Responsible AI does not happen by accident

Alma Media’s services and marketplaces are part of the daily lives of millions of people and numerous companies in Finland and international markets, and an increasing number of them feature artificial intelligence capabilities or are being developed using AI tools. As AI does more and more work behind data, decisions, and recommendations, its responsible use is no longer a technical detail but a part of the company’s core operations. Matti Venho, who is responsible for Alma’s data protection and has been developing the AI governance model, explains how this work has been done.

“The work was started long before AI became commonplace. As early as 2019, the European Commission’s expert group drafted general guidelines for the ethical use of AI. Based on these, we at Alma also began to build a model suitable for us”, says Venho.

Since the material prepared by the European Commission’s expert group contains general principles applicable to everyone, the project also utilized the practical AI Governance and Auditing (AIGA) model developed at the University of Turku, which, according to Venho, is also excellently suited to Alma’s business and provides an outstanding framework for the ethical use of AI.

Based on this work, Alma Media’s ethical principles for AI were created, forming the foundation of what responsible AI means at Alma Media. The principles are:

1. Safe, secure, and private: the principle covers the technical robustness, reliability and resilience against attacks as well as privacy and sound data-governance throughout the lifecycle.
2. Transparent and accountable: the principle covers the traceability, explainability and open communication about system capabilities, as well as clear assignment of responsibility, auditability and redress mechanisms.
3. Human-centric and fair: the principle covers the respect for human autonomy, meaningful oversight and the right to contest automated decisions, as well as promotion of diversity and the avoidance of unfair bias.
4. Sustainable and socially beneficial: the principle covers environmental and societal well-being, including climate impact, social cohesion and democratic integrity.

What AI governance means at Alma

At Alma, AI governance means common rules, processes, and designated responsibilities to ensure compliance with both regulatory requirements and the company’s internal guidelines. By doing so, AI is used safely and in line with business objectives. This covers risk management, documentation, approval of tools, division of responsibilities, and continuous monitoring of activities.

The goal is competitive and responsible AI. Competitiveness means that the use of AI is economically justified and makes business sense. Responsibility, on the other hand, means that AI capabilities are used in compliance with all applicable rules – legislation and ethical principles. Legislation sets only the minimum requirements for responsible AI, so merely complying with the law is not enough: the same development has already been seen in environmental responsibility, where stakeholder expectations have over the years become institutionalized as part of corporate sustainability reporting.

Safety and reliability as a starting point

The first principle concerns technical stability and resilience, as well as taking privacy and data management into account throughout the entire lifecycle – from data collection to model training, deployment, monitoring, and ultimately decommissioning.

“We have the EU AI Act, under which we must implement certain measures in our business,” Venho says. In practice, this means that the systems used are secure, data protection regulations are complied with, and the tools utilized are reliable. The division of responsibility between the service provider and Alma is ensured through contracts and technical restrictions.

Responsibility cannot be outsourced

The second principle concerns traceability, explainability, and open communication about the capabilities of the systems, as well as who is ultimately responsible – even when the work is done by a machine. When AI makes decisions, Alma is able to review them, explain them, and, if necessary, also handle potential requests for correction.

“Responsibility cannot be transferred to the AI service provider, letting go of the steering wheel,” Venho states. Alma itself must understand how the model used works so that its operation can also be explained. The responsibility for the use of AI and its outputs always lies with Alma itself, not the service provider.

A human reviews the outputs before they proceed

The third principle concerns human responsibility and control in the use of AI. This includes the obligation to review decisions made by AI, through which we aim to promote diversity and avoid unfair bias. The general rule is that AI assists, but humans decide.

“We review the outputs of the AI and ultimately, so to speak, under sign on the output that we move forward with and put into use”, Venho says. Thus, the output always passes through human verification in one way or another before it becomes part of a service or decision-making.

Energy and the transformation of work spark discussion

The fourth principle concerns societal and environmental well-being: climate impacts, social cohesion, and democratic functioning.

“AI consumes a lot of energy, that is an undeniable fact”, Venho states and continues. “At Alma, this means careful consideration of what kind of AI services are offered, how much they are used, and what kind of energy is consumed by their use, and that environmental impact is always weighed alongside performance requirements.”

The social side of the principle concerns, among other things, the discussion about the impact of AI on work. Venho does not deny that some tasks will decrease or change with AI – but he also highlights the flip side of the coin: “For example in Alma Media’s recruitment business, AI is utilized specifically to help people find better jobs more easily.”

Speed has changed the game

What new requirements has AI introduced, and which aspects of responsibility have already been considered in digital business previously? According to Venho, above all, the speed of information processing and the ability of new models to “think” and independently produce solutions have significantly changed the industry.

“Previously, you set a system running and it worked exactly as specified. Today’s models might produce an outcome that you didn’t even think of yourself”, he says. This is precisely why ethical principles for AI are practically important: outputs are reviewed and the desired outcome is ensured.

“The scale and speed of data processing, along with the volume of available information, have a significant impact on our operations right now. That is why we need good governance models and ethical principles to remain in control”, Venho says.

AI systems and their capabilities are currently developing rapidly. This also requires continuous monitoring and the updating of internal processes from Alma. “This work continues and evolves; we are by no means at the finish line yet”, Venho reminds.

Share article