Nordic Ethical AI Landscape

“When AI breaks, it breaks violently.” There are an infinite number of examples on the internet, of 
technology going wrong. The “AI risk” is a hazard not only to the company developing the technology, but to all stakeholders involved, not least to the end customer or citizen using the service. For investors, founders, and enterprises, minimizing AI risk is always profitable. Furthermore, it is the responsibility of the technology developer to make sure it does not amplify inequality in society. Technology should be strictly moving society in a positive direction and should work for all users equally. 

Absolutely. AI that is not properly governed, documented, controlled, and observed is volatile AI that can break at any time. The Nordics may have more of an incentive than other countries to monitor their AI, but developing ethical AI still requires monetary and temporal investment of some kind - which many companies from startups to corporations are not always willing to give. 

We define an “ethical AI company” as one that either offers services to help other companies make 
their AI systems more transparent, fair, trustworthy, and responsible or one that has developed 
technology that directly improves an area of human society that has previously been relatively 
unethical (e.g., toxic content moderation algorithms to improve internet safety, better facial 
recognition algorithms with less bias against darker facial tones, financial services companies 
improving existing lending algorithms). 

There is no denying the importance of startups fighting climate change or protecting vital information. However, we differentiate between startups that actively use ethical approaches / algorithms and startups that use regular AI to solve a social issue or environmental problem. For example, a company uses AI to quantify and track an organization’s climate risk, but the technology being used is not substantially more fair, transparent, ethical, or responsible than the industry average. Cybersecurity is an enormous space simply by itself - including it in this project would dilute the other smaller (but equally relevant) categories. 

The categories were created by studying the machine learning pipeline and how data flows through it. It begins at the data level (where the data must be observed and debiased), travels to the model (which also must be debiased and explained), then outputs results (which need to be interpreted and applied). The Governance Risk and Compliance (GRC) category comes from the idea that every machine learning pipeline needs to be audited, governed, and checked every step on the way to manage risk and enforce compliance. The global database (EAIDB) includes an open-source category, but this is replaced in the Nordic Landscape by institutions instead. Most consulting firms fall under the GRC category as well but are broken out separately in the Nordic maps to bring more focus to them. 

Vetting is done primarily through the company website as well as secondary articles, media, or 
research papers. Generally, companies deserving the title “ethical AI” are the first to advertise 
themselves as such since ethics is at the forefront of their overall approach. Some other key terms are “fairness / bias monitoring,” “transparency and trust,” etc. An understanding of the underlying 
technology is sometimes required in the case of startups with whitepapers or research explaining their methods. Overall, it is fairly obvious when a company does in fact belong in the list. However, this approach is not perfect - there are many companies that identify themselves as “ethical”, even when their technology is just average. A more thorough process would involve demos of each and every product, an endeavor that EAIDB has already begun but not yet finished.