In our globalized world, the future potential of Icelandic tech startups in the innovation sector is commonly evaluated primarily on opportunities for international expansion—that is, how likely the technology or product is to gain traction in foreign markets. This perspective often emphasizes innovation-driven exports and highlights the significant growth opportunities available in industries not constrained by natural limits, unlike tourism. This mindset is reflected in Iceland's innovation grant environment, where applications are frequently assessed partly on their growth potential abroad. This approach undeniably places software companies like Miðeind in a challenging position, since opportunities for expanding Icelandic language technology are naturally limited. However, it also gives us a unique position as a company developing advanced artificial intelligence exclusively for the Icelandic market and Icelandic society. So what, then, is the purpose of a business that intentionally sets fixed limits on its own growth in a capitalist society?

The Early Years and Participation in the Language Technology Program

At Miðeind, we often say we work at the intersection of research and application. When the company began operations in 2015, few core solutions existed for Icelandic language technology, so there was hardly any foundation for anything other than focusing on research and initial development if we intended to work in this field at all. Language technology was also fundamentally different then than it is today, as we recently discussed in an article about the evolution of language technology from rule-based systems to artificial intelligence.

Today, multilingual AI language models can be fine-tuned on Icelandic data, or even work in Icelandic without special training. In Miðeind's early days, however, most language technology solutions had to be built from scratch. This development was time-consuming and expensive, which is why a special four-year language technology program for Icelandic was launched in 2019, funded by the government. The result was high-quality language resources that have been used to train AI models in recent years.

Miðeind participated in the language technology program and flourished during this period. We started small, but by the program's end, our impact had become substantial. A vision emerged of a sustainable company that would serve the Icelandic public with outstanding language technology solutions.

As of this writing, Miðeind's team consists of 16 people, and the vision has partially come to fruition with Málstaður, a language processing platform offering the best available language technology for Icelandic. In addition to developing and operating Málstaður, Miðeind has created specialized solutions for various companies and institutions and has built a reputation and earned trust for its deep expertise in artificial intelligence.

Operational Structure and Funding of Miðeind

Miðeind was founded and funded by Vilhjálmur Þorsteinsson with the purpose of benefiting society, supporting our language—and having fun in the process. Beyond revenue from subscriptions and development projects, as well as funds from the language technology program, we have relied on various grants—for example, from the Technology Development Fund (Tækniþróunarsjóður) and the Strategic Research and Development Program for Language Technology (Markáætlun í tungu og tækni)—to ensure Miðeind's continued operation. However, the vision of sustainability has not yet been realized, so we have also relied on equity provided by the owner at the company's founding.

In any business operation, it's necessary to regularly ask pressing questions about goals and purpose, and the answers can affect potential revenue streams, market positioning, and operational structure, since the structure should primarily serve the objective. Those of us at the helm of Miðeind have been considering for some time where Miðeind fits within the landscape of Icelandic companies and institutions. We haven't reached a final conclusion, but we have formed a clear picture of what Miðeind is (and isn't!) and what goals we want to achieve. We have always strived to be open and transparent in our work, and in that spirit, we intend this blog post to give readers insight into the company's status and purpose.

What Tasks Need to Be Done?

In recent years, artificial intelligence has developed rapidly, especially in generative models. Certain fundamental discoveries, such as Transformer neural networks (Google Brain 2017), have opened up new possibilities and applications.

Artificial intelligence, with connections to language technology, multimedia, and functionality in three-dimensional environments (production lines, self-driving cars, home robots), will revolutionize all kinds of experiences, interfaces, services, and decision-making processes—in the daily lives of citizens, in business, in administration, and in strategic management. Given that the world stands on the brink of an AI revolution, what needs to be done about it here in Iceland, and what would be most beneficial?

First, this situation should present good general business opportunities on traditional terms, as examples from previous revolutions show—the advent of personal computers, the World Wide Web, and smart devices.

Second, there's a significant societal question about how the Icelandic language and Icelandic culture and cultural heritage will fare amid these expected changes. Icelandic is such a small language that it's usually not high on the priority list of major technology companies or global research entities when it comes to market-based support. And even if social responsibility considerations were factored in, there would likely be many languages with higher priority than Icelandic, based on speaker numbers.

As readers have undoubtedly noticed, considerable progress has been made in recent months regarding Icelandic inclusion in the world's largest language models. Companies like Microsoft have added Icelandic support to some of their AI products. This progress hasn't happened in a vacuum but results from extensive work and advocacy efforts carried out by organizations including Miðeind on a voluntary basis.

Despite relatively good progress so far, Icelandic still lags far behind larger languages in terms of the range and quality of language technology solutions. Without action, it's likely that AI functionality related to language—written or spoken—will become much more readily available in English and other major languages than in Icelandic, if available in Icelandic at all.

Beyond the language itself, Icelandic cultural context is at stake. If large language and multimedia models lack sufficient proportional access to Icelandic cultural heritage—history and various characteristics, for example—that culture won't be reflected in the models' responses and the text and visual content they generate. Since such outputs will likely form a significant portion of the content we encounter in our lifetimes, there's a risk that Icelandic cultural characteristics will diminish proportionally. One can debate how dire this prospect seems, but it's certainly not inevitable.

Who Should Do the Work?

The aforementioned challenges and tasks in Icelandic language technology fall into several categories. Some projects are inherently basic research or fundamental infrastructure creation, rather than product development. Their natural home should be within universities and research institutions, structured to allow technology transfer to companies when appropriate.

At the other end of the spectrum are projects that can sustain themselves through companies in a competitive market. For example, there's likely sufficient demand for Icelandic chatbots to finance the development and operation of such solutions, making them self-sustaining. In some cases, an initial boost might be needed to reduce risk and overcome profitability hurdles, but this gap can be bridged with grants from organizations like Rannís's Technology Development Fund and EU development programs.

Then there's a category of projects that should inherently be profitable as products in a competitive market, but where the Icelandic market's small size is the obstacle. Examples include solutions similar to Grammarly (grammar checking and proofreading) for English, which is self-sustaining in the international market. Meanwhile, Miðeind's experience operating Málfríður and Málstaður has shown that despite these solutions being popular and praised by users for quality and usefulness, it's uncertain whether subscription revenues alone can cover development and operating costs unless a substantial portion of the population pays for subscriptions.

Another category consists of projects that most agree are necessary, but where it's difficult to establish pure market conditions for development and sustainable operation. Examples include solutions that automatically translate television content, such as children's programs and movies. This may require pooling private sector efforts and enlisting funds and/or public entities that support projects with societal goals in mind.

Finally, we can mention the general building of knowledge in artificial intelligence and the distribution of that knowledge to business and public administration. Other Nordic countries operate special institutions or units with this role, sometimes within universities and sometimes through strong university-industry collaboration. Examples include AI Sweden, NorwAI, Alexandra Instituttet og Finnish Center for AI.

What, Then, Is the Purpose and Role of Miðeind?

To answer this question, we have prepared a mission statement that serves as our guiding light in all decisions about Miðeind's operation and focus. The mission statement can be read here.