Artificial Intelligence

Candidates for this exam are AI enabled learners and are prepared for the professional use of AI by understanding how AI can be used to solve problems.

Defining the AI problem

1.1 Identify the problem we are trying to solve using AI (e.g., user

segmentation, improving customer service)

• Identify the need that will be addressed

• Find out what information comes in and what output is expected

• Determine whether AI is called for

• Consider upsides and downsides of AI in the situation

• Define measurable success

• Benchmark against domain or organization-specific risks to which the

project may be susceptible

1.2 Classify the problem (e.g., regression, unsupervised learning)

• Examine available data (labeled or unlabeled?) and the problem

• Determine problem type (e.g., classifier, regression, unsupervised,

reinforcement)

1.3 Identify the areas of expertise needed to solve the problem

• Identify business expertise required

• Identify need for domain (subject-matter) expertise on the problem

• Identify AI expertise needed

• Identify implementation expertise needed

1.4 Build a security plan

• Consider internal access levels or permissions

• Consider infrastructure security

• Assess the risk of using a certain model or potential attack surfaces (e.g.,

adversarial attacks on real-time learning model)

1.5 Ensure that AI is used appropriately

• Identify potential ways that the AI can mispredict or harm specific user

groups

• Set guidelines for data gathering and use

• Set guidelines for algorithm selection from user perspective

• Consider how the subject of the data can interpret the results

• Consider out-of-context use of AI results

1.6 Choose transparency and validation activities

• Communicate intended purpose of data collection

• Decide who should see the results

• Review legal requirements specific to the industry with the problem being

solved

2. Managing data to solve the AI problem

2.1 Choose the way to collect data

• Determine type/characteristics of data needed

• Decide if there is an existing data set or if you need to generate your own

• When generating your own dataset, decide whether collection can be

automated or requires user input

2.2 Assess data quality

• Determine if dataset meets needs of task

• Look for missing or corrupt data elements

2.3 Ensure that data are representative

• Examine collection techniques for potential sources of bias

• Make sure the amount of data is enough to build an unbiased model

2.4 Identify resource requirements (e.g., computing, time complexity)

• Assess whether problem is solvable with available computing resources

• Consider the budget of the project and resources that are available

2.5 Convert data into suitable formats (e.g., numerical, image, time

series)

• Convert data to binary (e.g., images become pixels)

• Convert computer data into features suitable for AI (e.g., sentences

become tokens)

2.6 Select features for the AI model

• Determine which features of data to include

• Build initial feature vectors for test/train dataset

• Consult with subject-matter experts to confirm feature selection

2.7 Engage in feature engineering

• Review features and determine what standard transformations are needed

• Create processed datasets

2.8 Identify training and test data sets

• Separate available data into training and test sets

• Ensure test set is representative

2.9 Document data decisions

• List assumptions, predicates, and constraints upon which design choices

have been reasoned

• Make this information available to regulators and end users who demand

deep transparency 

3. Building an AI model that solves the problem

3.1 Consider applicability of specific algorithms

• Evaluate AI algorithm families

• Decide which algorithms are suitable, e.g., neural network, classification

(like decision tree, k means)

3.2 Train a model using the selected algorithm

• Train model for an algorithm with best-guess starting parameters.

• Tune the model by changing parameters

• Gather performance metrics for the model

• Iterate as needed

DOWNLOAD HERE