AI as opponent
Semantic Passwords: make AI guess without obvious clues
Semantic Passwords is a concept activation, built per occasion: a player explains a secret word without using the word, its synonyms or banned clue families, and the AI tries to guess it. The language model catches meaning leaks, not only exact banned words, so the clue has to be genuinely clean.
ConceptConcept, built for your event. We scope it, rehearse it with your team and agree a fallback.
The idea
The AI is the guesser and the referee at once. The tension is in the penalty for a leaky clue, which is why the round is short and loud. It runs on a phone, a voice kiosk or a table game.
How it runs on event day
- Attendee
- Gets a word, explains it by voice or text, and watches the AI guess until it is right or the time ends.
- Operator
- Loads word lists and banned families per theme, sets the strictness of the leak check and can override a penalty.
- LED or venue screen
- The countdown, the AI guesses as they land, any penalty flagged and a fastest-clean-guess board.
Inputs and outputs
Inputs
- Secret word and banned families from the organiser
- Voice or typed clues
Outputs
- AI guesses and the time of the correct one
- Leak penalties
- Points to the leaderboard
Moderation and fallback
Human approval. Word lists and banned families are fixed by the organiser, and the leak check can be set to lenient. An operator can overturn a penalty that looks wrong, since meaning checks can misfire.
If something fails
- A named operator can pause, skip or replay any round from the admin panel
- If speech-to-text fails, the round continues with typed clues
- If the model is slow, the operator marks guesses by hand
Privacy and consent
The microphone opens only after a consent tap. Speech is transcribed for the round; the recording is not kept unless the organiser configures it and the notice says so. A typed-answer mode is the alternative for anyone who prefers not to speak.
Under India's Digital Personal Data Protection Act, 2023 the organiser decides the purpose and the retention period, and Sociana configures the consent notice, the withdrawal route and deletion on request. Collect only what a round needs. Details sit on the data protection and AI pages in the trust centre; nothing here is a legal opinion. See responsible AI, data protection and DPDP and Smart Lens and biometric consent.
What drives the cost
Price band: ₹35,000–₹75,000 per activation per occasion, ex-GST (add-on menu). Per-completion model costs are quoted on actuals at proposal; no per-completion rate is published.
- Rounds per day times players per round, since every completion triggers one or more model calls
- Model mix for the round: speech-to-text and a language model that checks for meaning leaks
- Strictness of the leak check, which adds a model call per clue
- Peak concurrency, which sets queueing and the number of kiosks or tablets
- Retention of outputs: replays, galleries and downloads add storage and moderation time
- Rehearsal and operator time on site; kiosks, tablets, cameras and LED walls come from the venue or a hardware partner, not from Sociana
Questions people ask
How does the AI know a clue leaks the word?
A language model compares the clue with the secret word and its banned families by meaning, not only by spelling. It can misjudge, so the organiser sets the strictness and an operator can overturn a penalty.
Does it need voice?
No. It works with typed clues on a phone. Voice is offered for a kiosk or a table game, and typed play is always available as the alternative.
Can we use brand terms as the words?
Yes. A product line, a value or a sector term gives the round a brand theme. The organiser approves the list, which is also the answer key.
Want this on your screens?
Tell us the occasion and the audience. We will show you the activation running with your branding.