🗣️ Virtual Personal Assistants

Virtual personal assistants (VPAs) are software programs that use ML to understand human language, answer questions, perform tasks, and learn from interactions. Siri (Apple), Google Assistant, Alexa (Amazon), and ChatGPT (OpenAI) are the most well-known examples. They represent one of the most complex ML applications because they must handle the ambiguity, context, and nuance of human language.

When you say "Hey Siri, what is the weather like today?" a cascade of ML models activates. First, a speech recognition model converts your audio waveform into text (this is called ASR — Automatic Speech Recognition). Then, a natural language understanding model interprets the meaning — it determines you are asking about weather, for your current location, for today. Then, the assistant retrieves the information and a natural language generation model formulates a response. Finally, a text-to-speech model converts that response back into spoken audio. Four separate ML models, all in under a second.

The ML Behind Language Understanding

The breakthrough that made modern VPAs possible is the Transformer architecture, introduced in 2017. Before Transformers, language models processed words one at a time in sequence. Transformers process all words simultaneously using a mechanism called attention — they learn which words in a sentence are most relevant to each other. This is what powers GPT (Generative Pre-trained Transformer), BERT (Google), and the language models behind every modern assistant.

These models are trained on enormous datasets of text — books, websites, articles, conversations — using semi-supervised learning. They learn the statistical patterns of language: which words tend to follow which other words, how sentences are structured, what common questions and answers look like. ChatGPT (GPT-4) was trained on hundreds of billions of words.

💡 Real-World Examples

Google Assistant + Duplex: In 2018, Google demonstrated Duplex — an AI that makes phone calls on your behalf to book restaurant reservations and hair appointments. It uses natural conversation, including "umm" and "uh-huh," and the person on the other end does not realise they are speaking to an AI. This requires speech recognition, language understanding, language generation, and text-to-speech working together seamlessly.

ChatGPT in education: Teachers in Australian schools use ChatGPT to generate lesson materials, create differentiated worksheets, and provide personalised feedback on student writing. The NSW Department of Education released guidelines in 2023 for responsible AI use in classrooms.

Telstra virtual assistant: Telstra's virtual assistant handles millions of customer enquiries per year using natural language processing. It can diagnose network issues, process bill payments, and escalate complex cases to human agents — reducing wait times from 15+ minutes to seconds for common queries.

✏️ Describe how ML enables virtual personal assistants to work. (2–3 sentences)