Column: in AI projects, control matters more than capability
Organizations need governance for AI output, not just better tools. Four checks to pass before adopting AI.
In a column for Money Today, Lab Director Byung Jun Kang, head of the SNU EPM AI4PM initiative, argues that companies focus on the performance of AI tools while lacking systems to control AI output and keep projects on target.
Because AI output looks convincing, people skip cross-checks. In megaprojects with large budgets and workforces, hallucinations can turn into financial losses and legal claims. AI shortens the work but does not take responsibility for the result.
He proposes four checks before AI is adopted: legal risk, by identifying high-risk systems ahead of EU-style AI regulation; company-wide standards, such as an AI management system under ISO/IEC 42001; ongoing monitoring of bias and security vulnerabilities, following frameworks like the NIST AI RMF; and verified results through limited pilots, as PMI recommends.
Work can be delegated to AI. Accountability for decisions cannot.
Original article: Money Today