Artificial intelligence (A Level) flashcards

Topic 16: Artificial intelligence

When is a system often called artificially intelligent?

There is no one accepted definition. A system is often called AI if it solves a complex problem using a method like a human's, or one at least as good as a human's, on problems once thought to need human intelligence.

Narrow AI vs generally intelligent AI

Narrow AI works in one field and is what exists and is widely used today. Generally intelligent AI would learn and reason across any field; it is an area of research only.

What is a neural network?

A network of nodes connected in a similar way to neurons in the brain. Each node takes weighted inputs and produces an output, in input, hidden and output layers.

What is backpropagation?

A training method in which the output error is passed backwards through the network and the weights are adjusted to reduce it.

Deep learning vs machine learning

Machine learning is a type of AI whose performance improves based on experience. Deep learning uses neural networks with several hidden layers, allowing more complex problems to be solved.

A node gets inputs 0.6, 0.2 and 0.9 on weights 0.5, −1.5 and 0.8. What is the weighted sum?

0.30 − 0.30 + 0.72 = 0.72. With a threshold of 0.7 the node outputs 1.

How does bias arise in an AI system?

From the training data: if a group is under-represented, or the historical decisions in the data were unfair. Careful selection of data and testing each group separately reduce the risk.

Give three benefits and three risks of AI

Benefits: more consistent decisions; very large data sets analysed quickly; continuous availability; lower cost. Risks: job losses; bias; false information from bad training data; plagiarism; surveillance.