Artificial intelligence (AI) refers to a broad spectrum of technologies that combine to enable machines to imitate human intelligence. AI systems have been developed to learn, adapt and make decisions based on data.
Examples include Siri – Apple’s digital assistant – and Google Assistant, which can answer questions you ask.
Machine learning is a subcategory of artificial intelligence that involves computers learning from data. Instead of being programmed with specific rules on how to carry out a task, they learn by analysing the patterns in the data.
An example is the spam or junk folder in your email inbox. Email services learn to recognise spam by analysing many examples of genuine emails and spam emails. As a result, items that the email service thinks are spam never land in your inbox, but go straight to your spam or junk folder.
For Lerøy’s part, machine learning can help to recognise lice and predict prices and harvest volume, etc.
Machine learning operations are a practice that combines machine learning with development operations (DevOps) to develop, distribute and maintain machine-learning models efficiently. This involves automating and optimising the entire life cycle of machine-learning projects: from data preparation and model training to distribution and monitoring.
An example of a machine-learning model is when you finish watching a film and are immediately recommended another to watch. Most streaming services use machine-learning models to continuously update a recommendation algorithm based on users’ viewing history and preferences.
A practical example is Lerøy’s machine-learning model that predicts future salmon prices based on historic seasonal variations and price data. In an MLOps framework, this model goes through all phases of its life cycle: from development and testing to implementation, continuous monitoring and updating. This structured approach ensures that the model remains accurate and relevant over time, even in a dynamic and changing market.
Generative artificial intelligence differs from traditional machine learning in that it not only analyses and makes predictions based on data, but can also create new content, such as text, images, music or videos based on learned information. These systems can create something new that is similar to what they have learnt from.
A prominent example is the chatbot ChatGPT (link). This can conduct natural conversations, answer complex questions and help with tasks such as writing or programming.
“Once these data are available to us, we can use them to predict future events. They become a tool to assist with decision making,” says Satheshkumar, or Sathesh, as most people call him.
The job of getting these systems to a stage where the data are usable is part of Lerøy’s “Digital transformation” initiative, led by the IT department. One of the Group’s goals is to become more data-driven, and the initiative involves IT specialists working closely with management and other departments to achieve this, by means of improvements and adopting new technology.
“My job is about making data accessible. I collect and organise information so that others can easily find, understand and use it.”
Lerøy owns various digital systems as a result of acquisitions.
“Harnessing these data and being able to make decisions based on them is essential for Lerøy. My focus is on achieving even better control of the data so that, going forward, we can use them to make decisions,” he explains.
Specialising in artificial intelligence and machine learning wasn’t something he ever envisaged. At least not when he took a module in machine learning towards the end of his undergraduate degree.
“It was hard, and I never really got the hang of it,” Sathesh says.
But later, when he presented a master’s thesis on the subject, he became both motivated and committed to giving it another chance.
“I realised there were many more fields of application than I’d first thought. I thought wow, this is great, this is fantastic!”
With artificial intelligence and machine learning in the bag, he found himself working in the medical field. After a year as a researcher, he started a Ph.D.
“In layman’s terms, I wrote about and researched artificial intelligence in medical image analysis.”
Working with researchers in the radiology department at Haukeland University Hospital, he developed various machine-learning models for medical image analysis. The model was used on different patients with different medical conditions such as backache, and lung and gynaecological cancers. An otherwise time-consuming task that doctors would have carried out manually could now be handled by the model, providing answers in seconds. The results in his doctoral thesis showed improvements in accuracy and workflow compared with performing the tasks manually. His Ph.D. opened up opportunities, and he had the chance to go to the USA to work at the Mayo Clinic in a collaboration with Haukeland Hospital. But despite the prestigious job awaiting him, he couldn’t reconcile himself to the decision.
“It was a very medically specific direction to go in, compared with my developer background where I can focus on creating something from start to finish. I wanted to work on producing tools that can be used in real-world settings and not just in research environments,” he says.
A chance meeting on his way home from his job at the time opened his eyes to the seafood company. A former colleague talked about how Lerøy was working on digital transformation and the opportunities that were on offer in the company’s IT department. As luck would have it, they were on the lookout for someone with his background, and after a successful interview, a new path opened up.
Sathesh is also working on internal applications for generative AI within the company. (Faktaboks om Generativ AI). One of the projects he’s involved in is a recently launched HR assistant. This enables employees to ask HR-related questions and get a ready-generated answer based on what they ask.
“It makes the information easier to access and means that employees save time spent looking for information,” he says.
Artificial intelligence is a fast-developing field: if you don’t keep up, you miss out on important information and updates. This is why monitoring developments is an important part of his job. The fact that many are sceptical about developments in AI and what it means for their everyday working life is nothing new to Sathesh. He is keen for people to understand the opportunities the technology offers.
“It won’t replace us, but we should learn to use it as a tool, because it has a lot to offer,” he clarifies.
The project he has spent the most time on involves creating a machine learning operations environment to be able to predict salmon prices. Machine learning operations (MLOps) refer to the practices and tools used to develop and distribute machine-learning models in an efficient and scalable manner. The project is a multidisciplinary collaboration with employees in the analytics department. They are the end users of the program and also the people who know which data are relevant in setting salmon prices. Based on input from the analytics department, Sathesh has included data such as exchange rates, public holidays in various countries and many other factors that together play a part in determining salmon prices.
“These models are extremely good at identifying patterns that may be difficult for humans to discern, and so they can be used as a valuable decision-making tool.
It’s important to monitor the models over time to ensure the output remains of a high standard. If the standard drops, the models have to be retrained, which is a fundamental part of how machine learning works,” Sathesh explains.
