The Future of Policing Transformed by AI and Mixed Reality: Exploring the Potential of New Technologies at Umeå University in Sweden

AI Information in Police Officers' View—The Technology Supporting "Future Investigations" and Decisions That Should Not Be Entrusted

Looking at the vehicle in front, digital information emerges around it. Without lowering their gaze to a screen, necessary information can be confirmed—.

Such futuristic police activities are being researched by a team from Umeå University in Sweden and others. They are using an interactive system that combines AI and "Mixed Reality (MR)."

In the research introduced by Phys.org and the university on October 8, 2026, the potential to support decision-making on-site and the ethical issues accompanying its introduction were examined through training involving police officers and instructors.

It is noteworthy that this is not research to replace police officers with AI. The researchers emphasize the question of how to utilize digital information while maintaining human judgment capabilities.

What is "Mixed Reality" that Overlays Information on Real Scenes?

MR is a technology that overlays digital information and objects onto real spaces. Unlike VR, which covers the view with a virtual space, it allows additional information to be referenced while seeing surrounding people, cars, buildings, etc.

In this study, Microsoft's headset "HoloLens" was used. It can be operated by gaze and voice, and its use is being considered to assist in identifying objects and understanding potential dangers.

In police activities, the situation in front of you changes moment by moment. Attention must be paid to multiple pieces of information, such as the movements of the other party, the positions of surrounding people, and coordination with colleagues.

If necessary information can be confirmed within the field of view during such times, it may help in understanding the situation.

However, more information is not always better. If too much is displayed or incorrect information is emphasized, it can lead to missing changes right in front of you. What's important is not just what can be displayed, but what, when, and to what extent it should be shown.

39 Police Officers and Instructors Participated, Training Received Favorable Evaluation

The research was conducted in collaboration with police education institutions and law enforcement agencies in Sweden and Catalonia, Spain, in multiple stages. A total of 39 police officers and instructors participated.

Before practical tests, participants attended two workshops. One was to learn the basic principles of trustworthy AI. The other was to actually experience MR and HoloLens.

After that, HoloLens was used as a support tool in training that assumed high-risk situations during vehicle stops.

According to the university's announcement, participants were very favorable towards the system's usefulness, ease of use, and reliability. Anders Sjöstedt, a police officer and lecturer at the university, explained that participants were highly interested and saw many possibilities.

There is also significance in the cooperation between the technology development side and the police practice side. The functions developers find convenient and those truly needed on-site do not necessarily match. Bringing together knowledge from both sides becomes the starting point for considering practicality.

"It Was Well Received" and "Safety Was Proven" Are Different

On the other hand, it is premature to take the results as proof that "using AI makes police activities safer."

The main achievement indicated in the introduction article is the favorable evaluation by participants who experienced the training. It does not show effects such as how many street accidents were reduced or how much misidentification was prevented.

Feeling that the system is trustworthy and the system working accurately in various situations are separate issues.

For example, does the support function in dark places, rainy weather, crowded areas, or environments with unstable communication? Can police officers discern incorrect information when it appears? These points need to be verified when considering actual operation.

These are issues to be addressed in future verification and cannot be judged as resolved from the introduction article.

Because it is a promising study, it is important to read separately the user feedback, technical performance, and effects on society as a whole.

The More Automation Advances, the Heavier Ethical Issues Become

The research team analyzed functions by dividing them into three risk levels from low to high. They found that the higher the degree of automation, the more ethical risks related to privacy, transparency, and algorithmic bias increase.

There is a difference in impact between showing reference information to police officers and AI evaluating people or situations and strongly directing subsequent responses.

If the system mistakenly evaluates a person in front as dangerous, that display could create a bias in the police officer. Conversely, missing a danger could lead to weakened vigilance.

This is not an event reported to have occurred in this training. It is an issue to be considered when introducing decision support systems.

Also, if the design records images and information in view, how to handle information about unrelated passersby must be addressed. Consideration must include storage duration, who can view it, provision to external parties, and deletion procedures.

The performance of the headset itself cannot answer these challenges.

To Prevent "Humans Make the Final Decision" from Being Merely Formal

Researchers emphasize that the system should not be fully autonomous and that humans should make the final decisions.

However, it is not enough just for humans to operate it in the end. If they are merely accepting AI displays as they are, human judgment becomes merely formal.

When considering similar systems in Japan, it will be necessary to design them so that police officers can scrutinize AI information.

For example, being able to distinguish whether the displayed information is directly observed facts, past records, or AI estimates. Understanding the limitations when information is outdated or uncertain. And being able to reject AI suggestions when they contradict facts seen and heard on-site.

Furthermore, it is important to have a mechanism to verify responses later, confirming what information was shown and how humans judged it.

To make the phrase "humans take responsibility" substantial, it is necessary to leave enough information and discretion for humans to make judgments.

On Social Media, Focus Is on Introducing the Paper, General Public's Evaluation Is Not Yet Visible

On social media, public posts introducing this paper can be confirmed.

Co-author Bernat Vivolas Jordà reported the publication of the paper on LinkedIn, introducing it as a result of a European joint project involving the Catalonia Institute of Public Safety, Umeå University, and companies.

Erik Borglund also shared the paper on LinkedIn, describing it as research addressing AI challenges in police tactics by Umeå's police education department.

From the posts confirmed, there is a movement to connect research results to specialized discussions.

However, these are posts introducing the paper itself and do not collect public reactions to the news on October 8. From what can be confirmed, there is not enough material to judge public approval or disapproval or reactions on Japanese social media.

Therefore, expressions like "welcoming voices are flooding social media" or "fears of a surveillance society are spreading" cannot be used for this research. What is visible at this stage is that the sharing of research is progressing.

If Considered in Japan, Pre-Introduction Verification Is Key

This research is not news that the same system's introduction has been decided for Japanese police. Nonetheless, the issue of how to use AI for public safety is relevant to Japan as well.

What particularly needs consideration is not just the convenience for police officers but also the safety and sense of acceptance for the citizens being dealt with.

Even if information can be quickly confirmed, if unnecessary caution is caused by incorrect estimates or records are left without explanation, it does not necessarily lead to citizen trust.

If considered in Japan, it might first be verified in a training environment, checking the method of information display and response to misidentification. In addition to how the workload of police officers changes, it should also be evaluated whether it obstructs the view or allows concentration on dialogue with the counterpart.

Evaluation items should include not only processing speed but also accuracy of judgment, avoidance of unnecessary intervention, proper record management, and the ability to explain to citizens.

This does not mean that the effects for Japan have been confirmed in the research, but it is a point of discussion when considering the results in the context of Japanese society.

What Is Needed Now Is "Training to Use AI" and "Training to Doubt AI"

The research team believes there is significance in incorporating such technology into basic police education, allowing future police officers to learn both the possibilities and weaknesses early on.

It is not enough to just learn how to operate it. It is necessary to develop the ability to know under what conditions information becomes uncertain, what to do if the display and reality differ, and whether they can cope even if support is lost.

What this study shows is the potential of technology to increase information in police officers' view and the challenges to prevent that information from dominating judgment.

Can attention continue to be directed to the voice and movement of the person in front while looking at what AI shows? The quality of future police activities depends not only on the novelty of the equipment but also on how to protect the judgment of the humans using it.

Sources & Reference URLs

1. Phys.org:
   Published on October 8, 2026. Introduces the research overview, training with 39 participants, user evaluations, and ethical issues.
   https://phys.org/news/2026-10-ai-reality-police.html

2. Umeå University: Official Announcement by the Research Institution
   Confirms joint research in police studies and computer science, verification using HoloLens, the necessity of maintaining human judgment, and prospects for police education.umu.se
   https://www.umu.se/en/news/research-shows-how-ai-and-mixed-reality-can-enhance-police-work_12196568/

3. Research Paper: Trustworthy AI and Mixed Reality in Police Interventions: Challenges and Opportunities
   Published in the Journal of Artificial Intelligence Research. The description of the research content is based on the provided article and the university's official announcement.
   https://doi.org/10.1613/jair.1.19461

4. LinkedIn: Public Post by Co-author Bernat Vivolas Jordà
   A post introducing the publication of the paper and that it is a result of a joint project. Referenced as an example of research sharing on social media.Bernat Vivolas Jordà
   https://www.linkedin.com/posts/bernatvivolas_publicat-larticle-trustworthy-ai-and-activity-7471214915828158465-gYXJ

5. LinkedIn: Public Post by Erik Borglund
   A post introducing the paper as research addressing AI challenges in police tactics. Not indicative of general public opinion.Erik Borglund
   https://se.linkedin.com/posts/erik-borglund-5631874_alldeles-f%C3%A4rsk-artikel-fr%C3%A5n-bland-annat-polisutbildningen-activity-7471157629449465857-aUht