Goa’s AI ambitions must be matched by human oversight, transparent decisionsand accessible grievance redressal when automated systems fail residents
GLEB TSIPURSKY
Goa’s AI Policy 2026 was launched onOctober 22 with promises of better digital public services, new skills andopportunities. Ministers have emphasised inclusion and responsible technologyuse. The real test, however, is what happens when an AI-supported servicemisunderstands a resident who needs an answer urgently.
The government’s earlier draft policyproposed a Konkani-first strategy and multilingual public services, animportant direction in a state where residents move between languages anddialects. But fluent responses do not guarantee accuracy. Administrativedecisions must also be understandable, contestable and reversible.
When AI gets it wrong
Consider a hypothetical fishingbusiness applying for a routine certificate. A digital assistant misinterpretsa document because a term carries different meanings in Konkani and English. Itconfidently asks the applicant to resubmit material already provided. Afterfollowing the instructions twice, the applicant loses a working day and returnsto the same automated screen.
The system may record three completedtransactions. For the resident, it has failed three times.
Public services must ensure that suchfailures do not become the measure of digital success.
Build human oversight into services
First, every AI-supported governmentservice should identify a designated human office authorised to review disputedresults and correct underlying records. Residents should be able to requestassistance in the language they are using. Referring them to a general helpdesk is inadequate if its staff cannot access or amend the decision.
Second, services must distinguishbetween advice and authority. A chatbot may explain which documents usuallyaccompany an application, but it should not issue a definitive refusal orchange eligibility without a documented, authorised process. Residents mustknow whether they are receiving general guidance, an automated recommendationor a formal government decision.
Third, agencies should maintainrecords of submitted documents, applicable rules, decisions and subsequentcorrections. An AI-generated summary cannot substitute for this evidence, as itmay omit the very misunderstanding that triggered a complaint.
Services must also be tested withpeople likely to be overlooked, including speakers of different Konkanivarieties, residents switching between Konkani, Marathi, Hindi and English,older citizens and users of basic mobile devices. Realistic tasks should revealwhether people can identify errors and find correction channels withouttechnical expertise or personal connections.
These tests should be part ofprocurement. Vendors must demonstrate how errors are detected, appealed andcorrected, and how government departments can override systems even whentechnical providers are unavailable.
Make accountability measurable
The Goa AI Tech Conclave has exploredpublic-sector solutions through a government-sponsored hackathon. The nextphase should prioritise correcting mistakes. A useful prototype would letresidents report incorrect automated instructions, receive a case reference,track human review and access the corrected official response.
The government should publishstraightforward performance measures, including the percentage of disputesresolved, median time taken for human review, repeat submissions caused byincorrect guidance and whether errors are concentrated in particular languagesor services. Aggregated reporting can provide accountability without exposingpersonal documents.
Some errors will remain inevitable.The objective is to prevent automated mistakes from becoming endlessadministrative dead ends.
Goa can use AI to reduce queues,translate guidance and help civil servants work faster. Its distinctiveadvantage could be a stronger guarantee to every resident: when a computer getsyour case wrong, you can reach a person empowered to put it right.
(The writer is a PhD, a behaviouralscientist, CEO of Disaster Avoidance Experts, and author of The Psychology ofAI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026)
