Skip to main content
Search
Menu

Easy-to-use tools necessary as AI usage increases

AI investments are on the rise. Computing power has never been greater. But, according to Martin Körling, Senior AI Researcher at RISE, researchers and specialists in trade and industry and the public sector need to have access to more easy-to-use machine learning (ML) systems and other technologies. 

A quick read through of this year's Data and AI Survey among leading American companies confirms what many suspect. Almost everyone is increasing their investments in data and AI technologies. Almost everyone believes they get value for money.

And perhaps more interesting: the number of data scientists is increasing and 26 percent of companies today have operational AI systems in their production, which is double the year before.

“What’s happening now at the big Swedish companies is that they use specialists, referred to as data scientists,” says Martin Körling, who is an expert in AI platforms and infrastructure. “They help those who actually know the business area or the specific product technology.”

Körling says that it can be difficult today for an operator to set up an AI platform to store data, train, and run models:

“Domain specialists will eventually be able to work with data and ML development themselves. At present, the tools are often too complex. It can also be difficult to access infrastructure, such as calculation and computing power. These things are often found in cloud infrastructure.”

Support for future tools

It is important to choose a platform carefully. And to take the rapid development of tools into consideration.

“One requirement you should place on the infrastructure is to be able to freely choose tools. After all, there is no single application that can satisfy the needs of all specialists.

“There may be a new tool within a few months, and the infrastructure must be able to support it and be compatible,” says Körling and mentions MLOps, functions or methods that involve automating retraining of the model:

“When the model gets old, when new data is obtained to update the image recognition or language model, you have to retrain it automatically and continuously.”

One requirement you should place on the infrastructure is to be able to freely choose tools

Hardware requirements create success for cloud services

An AI platform is not only software, but also hardware. For example, Körling sees the success of cloud services in ML as a direct result of the availability of powerful processing units.

“At present, many people are choosing between European and US cloud providers,” he says, referring to the fact that based on, among others, the General Data Protection Regulation and the Schrems II judgment* in the Court of Justice of the European Union, authorities and public sector organisations are struggling with how to manage and geographically store data.

“It’s possible to comply with the requirements of the legislation by using European suppliers. The public sector is investigating how to deal with this.”

One option is to buy your own equipment and operate proprietarily. The upside is total control.

“The downside is that establishing your own data structure requires security expertise and other technical skills,” says Körling. “And that costs a lot of money.

“For example, a company as large as Spotify does not have its own infrastructure. They conduct their data analysis and store the data in cloud services. They use Google’s cloud.”

 

*: Following a judgement by the Court of Justice of the European Union, the transfer of personal data to the US without a valid basis in the GDPR became illegal.