Page 53 - 2622_Delhi Police Public School_C-7

Errors in Text and Speech

When a computer is trained with American or British accents, it may misinterpret a strong
accent saying "Turn on the light” as “Turn on the lice". It may also incorrectly mark a text as
misspelt or grammatically wrong based on regional language.

Use of Slang and Colloquial Words

The use of slang or colloquial language can sometimes be misinterpreted or may not elicit
a response. For instance, when a user says, “That game was fire", intending to convey that it

was amazing, NLP might misinterpret this as a reference to danger or damage, leading to a
negative connotation.

Privacy Concerns
NLP tools process large amounts of text and speech data to understand what people say. If this
data isn't protected properly, it can expose private conversations, personal details, or sensitive

information without the user's permission.

Misinformation Bias Use of Slang
• Historical Errors in Text and Colloquial
If an NLP system is trained on incorrect
• Representation and Speech Words
or biased data, it might produce false
or misleading answers. This can lead
people to believe wrong facts, spread
Privacy
rumours, or make poor decisions based Concerns Misinformation

on misinformation.

Brainy Fact

Google’s BERT model, despite its success, has shown gender bias, associating professions
like "nurse" with females and "doctor" with males, highlighting the need for addressing bias

in NLP models.

Ethical Considerations in Using Statistical Data

Ethical consideration in using statistical data means respecting people’s rights, privacy, and

dignity. For example, while working on statistical data of healthcare industry, the privacy of
personal information like patients’ names and, contact details should not be shared unless
people agree.

AI and Ethical Considerations 51