How the AI Training Bot uses AI

The AI Training Bot (AITB) is an artificial-intelligence system. This page explains, in plain language, where AI is used, what it does and does not do, and its limits. For more on how your personal data is used, see the Privacy Policy.

This product uses AI

AITB uses generative AI to:

Understand your speech and detect the language you speak (English/Singlish, Mandarin, Malay, Tamil).
Play the role of a client — an AI avatar that responds in character and shows changing emotions, so you can practise difficult conversations.
Assess each session — after a session, an AI model reads the transcript and produces feedback and scores against the training rubric.

What the AI does not do

It does not make real decisions about real members of the public. The clients you practise with are fictional, AI-generated characters — no real client data is involved.
The scores and feedback are indicative training signals to support your learning — they are advisory, not determinative. They are not an official appraisal and are not used to rank, grade or make employment decisions about you.
A human remains responsible for any real decision; the AI supports practice and reflection only.

Limitations to be aware of

Speech recognition and per-language accuracy can vary, and the avatar can occasionally be inconsistent or imperfect.
The avatar is a simulation, not a real person — treat it as realistic practice, not a definitive model of any individual.

Bias and fairness

Like all AI systems, the models behind AITB are trained on large amounts of data and can carry biases from that data. In practice this could show up as uneven quality across the four languages, or as a simulated client that leans on a stereotype. Because the assessment involves real people, we take this seriously:

We monitor the system for biased or uneven behaviour — including how it performs across English, Mandarin, Malay and Tamil — and test it for stereotyped or culturally insensitive output.
Where issues are found, we work to reduce them, and the personas and assessment criteria are reviewed with MSF subject-matter experts.
AI bias cannot be fully removed, so the AI's scores remain advisory only and a human stays responsible — which keeps the impact of any residual bias limited.

Your data

Your session voice recordings, transcripts and assessment results are stored to give you feedback and progress over time, to allow trainers to review sessions, and to check the accuracy of the speech-to-text transcription.
Your voice recordings are retained for two reasons: so trainers can review your training interactions, and so we can check the accuracy of the speech-to-text (STT) transcription against the source audio. Recordings, transcripts and assessment results are stored securely — encrypted, access-controlled, held in Singapore — and handled under the PDPA, with a set retention period (90 days by default) after which they are securely deleted (your administrator can configure this).
Your training data is not used to train AI models or for external analytics without explicit approval.
Data is held in line with the Personal Data Protection Act (PDPA) and the project's security and data-protection measures.

Reporting a concern

If the AI behaves in a way that seems harmful, unfair, biased or inappropriate — or you have any concern about how it works — you can report it:

Officers: tell your system administrator, or email report@tiny.codes.
Anyone else: email report@tiny.codes.

You can report in good faith without fear of penalty or retaliation, and confidentially if you prefer. We acknowledge reports, look into them, and act where needed.

Questions

For anything else, contact your system administrator.