions. The latter perspective is also the one of this paper. It has become increasingly common in international digital competency frameworks, as for example in the DQ Framework [13], which aggregates more than 25 prior leading frameworks about digital literacy and skills. The DQ framework lists Data and AI Literacy as one of 24 digital competencies.

The competence framework proposed in this article is very detailed and complex in relation to the examples found in the literature, but the ambition was to create a framework comparable to the standardised European competence framework. The procedure for developing the competence framework was therefore based on the methodology used in comparable projects of the European Commission, such as e-CF or ESCO.

2.2Justification for another data literacy competence framework

The following sections summarize in short, several requirements for a data literacy framework that are not fully met by existing frameworks. In particular, this is true for the focus on interdisciplinarity and the competence dimension of attitudes and values.

Ridsdale defines data literacy as the ability to collect, manage, evaluate, and apply data in a critical manner. If data is to support decision-making processes, competent answers to four basic questions are needed:

3.1Definition of “Data Literacy”

Therefore, the HFD Data Literacy Framework defines the term “data literacy” very comprehensively:

“Effective execution” is one that is target-oriented with respect to the target of the whole process, which is creating value or knowledge from data, i.e., it answers the question of what needs to be done to achieve this goal (“doing the right things”). “Efficient behaviour and attitudes” allow to achieve this target with the least possible use of resources, answering the question of how the target can be achieved in the best possible way (“doing things right”). The explicit highlighting of attitudes is in accordance with Seidl et al. who emphasize the importance of attitudes, values, and ethics for key competences [15].

3.2Structure of the framework

The HFD Data Literacy Framework is structured in four outline levels. They reflect the different levels of the process and its six steps as follows:

Outline level 1 determines 6 fields of competence, derived from the process steps:

The competence fields (A) to (C) correspond to the production process (“coding” from the real world to data and then to data products, the competence fields (D) to (F) correspond to the reception process (“decoding”) from data products to data and then to the real world.

Outline level 2 describes essential competencies for each field, each with a generic description. A further subdivision is made in competence field (B).

Outline levels 1 and 2 of the data literacy framework are summarised in the following Table 2.

Outline level 3 gives examples of knowledge, skills/abilities, and motivation and (value) attitudes. The dimension “knowledge” refers to the knowledge needed to master the respective process step. This dimension deals with the (complex technical) knowledge. The dimension “skills” describes the corresponding abilities. It refers to the application of knowledge, i.e., the abstraction of what has been learned. The dimension “Attitudes, Values, Ethics” (AVE) describes motivations and values that an individual should possess. It covers ethical requirements, e.g., to enable objectivity and to avoid misuse of data and analyses. In addition, it is about motivation, openness, and the willingness to learn from mistakes. The AVE dimension is often neglected in competence frameworks. Although values and ethics are sometimes formulated as separate competences or as competence bundles, the personal attitude of the individual is rarely the focus of attention.

For competence “C1: Analyse data” in the decoding subprocess of the value creation process, the framework gives, among others, the following examples:

Knowing that statistical artifacts do exist is a necessary precondition to anticipate how they could be misleading when results are used in data-driven actions. This ability in turn is a necessary precondition to decide which and how respective analyses should be performed, interpreted, and eventually provided with a warning notice.

For the corresponding competence “D3: Interpret analyses” in the coding subprocess, the framework gives, among others, the following examples:

Again, the dimensions are closely related. Being aware that previous knowledge about the context of a problem affects one’s perspective on the results to be expected or their meaning, respectively, is a prerequisite to critically question their interpretation.

Outline level 4, finally, specifies competence levels that represent a simplification of the EQR levels. They provide European references for the levels of complexity to which a competence applies. Here, the competences have been specified at three levels of complexity of the underlying requirements, which are to roughly outline a “basic level”, an “advanced level” and an “expert level”. They are defined via requirement levels that build on each other.

These specifications must be worked out in detail to enable the derivation of concrete learning objectives. The following section shows how such learning objectives were derived for a learning application in adult education and how the framework was adapted accordingly.

4.1Target groups and approach

The D21 Digital Index, co-funded by the German Federal Ministry for Economic Affairs and Energy, has been providing a comprehensive annual picture of the state of digital society in Germany since 2013. For the year 2021/22, the study shows that the group of users over 50 has increased, digital progress is viewed more positively than in previous years, and access to digital infrastructure has improved significantly. 80% of the population also have mobile access to the internet [16].

There is also a need for action regarding the confident use of digital technologies: A total of 15% of the German population consists of “digital keepers” (40%) and “digital laggards” (15%), while only 46% are “digital pioneers”. In 2016, 31% were classified “digital pioneers” and 26% “digital laggards”. This means that the “digital skills gap” is closing but still, most Germans are at risk of being left behind in the process of the digital revolution – or already are. People with a low level of education are particularly hard hit: across all age groups, there is a sharp drop in the use of digital technologies compared with people with a formally higher level of education (Internet use: 70% vs. 98%). This goes hand in hand with a low awareness of the problem and great carelessness regarding personal data. But the “digitally literate” also have some catching up to do in terms of digital competence: they often tend to underestimate the socio-economic impact of the digital revolution and are hardly or only hesitantly expanding their digital skills. This is especially true for data literacy.

Creating data literacy for all, raising awareness of data handling issues and establishing a confident approach to new technological developments such as AI, IoT or Big Data and one’s own data – these are some of the goals that the German Adult Education Association DVV (“Deutscher Volkshochschul-Verband”) is pursuing with the app “Stadt | Land | DatenFluss” SLDF (“City | Country | Data Flow”, which is a word game in relation to the famous game “Stadt Land Fluss”, known as “Categories”, “Tutti Frutti” or “Jeu du Baccalauréat” in other languages) [17].

SLDF therefore addresses both “digital keepers” and “digital outsiders”. It must be considered that these groups are characterized by a high degree of heterogeneity in terms of their digital preferences and interests, competencies, expectations, experiences, habits, and demands – a “one size fits all” offering is not appropriate under these conditions. That is why the app takes a modular approach.

The first target group can be described as citizens who are in general interested in the topic of digitization and datafication, but struggle with some basic terms and concepts. Their thoughts and needs could be described by questions such as:

A second target group can be characterized with the question:

For both target groups, it can be assumed that digitization has already changed (and is still changing) their learning habits. Attention spans are getting shorter, the desire for entertainment is growing. Numerous offers compete for the attention of learners. This requires new didactic concepts include playful learning with digital formats, especially if target groups are focused that have unlearned how to learn during their lives or have never learned how to learn.

The workload of the app takes these considerations into account. For the first version, 12 lessons of 3 learning units each, one basic module on fundamental knowledge and several outlooks were developed. With 15 minutes workload on average, each learning unit can easily be completed on the way to work or during a short break. This results in a total workload of 10 weeks when working on one learning unit per (work-)day, which is a sufficient period to establish a new habit of self-directed learning.

4.2Derivation of learning objectives

According to the tradition of empirical research, learning from data starts with one or more research questions (“science starts with a question”) and models a hypothesis-driven or confirmatory approach. To answer that question, data is acquired, prepared, evaluated, and interpreted according to the respective purpose of analysis. In public statistics, the process starts with a need for information brought forward by political decision-makers.

Digitisation, however, creates data “purposelessly” in large quantities and heterogeneity and even raises new questions (“data science starts with the data”). Data-driven or explorative procedures are becoming increasingly widespread in practice, so that new skills are needed in dealing with new forms of data – text, sound, image – beyond the familiar scale levels and forms of storage. An explorative approach of “letting the data speak for itself” is becoming more popular. However, the risk of drawing wrong conclusions is high if there is a lack of a basic understanding of statistical errors such as the confusion between correlation and causality, or if there are unrecognised shortcomings in the quality of the data (e.g., a bias). Additionally, ethical literacy is required when data is freely combined and analysed for purposes other than its original purpose of collection. Regarding public statistics, those developments need to be reflected in regulations that define outputs (= information needs) instead of inputs (= data).

University curricula for undergraduates and graduates traditionally still focus on the “coding side” of the value creation process, for example in statistics courses, as researchers need to learn how to formulate hypotheses, how to collect and manage data, e.g., by conducting surveys, and how to analyse that data statistically and to present it for the scientific community, even though it is likely that this will change in the near future, due to the possibilities of such a “datafication”. The HDF Data Literacy Framework aims to integrate both perspectives.

With respect to the target groups described above, it seems obvious that the HFD Data Literacy needed to be adapted for the purpose of adult education. Its objective is neither the acquisition of expert knowledge nor the ability to apply statistical methods or work with data management and data analysis tools. Those are not the tasks at hand to be performed in everyday life by the average citizen or employee. To them, data literacy should rather provide orientation in a world where “we are drowning in information and thirsting for knowledge”, as futurologist John Naisbitt puts it [18], and be part of a general education for everyone. “If we want responsible citizens in a modern technological society, we have to teach them three things: Reading, writing and statistical thinking, that is, the rational management of risk and uncertainty”, which is a famous (but false) quote by science fiction writer Herbert G. Wells [19].

For this reason, the six original competence fields were renamed and clustered into three competence areas. For each of the competence fields, a leading question was formulated to derive the corresponding learning goals, which was done by answering the question as follows: What does one need to know, to be able to, and to be willing to, so that they can solve the problem(s) described by the respective leading question?

The most central competence area, “Act on data”, groups the fields “(A) Establish data culture” and “(F) Derive actions”. Reflecting the needs of the target groups, they were renamed into “Data culture” and “Data sovereignty”. This area covers the chances and risks arising from novel applications of data, but also the relationship between human and artificial intelligence. A second area, “Use and protect data”, combines the fields “(B) Provide data” and “(C) Evaluate data”. They were renamed into “Consciously share data”, which puts emphasis on data protection issues, and “Gain data and information”, which gives an overview how data is transformed into information by using various analysis techniques. The third area, “Classify data and information”, aggregates the fields “(D) Interpret results” and “(E) Interpret data”, to raise awareness for the need of contextualizing information, so that knowledge and action can be derived.

The following Table 3 summarizes competence areas, leading questions, and respective learning objectives on competence levels L1 as a basic level for the first target group, and L2, an advanced level for the second target group. For each of the dimensions: knowledge (“understand”), skills (“apply”), and attitude (“evaluate”), one learning objective was formulated. Dimensions were renamed as well, so that they can be mentioned in the app without using scientific terms.

4.3Structure and content of the learning app

As a starting point, the app provides a virtual “adult education centre” that contains a glossary and six lessons covering the three competence areas with two leading questions each, i.e., the process steps of learning from or decision making with data. Four additional lessons deal with key drivers behind digitization and datafication: big data, artificial intelligence, internet of things, data flows and digital communication. The purpose of that virtual centre is to understand what the key technologies behind digitization are. In upcoming versions of the app, other technologies like robotics, virtual/augmented reality and others can be included. In terms of competence dimensions, it is the place where “knowledge” is stored and can be explored, comparable with a library or a museum. The learnings paths can be followed according to individual preferences, lessons may be left out or completed only in parts.

To apply that knowledge in concrete problems and to evaluate potential solutions, real-world situations seem a more appropriate environment: “skills” and “attitudes” should be practiced where the change through digitization takes place. The app presents three fields of application: health, mobility & smart city, work & economy. In the future, more fields can be added, for example leisure, education, energy, or sustainability. Within each field of application, users can work (or play) their way through four learning stories, two on competence level L1 (basic) and two on competence level L2 (advanced). Table 4 gives an overview of the stories.

Figure 1.

Typical learning path in Stadt | Land | Datenfluss.

One story, “Monitoring of body data”, describes a conversation of Timm and Martin on a bike trip, where one of them frequently monitors his heart rate, calory burning rate, and other measures with the help of his smart watch and an app. Several simple statistical concepts like the average are explained, as well as some basic chart types. Also, the difference between a measurement (the reflection of a light signal in a person’s blood) and a key figure (percent of oxygenic saturation, calculated with an algorithm) is discussed to clarify the relationship of data and information. It is mentioned that both during the measurement and the calculation, errors may happen, so that results may show random variation.

Or there is Mara, a young employee who works in her home office and reflects on how video conferencing has changed the way she interacts with her colleagues. Privacy and security issues of dataflows caused by the digitization of communication are in the centre of the story “The flexible workplace”. Later that day, Mara will apply online for a job and experience AI-assisted “Intelligent recruiting” – and find out about chances and risks of algorithms in digital recruiting. Then the user meets Lena, the 16-years-old trainee in car mechanics, who is curious about the large amount of sensor data that is collected in a car and can be analysed to identify potential problems. In the story “Digital learning in the work process”, she talks about digital learning approaches in dual education. Visiting the “Smart factory” where her uncle works, she gets deeper insights into the use of data and algorithms to monitor, control, and optimize production processes with so-called statistical process control.

Whenever a specific term such as “AI”, “internet of things” or “wearable” appears in one of the stories, a link to a glossary is provided. Additional elements used to build the bridge to real-world applications are short videos (excursions), where experts from different research fields describe their work with respect to data literacy. One of the videos for instance deals with robotics, another one with urban planning. Figure 1 schematically shows a typical learning path.

4.4App and MOOC as complementary offers

SLDF can be accessed via two different channels. On the one hand, it is implemented as an app for iPhone and Android phones. Both app versions are available for free and can be downloaded from the respective app stores. They do not require a log in, and the learning status is stored on the user’s device.

A second way of accessing is the browser version, hosted by KI-Campus, the German learning platform for artificial intelligence, which is founded by the German ministry of education and research (Bundesministerium für Bildung und Forschung, BMBF). KI-Campus offers courses in German and English, most of them on university level. SLDF can be reached on the KI-Campus website and is fully integrated into its learning management system. A login is optional but required if users want to save their learning status, because it is stored centrally on the platform.

Regarding the expected learning situation, the app allows an approach that can be described as “affective”, that means, the primary desire is one for entertainment and users want to play. Short learning units that can be completed in approximately 15 minutes allow that content is learned selectively. Learning units within a lesson or learning story of a field of application must be completed one after another on a predefined learning path. The approach of the browser version addresses a more cognitive learning situation where users primarily want to learn in focused, probably longer sessions that require more concentration. Content is linked to context, which means that no strict learning paths are given. Users can decide by themselves where they want to start, stop, or continue.

All that is reflected in the functionality of the two different versions. “Everyday life” is the principle behind the app’s design and approach. Short texts are optimized for smartphone displays, animations and short quizzes are included regularly, and the graphical design reminds of a mobile game. Also, one can earn bonus points that can be exchanged for short and entertaining games on the virtual city’s “playground” which is shown in the northwest of the map (see Fig. 2). The browser version, in contrary, is designed as a classical Massive Open Online Course (MOOC). Reduced graphical design prevents distraction, there are links to additional content on the internet, or external content is embedded, e.g., articles from Wikipedia. Advances courses that are available on the platform can be taken to broaden or deepen the topics where the course itself can only give a first insight. Furthermore, the MOOC can be extended with other didactic scenarios, such as tutoring or peer-learning. The two offerings, app and MOOC, are intended to complement each other.

Figure 2.

Map of Stadt | Land | Datenfluss.