Machine Learning (ML)
Qu'est-ce que Machine Learning (ML)?
Cette entrée n'est actuellement disponible qu'en anglais.
Machine Learning (ML) is the subset of artificial intelligence in which a system learns patterns from data instead of following hand-written rules. Engineers train a model on examples — labeled call transcripts, audio, messages — and it generalizes to inputs it has never seen. ML powers speech recognition, sentiment analysis, spam and fraud filtering, and the large language models behind today's conversational AI.
More detail
The classic split is supervised, unsupervised, and reinforcement learning. Supervised models learn from labeled examples and produce the classifiers a contact center uses daily: is this message spam, what is the caller's intent, is this call a robocall. Unsupervised models find structure without labels, such as clustering support conversations into emerging topics. Reinforcement learning trains a policy by reward, and it is part of how speech and dialogue models are tuned after their initial training.
In customer communications, ML is the reason speech-to-text is accurate enough to drive automation, why a message can be scored for sentiment the moment it arrives, and why traffic that artificially inflates billing (AIT fraud) can be flagged before it costs money. Every one of these is a model trained on historical examples, evaluated on held-out ones, and monitored for drift once live — ML in production is an operational discipline, not a one-off training run.
On Devotel Orbit, machine learning runs inside shipped surfaces: sentiment analysis on conversations, speech recognition and synthesis in AI voice agents, and the language models orchestrated by the AI agent runtime. When a vendor claims AI, the practical question is which of these models sits behind the claim and how its accuracy is measured — the related terms below name the ones that matter for communications.
Questions fréquentes
- Is machine learning the same as artificial intelligence?
- AI is the wider ambition — machines performing tasks that would require human intelligence — and machine learning is the approach that got there: learn from examples instead of encoding rules by hand. Practically everything marketed as AI today is ML, and the current generation of conversational products is ML in the form of large language models.
- What is the difference between machine learning and an LLM?
- An LLM is a specific kind of machine-learning model: a transformer trained on very large text corpora to predict and generate language. Machine learning also covers the smaller, single-task models a contact center runs alongside it — spam classifiers, sentiment scorers, speech recognizers. Think of ML as the discipline and the LLM as its most capable current artifact.
- Does a machine-learning model need my data to work?
- General models are trained once on broad corpora and work out of the box for language tasks. Accuracy on your specific conversations improves when the model is grounded in your knowledge base at query time, or fine-tuned on your labeled examples — which is also why data handling, consent, and retention policies matter when a vendor tells you their AI will learn from your traffic.
See also
Build it on Orbit
Voice, messaging, email, video, and AI agents on one platform and one pay-as-you-go bill. Start free — no credit card required.