The emergence of Industry 4.0 in the post-Covid era has forced companies to adopt new procedures heavily relying on technologies. While automation and the adoption of new technologies can be helpful in many ways, it also jeopardizes certain jobs, rendering the training of large numbers of workers obsolete. AI4Labour is an H2020 European Project aiming to predict which of these jobs will be at risk of automation in the near future, in order to allow individuals, companies, institutions and policymakers to prepare themselves for the transition. The challenges that AI4Labour faces are therefore threefold. First, being able to predict the occupations or tasks that might be replaced by others in the near future. Second, enhancing the relation between task tasks and skills with education and training options, to enable workers to learn new skills or earn the required qualification for the job at hand. Finally, recommending training options to workers for the new skills required by the market. As a result, the innovative skill-based modelling and skill development methodology designed in this project will help reduce the possible negative AI-based impacts on the labour force.
introduces the terminology and the use case of the project. Section 3 presents the main tasks tackled in the project, grouped as data gathering (Section 3.1), skill modelling (Section 3.2) and recommendation portal (Section 3.3). Section 4 breaks down the participants in the project, both on the industry and research sides. Finally, Section 5 presents the eforts related to the dissemination of the results, while Section 6 summarizes and closes the paper.
Before going deeper into the diferent tasks of the project, it is necessary to clarify both the terminology used in the project and the use case.
In the context of the project, we can distinguish four core concepts, described below: recommend users a training option that allows them to obtain a new job. Below we clarify each of the steps in this scenario.
1. We have two employees, Alice (P1) and Bob (P2), who have certain skills that allow them to perform certain tasks in certain jobs (J1 and J2, respectively). 2. Nevertheless, while Alice’s job looks safe for now, automation will soon be able to perform the tasks required for Bob’s job (J2). 3. Therefore, we inform Bob that there is a high probability of he will soon be losing his job. 4. Luckily, there is one job for which Bob has almost all the skills required (J3), and he would only need to acquire a new one (S8). 5. We inform Bob of this situation and recommend him a course he could take in order to acquire this skill. 6. Bob takes the course. 7. Now Bob acquires the skill and is able to perform the tasks in J3. 8. Bob gets J3 and the impact of automation is minimized.
We summarize below the main tasks tackled in the project, which we can split into diferent groups (data gathering, skill modelling and recommendation portal). Figure 3 shows the workflow of the project.
The interaction among these basic concepts is depicted in Figure 1.
• Job: an occupation in the labour market. Each job
entails a series of tasks to be performed. 3.1. Data Gathering
• Task: a piece of work to be undertaken in a certain A fundamental part of the AI4Labour project includes
job, such as managing people, writing reports the design of surveys to help to detect undesired
situor documenting business processes. Each task ations or changes in the labour market. One of these
requires a set of skills to be carried out correctly. possible situations is gender bias: there may be gender
• Skill: ability that allows a person to undertake diferences in the predictions, such as some jobs being
certain tasks. Some examples of skills are the more often associated with a particular gender than with
ability to speak languages or to handle certain the other, being these either male- or female-associated
tools. They can be acquired via training activities. jobs primarily disappearing, and, if so, considering an
• Training: an activity that allows a person to ac- equitable approach to take this into account when
makquire certain skills. Training activities include ing recommendations [
In addition to the information retrieved from these surveys, we also rely on previous and ongoing eforts for data gathering, such as the available occupations and skills databases O*NET2 and ESCO3. S6
T4 S7
S6 S1 S1 T1 T1 S6 S6 S1 T1 S6 S1 T1 S6
P2 P2 S5 S5 J2 J2 T3 T3 P2 S5 J2 T3
S8 P2 S5 J2 T3
S7 S7 S6 S6
T4
T4 S7 S7
S7 S6
T4 S7
S7 S6
T4 S7
S6 S6 S6 S6
T1 T1
T3 T3
J2 J2 T4
T4 S5 S5
J3 T8 J3 T8
S8 S8 S8
S8 S1
S3 T2
S1 P1 S2 J1
S2 S3 S3
S1
S1 S1 S1 5 S1 7
S1 S1
T2 T2
S1 P1 P1 S2 S2 J1 J1
S2 S2 P1 S2 J1
S2 P1 S2
J1 S1
S3 T2
S1 S1
S3
T2 S1
S2
S1 S1 4 S1 6 S1 8
S1
S3 T2
S1 S1
S3 T2
S1 S1
S3 T2
P1 S2 J1
S2 P1 S2 J1
S2 P1 S2 J1
S2
S1 P1 S2
J1 S1
S3
T2 S1
S2
S1 T1 S6 S1 T1 S6 S1 T1 S6
P2 S5 J2 T3 P2 S5 J2 T3 P2 S5 J2 T3 S6
T4 S7
S7 S6
T4 S7
S7 S6
T4 S7 S1
S6 S6 S6
S8 P2 S5 J3
T8 S5
S8 T1
T4
T3 S5
S8 J3 T8 J3 T8
S8
S8 S5 S7 S6
The Occupational Information Network Database, Similarly, the European Commission’s multilingual known as O*NET, is the primary source of information on classification of Skills, Competences and Occupations, occupations in the United States, and it is developed with called ESCO, is a project led by the Directorate-General the support of the US Department of Labor/Employment for Employment, Social Afairs and Inclusion. It is freely and Training Administration. This comprehensive available to access and consult, and it can also be downdatabase contains standardized and occupation-specific loaded through the ESCO API, which provides access descriptors on nearly 1,000 occupations across the US for software agents to analyze the data. ESCO contains economy, and it is updated periodically. Each occupa- descriptions of 3000 occupations and more than 13,000 tion is defined through a set of worker-oriented descrip- skills related to these occupations, translated into 27 lantions, which include knowledge, skills, and abilities that guages, including all oficial EU languages, as well as a worker should have, as well as a range of activities and Icelandic, Norwegian, and Arabic. tasks they should perform.
WP2
WP1
TASK 1.1 CREATION of DATA DICTIONARY & ML
METHODOLOGY DEVELLEOAPRMNIENNGTMofOMDLE/LDEEP
VERIFICATION of THE
MODEL with MODEL DATA NO
ACCURATE ?
YES NO
TASK 2.3
these documents, which are usually "hidden" on university websites and in very diferent formats), we will focus for the time being on those of the centres participating in the project.
Once we have models able to relate the diferent pieces of information of interest for the project (see Figure 1), we are ready to use them for user recommendation.
We are currently using the previously mentioned data sources ESCO and O*NET for two diferent tasks. First, for training the natural language models that relate skills and tasks, since the input by the industrial partners would not be enough for training and testing. This process involves diferent steps, such as semantic similarity between the diferent jobs/skills naming among diferent countries, task/skill matrix building to assign weights to 3.3. Recommendation portal their relations, and skill clustering and naming. The models created in the previous step allow to:
Once all this information is acquired, it will be represented in an ESCO compliant format, building a Knowl- 1. Predict which task will be automatized, and thereedge Graph able to support diferent applications, such fore the jobs that are at risk. as QA systems or pattern detection. 2. Based on the input of the user, find the most sim
Second, as useful as the previously mentioned re- ilar skills/tasks/jobs to theirs in the database. sources are, they do not include information about the 3. Tell the user if their job is at risk, and if this training required to acquire the diferent skills necessary was the case, recommend available training to for the performance of the labour activities mentioned. increase their possibility to get a new job related Unfortunately, we have not found either in the litera- to their skills. ture or industry any training repository that includes the These functionalities will feed a portal that allows the skills or tasks they cover. That is why it has been neces- user to input their data (current job and skills) and get sary to extract that information directly from the sources; the risk prediction and training recommendation. in the context of the project, diferent code snippets have been created in order to extract course listings (with their respective skills) from diferent MOOC platforms. In the 4. Consortium future, we plan to work on extracting this information from university teaching guides; however, given the absence of a global repository (and the dificulty of finding We can divide the partners in the AI4Labour project into research institutions and industrial partners. This division responds to the purpose of this Marie Sklodowska
KHAS Kadir Has Universitesi4 is the leader of the AI4LABOUR project. Their research focuses on topics such as Gender and Women’s Studies, but also leads the technical developments in the project.
University of Wolverhampton This English university5 work focuses on addressing real-world problems, such as the health of ageing populations and sustainable development, in a variety of diferent ways.
ITCL Instituto Tecnológico de Castilla y León6 is a Spanish research institute whose research covers areas such as Energy Technologies, Artificial Intelligence/Electronics, or Simulation (Virtual and Augmented Reality).
One of the main objectives of the project is to make AI-related information accessible to layman people, companies and any agent in society. Due to this, a big efort of the project is devoted to the organization of dissemination activities, such as seminars and workshops. Besides the website of the project and the broadcast of a newsletUniversity of Limerick , in particular the Kemmy ter, also some seminars have been and will be organized, Business School at this university7 bring their exper- including topics such as "Machine/Deep Learning Aptise on strategic HRM, Strategy-As-Practice and Strate- plications at Industry", "Skill Gap in Industrial Manufacgic Change, as well as some key areas which under- turing between Today and Tomorrow" or "AI and Data pin organisational strategy such as the Digital Trans- Analytics", as well as workshops on automation, Industry formation of Work, People Analytics and MNC Leader- 4.0, education and AI, or gender. Updated information ship/Management. about these events can be found in the webpage of the project12.
Curie Research and Innovation Staf Exchange project in research and development, they develop cloud-based itself, which seek to promote exchanges between compa- platforms in the fields of voice, vision and health and are nies and research centres and only funds secondments supported by machine and deep learning techniques. between companies and research centres, never between organizations of the same category.
In conclusion, the progress of computing power and the emergence of Industry 4.0 is expected to have a profound impact on the socio-economic dynamics of society. The recent COVID-19 pandemic has highlighted the importance of technology in shaping the future of work. The AI4Labour project aims to help workers, companies, and policymakers navigate these changes by predicting the types of jobs and skills that will be in demand in the future and designing the required training to develop these skills.
The results of the project are publicly available both in its Cordis profile 13 and the webpage of the project14, and the whole consortium is open to collaboration in order to positively impact the labour market.
10https://icbe.ie/ 11https://www.datalobster.io/ 12http://www.ai4labour.com/index.php/events/ 13https://cordis.europa.eu/project/id/101007961/results 14http://www.AI4Labour.com/ UPM Universidad Politécnica de Madrid8, and more specifically the Ontology Engineering Group, works in diferent areas of information extraction and management, such as the representation of knowledge (e.g., Ontology Engineering, Linked Data) or Natural Language Processing.
Among the business partners, we find companies operating in diferent fields.
Arçelik Better known in Europe for being the parent company of Beko, this international Turkish company9 designs intelligent and networked products at the intersection of AI, software and hardware. As a result of their cooperation with global and competent partners 4https://www.khas.edu.tr/en 5https://www.wlv.ac.uk/ 6https://itcl.es/ 7https://www.ul.ie/ 8https://www.upm.es/ 9https://www.arcelikglobal.com/en/ ICBE The Irish Centre for Business Excellence10 is a centre based in Limerick (Ireland) that facilitates members to access solutions to organisational challenges. Their principal aim is to promote and develop business excellence through benchmarking, knowledge-sharing forums, and training and development.
DataLobster The France-based company DataLobster11 optimizes operations and maintenance, services and develops digital transformation strategies for globally renowned companies.
We want to thank all the colleagues involved in the AI4Labour project.
This work received financial support from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 101007961 (AI4Labour project).