Week 3 Flashcards
Algorithmic thinking and management of AI
What is AI?
AI involves an attempt to build computer systems that think and act like humans
Grand vision: Computer hardware or software systems that are as smart as humans
Realistic vision: Syt
stems that take data inputs and produce outputs, and that can perform complex tasks that are difficult or impossible for humans to perform.
Examples of AI:
- Recognize faces
- Intepret CT scans
- Analyze millions of financial records
- Detect patterns in very large Big Data databases
- Answer questions from humans
- Play chess
- Navigate a car
Major types of AI
- expert systems
- neural network
- machine learning
- deep learning
- intelligent agents
- genetic algorithms
- natural language processing
- robotics
- computer vision systems
What do expert systems do?
Capture tacit knowledge in very specific and limited domain of human expertise. Capture knowldge as a set of rules. Peform very limited tasks.
What are expert systems used for?
Expert systems are used for discrete, highly structured decision making.
Components of Expert Systems
- User interface: most crucial part- takes the user’s query and sends it back to the interface engine following that, it displays results to the user
- Inference engine: searches through the rules to solve a specific problem, formulates conclusion
- Knowledge base: repository of hundreds of thousands of rules
Benefits of expert systems:
- improve decisions
- reduces costs and training time
- reduces errors
- enable better quality and service
Limitations of expert systems:
- experts can’t explain how they make decisions
- rules change and need to be continually updated
- knowledge base can become chaotic
- not useful for unstructured data
- do not scale well to very large datasets
Machine learning
- ML begins with a very large data set
- Main focus is on finding patterns in data and classifying data into known and unknown outputs
- Different pardigm than expert systems
- Most of today’s Big Data analytics use machine learning
- Patpal, spotify
How do we classify machine learning?
Supervised learning: Humans provide examples of desired inputs and outputs. defined by its use of labled datasets to train algorithms to predict desired output.
Unsupervised learning: Humans do not provide examples. Uses machine learning algorithms to analyze and cluster unlabled datasets.
Neural networks
- insipred by the structure and function of human brain
- finds patterns and relationships in massive amounts of data too complicated for humans to analyze
How do neural netowrks learn patterns?
- learn patterns by searching for relationship, building models and correcting over and over again
- humans train it by feeding it inputs for which outputs are known, to help neural network learn from human experts
Examples of neural networks
- face recognition, speech recognition, and natural language processing
What are deep learning neural networks?
- more complex, with many layers of transformation of input to get to desired output
- used almost exclusively for pattern detection on unlabled data
- come closest to grand vision of AI
Genetic algorithms
- useful for finding a solution to a rpoblem by examining many possible solutions
- based on process of evolution
- search among solution variables by changing and reorganizing parts through processes such as: inheritence, mutation, and selection.
- used in optimization problems
- able to evaluate many solutions quickly
Limitations of machine learning and neural networks
- require large data sets
- hard to understand how they arrived at a solution
- most important life decisions don’t have large data sets
- not all patterns are sensible (require human judgement)
- AI systems have no sense for ethics
Natural Language Processing
- understand and speak in natural language
- read natural language and translate
- google translate, siri. alexa
- not useful for a common sense human convo
Computer Vision Systems
Digital image systems that create a digital map for an image and recognize this image in large data bases of imagies near real time.
Computer vision examples
- Facebook DeepFace can identify friends in photos across their systems and the entire web
- Autonomous vehicles
- Passport control
Robotics
- Design, construction and operation of movable machines that can substitute for humans in many factory, offices and home applications.
- Generally designed to perform detailed and specific tasks in limited domains.
- Used in dangerous situations like bomb disposals.
- surgical robots
Intelligent Agents
- Work without direct human intervention to do repetitive, predictable tasks. (deleting junk emails, finding cheapest airfare)
- Use limited built-in or learned knowledge base
- Chatbots
What are agent based modelling applications of intelligent agents?
Model behavior of consumers, stock market and supply chains, used to predict epidemics.
Use of AI at work
Algorithms can act as invisible managers:
- Algorithms can engage in task coordination (Uber)
- Algorthims can exercise soft-surveillance through data collection
Risks associated with the use of AI at work
- algorithms can nudge workers to behave in a certain way
- AI can be biased in recruitment selection
- lack of transperancy: we don’t know how decions were made
- secuirty risks: cyberattacks
- ethical concerns: privacy, autonomy and accountability
AI in HR
- AI can be used in the selection process of candidates
- it can interpret external data, learn from it and use that to perform recruitment tasks efficiently
Stages of recruitment and AI application
- Used across various stages: outreach, screening, assesment and facilitation
- Help target communication, screen CVs, de-bias job ads, analyze video interviews, and faciliatet selection process like scheduling
Challenges for AI in HR
- ethical concerns, such as the risk of algorithm bias and reduction of human interaction
- skepticism about wether AI can adaquately capture the complexity of human behavior, subtleties of interperonal interaction, and emotional intelligence
Amazon AI tools
- implemented to enhance recruitment efficiency
- designed to identify patterns in resumes to predict candidate success
- trained on data from a 10-year period
Issue with amazon tools
- developed bias against female candidates
- learned from male-dominated resume patterns in tech industry
Consequences of Amazon Tool
- discrimination against women in technical job applications
- ethical concerns regarding fairness
- legal implications related to equal employment opporunity
Interactive AI
Also called multimodality, allows text, images, videos and other input to be combined.
