How artificial intelligence will prevent Iceland from losing its language

Iceland He has been fighting for several years so that his mother tongue does not disappear, due to the numerical growth and the low number of people who speak it. A solution to this problem would be in the implementation of artificial intelligence GPT-4recently announced by OpenAI.

Using this technology would allow them to train systems to learn their language and thus develop applications and digital environments for Icelanders to preserve their traditions in the face of the invasion of English.

The country’s president, Guðni Th. Jóhannesson, has already started discussions with the company to use AI as a solution. “We need to integrate our language into the software and applications that people use every day,” said Jóhanna Vigdís Guðmundsdóttir, executive director of Almannarómur, a nonprofit language technology center.

It may interest you: GPT-4 is exciting and scary

The fight against this possible disappearance of the language has led the government to create the Department of Language Planning, which has the task of inventing global terms in Icelandic. For example, “computer” is computer, which is a mix between slaughter (number) and skip (prophetess).

All this with the aim of maintaining the language in the new generations, taking into account the numerical growth since officially this language is only used by approximately 340,000 people.

Which leads to platforms like Siri and Alexa not having it among their options, and users opting for English as the global option.

In this context, a local technology company called Miðeind ehf formed GPT-3 adapt other systems, but the process did not go well, so with the new version of the artificial intelligence they expect to have better results because of their advances.

It may interest you: Here’s how Siri, Alexa and Google Assistant lost the battle of artificial intelligence

To achieve this, the company has assembled a team of 40 people who will be responsible for training the system on two key topics: grammar and cultural knowledge of Icelandic.

They will do this through a process called reinforcement learning from human feedback, or RLHF. What do humans give to GPT-4 a series of commands to generate four possible answers, which are then edited and the best chosen to create a list.

With these data, training is carried out and in this way, better answers are obtained.

The process has to be done this way, because for its operation the artificial intelligence takes information from the internet, where most of the content is in English and it does not have adequate knowledge of Icelandic, so the translations often contain many errors.

The team expects a long training work to refine the interaction with the AI ​​and that it generates creative and concrete content with this language. But if they do, they will be able to use this information to develop applications, such as making the voice assistant Embla, which belongs to the Icelandic company, able to interact fluently using this language.

It may interest you: LinkedIn integrates ChatGPT to create profiles

After strong growth of ChatGPT end of 2022 and the appearance of this artificial intelligence on multiple platforms, OpenAI, The company that owns the project has announced a new version that promises performance improvements.

The most remarkable point of its improvement is that the AI ​​will allow a multimodal language, that is to say that the requests will no longer be made only through text, but can be associated with images, videos and audios. Something that in practice would serve to summarize a YouTube video, for example.

Another important aspect will be the fidelity of information, since many of these chatbots have started to invent facts and data. But from the company, they assure that this new version “is 82% less likely to respond to requests for illegal content and 40% more likely to produce factual responses”.