Field: Strukturomvandling
Implementing and using AI
The aim of this report is to highlight examples of how established companies are implementing and using AI in their day-to-day operations and identify common characteristics that can serve as a basis for further research and policy development in this area.
There is currently no established definition of AI. The common denominator of different types of AI is computer technology that can perform tasks that require cognitive work of some kind. Today's AI development is based on an empirical approach, which involves training a computer to analyse large amounts of data in order to perform a task. The task may be detecting errors, translating speech into text, or playing games (Domingos 2015, Polson and Scott 2018, Gerrish 2018, McCorduck 2004).
Often, the work of large technology companies or small technology-driven start-ups is used as an example to describe AI’s business potential, and the implementation of AI in established companies is described as an inevitable transition. None of the descriptions, however, leave room for those companies that implement AI gradually or only in parts of the business. Moreover, most companies in Sweden have not implemented AI in their organisation at all.
Nevertheless, the aim of this report is to highlight examples of how established companies are implementing and using AI in their day-to-day operations and identify common characteristics that can serve as a basis for further research and policy development in this area. Six major Swedish companies participated in the study: ABB, Cevt, Cambio, Roche, Schibsted and Trelleborg AB. The study highlights how companies are working to implement AI in their operations and examines their perspectives on the prerequisites for, opportunities with, and barriers to AI. The company interviews were supplemented by two expert interviews.
AI - a technology among others
Our study shows that companies are treating AI technology as a tool amongst others and as part of data-driven analytics (advanced analytics). AI is primarily used as a complement, rather than a substitute. Companies do not have overarching strategies for how AI will transform their entire business but instead apply the technology incrementally wherever it can demonstrate business value. How it is used depends on the conditions and needs of different parts of the business organisation.
The infrastructure that surrounds data and how it is structured is many times more important to companies than the AI model or any other technology used for analysis. Collecting, structuring and quality assuring data, as well as agreeing on responsibility and ownership when data is shared between different parties, is time consuming and difficult. To coordinate internally all data-driven analytics work, companies are moving towards centralising data infrastructure and data management.
The clearest indication of the impact of digital technologies, and ultimately AI, on companies' business models is the development of new services based on software and data-driven analytics. There is a tension in companies' AI implementation between, on the one hand, incremental improvement projects that generate small gains quickly, and on the other, innovation projects that generate larger gains in the longer term. The latter shows a greater potential of AI technology; however, it requires more resources from the beginning.
Regulation, skills shortages, and data access hamper companies' AI use
The lack of clear regulatory frameworks, future regulations, skills shortages, and access
to data are perceived by companies as the main barriers to further implementation of AI
in their business on a larger scale. We specifically asked about companies' skillrequirements,
and companies are looking particularly for skills that bridge and link the
technology to different application areas within their own business. There is also a need
for general AI skills to enable more people in the company to contribute to the
implementation of AI and benefit from the technology across the different business areas.
Several of the companies are running their own extensive training initiatives to meet
some of their own needs for skills.
Policy implications
Historically, it has been very difficult to define and delimit what AI is. It would be easier
for companies to relate to data and data-driven analytics frameworks regardless of the
tools used by different parties in a collaboration. Policy initiatives could therefore be
usefully directed towards software-based and data-driven solutions. This would allow
for competition between different technological solutions but would contribute to a
policy that is persistent over time, regardless of technological developments.
Another policy implication is based on the assumption that increased diversity in skill
needs challenges the education system. Businesses show a heterogeneous need for skills
related to AI. This is not only a matter of variation in the need for specialist technical
skills, but also variation in the combination of technical skills and knowledge in relevant
application areas. With increased heterogeneity, it may become more difficult to forecast
and train the right number of engineers or technicians to meet the full variation in
demand. This places demands on formal education and training systems and raises the
question of the division of responsibilities between education, training systems and
enterprises.
The companies interviewed highlight the need for skills development;
several of them run their own training initiatives. This is a type of investment that larger
companies can make, but it is likely to be significantly more difficult for smaller
companies to invest in skills development. In the current context, this means that
investment in skills development and lifelong learning risks being unevenly distributed
across businesses and the labour market.
Implementing and using AI
Serial number: Rapport 2022:11
Reference number: 2022:152
Varför AI - förutsättningar, möjligheter och hinder för företag att använda AI Pdf, 1 MB.