MM1041 Chap.5 AI, Machine Learning and Deep Learning
AI, Machine Learning and Deep Learning
Three nested fields
Artificial intelligence, in the course's definition, is the study and building of computer systems that can do jobs which would ordinarily need a human mind, such as understanding language, reasoning, recognising speech, making decisions, navigating the visual world and manipulating objects.
Machine learning is the part of AI focused on getting computers to learn without being explicitly programmed. Deep learning, the most recent stage in the e-learning module's sequence of models from rule-based through statistical, uses many-layered neural networks.
Each field sits inside the one before it.
Weak and strong AI
The first lecture separates weak AI, very good at one specific task such as playing chess, from strong AI, which would do all the intelligent things a person can do. Every business system the course discusses is weak AI in this sense.
Keeping that in mind prevents an answer from claiming that a recommender or a chatbot understands a business in general.
Why firms adopt AI
The lecture gives three business reasons. AI reduces costs because automation creates savings in almost every industry.
It increases revenue because recommendation and prediction systems lift sales and efficiency, with a short-video feed that learns preferences, predicts interaction and ranks videos as the example.
And it launches new lines of business, illustrated by AI home management.
What learning means
The course quotes Tom Mitchell, whose definition has three parts: a family of tasks T, a way of scoring performance P, and experience E; a program is learning when its P score on T rises as E grows. The e-learning lecturer adds that the simple version is enough, a program that gets better at a task from experience.
The lecture explains why learning is needed at all with tacit knowledge: much of what people know cannot be written as instructions, so machines are trained on examples instead.
A short history
Machine learning is not new.
The e-learning lesson places the first perceptron in the 1950s, built as a circuit, and backpropagation in the 1980s, followed by a winter when programs could not meet expectations, before the recent wave marked by a program defeating a human Go champion. It also explains the slow start: the field draws on probability, statistics, optimisation, biology, psychology and neuroscience.
The module's lesson on neural networks and deep learning continues this story, and the first lecture warns that deep neural networks often cannot explain how they reach an output.
What this chapter covers
- 01
Artificial intelligence as tasks needing human intelligence
- 02
Machine learning without explicit programming
- 03
Deep learning and neural networks
- 04
Weak AI against strong AI
- 05
Cost, revenue and new business lines
- 06
Mitchell's experience, task and performance
- 07
Tacit knowledge and why machines are trained
Worked example · free
Decide whether a pricing tool learns
- 3Apply the definition to the first tool.
- 3Apply the definition to the second tool.
Key terms
- Artificial Intelligence
- The building of computer systems that can do jobs which would ordinarily need human thinking.
- Weak AI
- AI that is very good at one specific task, such as playing chess, and nothing beyond it.
- Strong AI
- Hypothetical AI able to do all the intelligent things a person can do.
- Deep Learning
- Machine learning with many-layered neural networks, the latest stage after rule-based and statistical models.
- Tacit Knowledge
- Knowledge people use but cannot fully put into instructions, such as how to ride a bicycle.
- Neural Network
- A model built from layers of connected units whose connection strengths are adjusted during training.
- Perceptron
- An early learning algorithm from the 1950s, first built as an electronic circuit.
AI, Machine Learning and Deep Learning FAQ
Is machine learning the same thing as artificial intelligence?
No. Machine learning is one part of artificial intelligence. AI covers any system that performs tasks normally needing human intelligence, including systems built from hand-written rules. Machine learning is the part in which the system improves from examples instead of being explicitly programmed, and deep learning is a further subset that uses many-layered neural networks.
Why can't programmers simply write rules instead of training a model?
Because much human knowledge is tacit. We know more than we can tell, and nobody can write complete instructions for recognising a face or riding a bicycle. Training on examples lets the machine find the pattern that a person uses without being able to state it, which is why the lecture concludes that the machine must be trained.
Do I need to memorise Tom Mitchell's definition word for word?
The e-learning lecturer says you do not. What you need is to use it: name the task, the performance measure and the experience for a given system, and check that performance improves as experience grows. Being able to do that for an unfamiliar product is worth more in an answer than reciting the sentence.
Exam move
Make a three-column table of five products you use, recording T, P and E for each, and mark any that turn out to be automation rather than learning. Then write the three business reasons for AI and attach one of your products to each. Both exercises take fifteen minutes and cover most questions this chapter can produce.
Working through AI, Machine Learning and Deep Learning in MM1041? Sia is AskSia’s AI Information Technology tutor — ask any MM1041 AI, Machine Learning and Deep Learning question and get a clear, step-by-step explanation grounded in how MM1041 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
