What should a nurse do when non-speakers of the local language come to the ward seeking information about a loved one? What should a receptionist do when they need to book an appointment and a language barrier takes them by surprise? How can an emergency call handler let a caller know that a human interpreter is being contacted? Chapter 3 examines circumstances in which the risks of multilingual AI are ostensibly low. It proposes a distinction between ancillary and core communication but argues that communicative settings are fluid. What starts as ancillary communication can easily turn into core care, so risk is not associated with specific roles or with levels of professional seniority. The chapter argues that, in the sectors under analysis, communication is rarely risk-free. Even where machine translation may not directly lead to harm or loss of life, it may be a feature of complex communicative environments which pose significant systemic risks.

Chapter 4 examines communicative settings where the use of machine translation is particularly likely to involve high levels of risk. The chapter looks at guidelines about machine translation use and at the issue of consent. Two types of consent are examined, namely using machine translation to seek some type of consent, and consent that concerns whether the use of machine translation itself is consensual. The chapter then explores some of the direct reasons why machine translation is used in high-risk scenarios. These reasons include urgency, service user preferences and unreliable human language services. The project’s participating professionals were not short of stories to tell about human interpreters who had not turned up for appointments or telephone interpreting connections that frequently crashed. Incidents of this nature are considered within a broader context where limited resources and outsourced human language services normalise the reliance on machine translation in ways that increase risks and affect standards of care.