There's a lot of hype about AI at the moment, but what can AI, ML, and data sciences do?
- [Instructor] A discussion about cognitive services demands an understanding of artificial intelligence. So let's start with that. Artificial intelligence allows computer to mimic capabilities of the human brain, such as learning, understanding, recognizing, and remembering things. It consists of reasoning, the capability of drawing conclusions from imperfect data, understanding, being able to interpret the meaning of some data such as images, video, or voice, interacting, allowing computers to communicate with people in more natural ways, such as voice or text. That's a huge contrast to algorithm computing, where in the end of the day, the computer is just executing the instructions that you have given, and it enables capabilities never possible before, such as object detection, speech recognition or language translation. So here are some common AI-related workloads that will be relevant for the exam. Machine learning, the ability to make predictions and draw conclusions from incomplete data, such as predicting a stock price, or defining the likelihood of a patient to get diabetes based on his blood exams. Anomaly detection, the automated ability to detect unusual activity in a system. That's quite useful for workloads such as credit card fraud detection, or to predict if a machine is going to fail. Computer vision, the capability of interpreting inputs from video, cameras or images through techniques such as classification, object detection or semantic segmentation. Self driving cars, satellite technologies, or drone footages can benefit from this. Natural language processing, or NLP, being able to interpret and respond to written or spoken language by performing key phrase or sentiment detection, for example. Conversational AI, the ability of a software agent to participate in a conversation. Some very good use cases for this are bots and automated agents. What enabled AI to be widespread nowadays is a combination of three technology advancements. Big data that can be more easily retained with cheaper storage solutions. In general, the more data is available, the more accurate the AI models will be. Higher processing power through both the advent of GPUs, which are much faster at processing machine learning models, and the scale-out capabilities of the cloud, which allows a workload to be processed by potentially hundreds of machines. And also better algorithms, as both ML and AI research have been quite active in the academia. And better algorithms, as both ML and AI research have been quite the active in academia. What we can conclude from this is that AI models generally require a lot of data and processing power, neither of them being cheap. Besides, these models are quite cumbersome to create, test and improve, before you get to a high level of accuracy. Once these models are built, though, they're quite cheap to consume. All these barriers could make adopting AI difficult for the average company, which might not have the data, skills or computing power to get started. Microsoft, however, has the commitment to democratize AI, making it available to every user on various levels, as you'll see on the next video.