How does a future AI engineer see the future?

With machine learning, a system can identify correlations and anomalies in complex patterns. Card fraud, for example, can be prevented this way. So says Marcus Collard, 23, who is in his fifth and final year of a Master of Science programme at KTH in Stockholm, a master's programme in machine learning and artificial intelligence, AI.

Machine learning means letting the computer learn to find likely answers to different questions based on large amounts of data. The calculations are sometimes simple and sometimes more complex. This is being applied to more and more areas.

— Right now there's a lot of development happening in systems for customer service, information retrieval and smart assistants built on large language models such as ChatGPT, says Marcus Collard.

There have been significant advances in logistics optimisation through so-called Smart Inventory Management.

— Machine learning lets us detect sales trends for goods or the consumption of components in manufacturing. This helps avoid both shortages and surpluses in inventory management.

— It can also be about preventing faults in a production line. Or teaching the system to recognise the sound of a properly functioning engine, so it can raise an alarm when something starts to sound off.

— 10-20 years from now, even physical processes such as manufacturing will be able to be replaced. In the future, industrial robots will be able to perform more tasks, instead of being specialised as they are today.

Research into language models is now underway to interpret what is happening in the physical world from sound and images.

Google's PaLM-E is a robot that can perform tasks it receives through natural speech. Using sensors, it interprets data from its surroundings.

— It doesn't work that well yet, but in the future it could be a great help, says Marcus Collard.

AI also means a changing threat landscape when it comes to IT security, both technically and socially.

— The biggest technical threat from machine learning is that you can't verify or explain the solution, whether the system is right or wrong, says Marcus Collard.

Problems arise when it's important to be able to justify the answer. For example, pricing on insurance premiums when selling insurance. Or, in terms of safety, being able to analyze afterwards what happened when an accident occurred involving self-driving cars.

It's important to be aware that a service like ChatGPT tries to answer even when data is missing or incorrect. This is usually referred to as the system hallucinating.

In general, there can be risks when AI can mimic a real person's handwriting, voice or appearance. This can be exploited with malicious intent.

— Companies can protect themselves by being aware that everything nowadays can be forged, says Marcus Collard.

The programme belongs to the computer science department at KTH, but the field sits somewhere between mathematics and computer science. Many who take the programme have a background in mathematics or physics. Programming skills come in handy, but you also need to master the underlying mathematics, particularly linear algebra and statistics. The programme emphasises sustainable development and ethical aspects of data security.

— It's very important that personal data remains anonymous and that you consider the ethical consequences of the systems, taking known shortcomings into account, says Marcus Collard.

After graduating, Marcus Collard can look forward to choosing between exciting assignments in several different industries. Preferably in an area with good conditions in terms of data. The volume, format, reliability and relevance of data are decisive.

— As a machine learning developer, assignments are much easier if there's an abundance of data and well-defined goals for the system, says Marcus Collard.

But there are other aspects too.

— Of course, a good atmosphere and a matching culture are the most important things when comparing employers, says Marcus Collard.

Seclight is a consultancy focused on digital security.

Contact: Ylva Hansson, phone 08-94 55 99 ylva.hansson@seclight.com