https://www.coursera.org/learn/ai-for-everyone
ANI = narrow (single function, self-drive car, smart speaker)
AGI = general (what human can do)
Supervised learning = translation, text transcript, spam filtering, visual inspection [ take input A->output B] (easy things that can be done in less than 1 second)
More data (Big data) + Large neutral network —> good performance & good values
Data – manual labelling / from observing user & machine behaviour / download from websites & partnerships
IT feed data to AI team —> AI advises IT on data collection. Not all datas collected are valuable, let AI team have a check first!
Data is messy
data problem —> incorrect labels, missing values
Multiple types of data (unstructured data) —> images, audio, text
Machine learning – field of study that gives computers the ability to learn without being explicitly programmed [output: software]
Data science – science of extracting knowledge & insights from data [output: PPT]
Deep learning = artificial neutral network
AI company – strategic data acquisition, unified data warehouse, automation, will have Machine Learning Engineer (MLE)
ML works well when (1) simple concept, (2) plenty of data available
Smart speaker AI pipeline
1) trigger word / wake word detection (hey Siri)
2) speech recognition
3) intent recognition —> timer
4) specialised program to execute command
Self-driving car AI pipeline
1) Image/Radar/Lidar
– Car detection + trajectory prediction
– Pedestrian detection + trajectory prediction
– Lane detection
– Traffic light detection
– Obstacle detection
1) GPS + maps
2) Motion Planning (steer/accelerate/brake)
Roles
Software engineer
Machine learning engineer (A->B mapping)
Machine learning researcher
Data scientist (examine & provide insights)
Data Engineer (organise data)
AI Product Manager
AI limitation
– performance limitation (when data is limited)
– Hard to explain themselves
– Biased AI through biased data
Discrimination/Bias
– Learning unhealthy stereotype from data (e.g. man as computer engineer, woman as homemaker)
– may affect hiring tool, facial recognition
Adversarial attacks on AI
– fool the AI by physical changes (e.g. cannot recognise stop sign once have graffiti)
Adverse uses of AI
– DeepFakes, Oppressive surveillance, generate fake comments
impromptuz 