When an AI service disappears, business operations may slow

When an AI service stops working, staff may take over some automated tasks, but they may not be able to handle as many cases. That can lengthen processing times, and some functions may stop altogether, Telia Cygate CEO Fredrik Sidmar told Techtidningen.

In a recent IBM survey on companies’ control over technology and operations, AI problems were the most common category of issue. Forty-four percent of respondents said they had experienced AI-related problems in the previous three years, including incorrect results and problems with models or suppliers.

IBM Sweden technology chief Håkan Andersson said digital sovereignty is ultimately about having the freedom to act if a model changes, a supplier disappears or new requirements arise. Telia Cygate uses open models, Kubernetes and standard formats for data, code and logs so customers can move models, data and workloads to another environment. That environment must have enough capacity to run the models. Sidmar said GPU hardware and some infrastructure layers remain difficult to replace quickly, while platforms and applications are more portable. Proprietary software is another difficult dependency.

Only 33 percent of respondents said they could fully demonstrate their ability to switch suppliers or restore operations; the measure covers AI and other parts of business. IBM’s index is based on what companies say they can demonstrate, including responsibility, what happened and how operations could be restored. Twenty-six percent reached the highest level. The survey included 2,003 decision-makers at large companies in seven countries; Sweden was not included.

Andersson described a hypothetical case in the report: a bank’s AI credit-assessment service starts producing questionable results after a cloud provider makes a change. Reconstructing later what changed, which controls were in place and who could intervene could take time away from other development work, he said.