Multilingual AI assistants: which working language in Basel?
In Basel’s pharmaceutical, life sciences and Rhine logistics organisations, an AI assistant can connect colleagues with different working languages. The management question is specific: should an assistant shared with international R&D operate exclusively in English, or should people be able to work in the language they know best?
What the literature establishes
Helene Tenzer, Markus Pudelko and Anne-Wil Harzing (2014, Journal of International Business Studies) show, through a qualitative study of multinational teams, how language barriers affect trust formation. Difficulties expressing oneself can shape judgements about colleagues’ competence or reliability, beyond the immediate problem of comprehension. For a shared assistant, this suggests examining who can formulate requests, discuss answers and challenge them, rather than simply checking which languages are available.
A different language can also change judgement
Boaz Keysar, Sayuri L. Hayakawa and Sun Gyu An (2012, Psychological Science) observed reductions in certain decision biases when participants used a foreign language in experimental tasks. Their findings are consistent with greater distance from intuitive and emotional reactions. They do not show that a foreign language improves every decision or makes working with AI easier.
The premature conclusion
One might conclude that everything should be in English to improve decisions, or that everything should be translated to remove obstacles. Neither rule follows from the cited research: a foreign language can change some judgements while making expression and discussion more difficult. The relevant choice therefore depends on the activity: understanding an instruction, proposing a research hypothesis and challenging a recommendation involve different linguistic demands. (our executive and employee training programmes)
What the evidence cannot settle
These studies examine neither generative assistants nor agents built without coding. Nor do they establish that a given model provides equivalent answers in German, French and English, particularly where technical terminology is concerned. A multilingual interface therefore leaves a practical question unanswered: do users notice and understand the same qualifications in each language?
A practical check in Basel
Within SHR’s “Creating your own AI assistants and agents without coding” programme, one exercise is to configure an assistant that helps prepare sample transfers between an international R&D team and a Rhine logistics partner, using fictional files. Have participants handle equivalent cases in English and in their strongest language, alternating the order and having a bilingual domain professional validate the translations. For each language, measure the proportion of constraints that participants correctly report against a checklist prepared in advance: temperature, deadline, recipient and a condition that prevents the transfer. This check does not certify the assistant for regulated use; it tests whether the choice of language leaves some constraints less well understood. To go further: explore the Building AI Assistants and Agents Without Coding training in Basel in the canton of Basel-Stadt, or browse our executive and employee training programmes in Switzerland.
In pictures: Building AI Assistants and Agents Without Coding in Basel
- building ai assistants and agents without coding
- Basel
- research
- training basel