Page 359 - CyberSecurityEssentialsEnglish
P. 359

Chapter Eight

                Responsible Institutional Adoption and the Human

                                                   Advantage


         From experiment to institutional capability
         Governance, risk, people, and sustainability
         Chapter question: who bears responsibility when an AI system proposes a wrong decision or

         carries  out  an  unsuitable  action,  the  vendor,  the  employee,  the  manager,  or  the
         organization?


         The practical answer is that responsibility is distributed, but it is never delegated to the

         machine.

         Story: When the Bot Got It Wrong, the Organization Was Still Responsible
         In 2024 a civil resolution tribunal in British Columbia, Canada, heard the case of a customer
         who relied on information given by a chatbot on an airline’s website about bereavement
         fares. The bot told him he could request a reduced fare after traveling, within a defined

         period, while the policy published on another page said the opposite. The customer made
         his decision and paid the full price, and the company later refused his request. When the

         case reached the tribunal, the company attempted to separate itself from what the bot had
         said, but the decision affirmed that the system is part of the organization’s website and that
         the  organization  is  responsible  for  the  accuracy  of  the  information  its  digital  channels
         provide [1].


         The story matters neither for the sum of money nor for the type of bot, but for the question
         it exposed: can an organization take the benefit of speed and automation, and then treat
         the  system  as  a  separate  entity  when  harm  occurs?  Practically,  legally,  and  ethically,

         responsibility  does  not  vanish  when  AI  enters  a  process.  The  organization  remains
         responsible for choosing the use, defining the data and the permissions, testing the system,
         monitoring performance, enabling a human being to intervene, and remedying an error

         when it happens.

         This chapter therefore treats governance not as a set of constraints written after a project
         is finished, but as the capability that lets an organization benefit from AI faster and with
         greater confidence. Good governance reduces rework, surfaces problems early, prevents

         the purchase of tools that serve no clear objective, and helps staff know what they may do
         and what needs approval or review.

           “An organization can adopt automation, but it cannot delegate its responsibility to an
           interface.”





        358
   354   355   356   357   358   359   360   361   362   363   364