Learning Objective: Examine artificial intelligence impacts on corporate efficiency
Today’s managers state that artificial intelligence (AI) will be among the most transformational technology in the workforce. It affects our lives in a multitude of ways, from banks using anti-money laundering algorithms to call center chatbots that augment customer interactions. All of these are led by AI’s power to increase operational efficiency and make faster, more informed decisions. The more AI influences customer expectations, the greater the impact it will have on the future of work and business decisions.
At the end of this session, participants will be able to:
1. Identify what artificial intelligence is and what its business applications are.
2. Examine the reasons for integrating AI into management and the practical, data-driven methods to quantify opportunities that create a competitive advantage.
3. Understand how to extract business value from AI to develop new, innovative methods for changing how their business operates.
2. Artificial Intelligence:
Improving Corporate Efficiency
• Josephine Bolotski, Sr. Principal Engineer,
Amazon Robotics AI
• Sivakami Sekar Miyamoto, Principal Director,
The Aerospace Corporation
• Darnell Moore, Principal Technical Program Manager,
Amazon University Hubs
• Lesley Yu, Sr. Technical Program Manager,
Amazon Robotics AI
3. What is Artificial Intelligence?
• “Artificial intelligence (AI) is intelligence
demonstrated by machines, as opposed to the
natural intelligence displayed by humans or
animals.” – Wikipedia
• Machine Learning (ML) is a subset of artificial
intelligence that solves specific tasks by learning
from data and making predictions.
Today, we will focus on Machine Learning
4. ML vs. traditional programming
Computer
Computer
Output
Input
Input
Output
Training Data Model
ML
generates
the program
SDEs write the program
ML can detect patterns that are not obvious, where
SDEs wouldn’t be able to write the rules.
5. How a model is trained
Evaluate model on
Validation Set
Evaluate model on Test Set
Train model on
Training Set
Tweak model according to results on
Validation Set.
Pick model that does best on
Validation Set.
Very important to
separate data sets!
6. Different types of ML
Source: https://wordstream-files-prod.s3.amazonaws.com/s3fs-public/machine-learning.png
7. ML Use Cases:
Website Recommendations
Personalized Recommendations:
• Inspired by your shopping trends
• Frequently bought together
• Related products
10. ML Use Cases:
Fulfillment
Quality
Defects: wrong item, missing
item, extra item, damage.
Eligibility
Certain process paths have certain
restrictions. Fragile items, hazardous
items, liquid items, round/spherical
items, weight < X lbs, length < Y in.
14. Resources for
Women of Color in AI
• https://www.latinxinai.org/
• https://blackinrobotics.org/
• https://blackinai.github.io/
• https://peopleofcolorintech.com/category/ai/
The biggest difference between ML and traditional programming is who writes the program. In traditional programming, software development engineers (SDE) write the program by specifying the rules that tell the computer how to turn input into output. ML, instead, uses learning algorithms to find patterns (rules) between input and output from numerous known input-output pairs (training data) to automatically generate the program (model). The model is then used to turn new inputs into outputs. ML can detect patterns that are not obvious to humans (SDEs wouldn’t be able to write rules for such latent relations). However, for the same reason, ML models also have low interpretability.
It learns from humans using the following process (supervised learning):Data labeling: Humans classify thousands of inputs and assign them an output. This is called Ground Truth. E.g. this is an image of a dog vs. cat.
Data splitting: The labeled data is split into two sets: a training set for constructing the model and a test set for testing its performance. The training set is split into a training set and a validation set.
Training: A learning algorithm goes through input-output pairs in the training set and learn the patterns - what kinds of inputs tend to be what type of output - to produce the ML model.
Tuning: the ML model predicts the outputs for the inputs in the validation set. Because we already know the ground truth, we can evaluate the model’s performance and adjust the model’s parameters to fine-tune the results.
Testing: Once a final model is chosen, the ML model predicts the output for the inputs in the test set. We can evaluate the model’s performance on the test set to estimate how it will perform in production.
There are three major types of ML: supervised learning, unsupervised learning, and reinforcement learning. Supervised learning learns a function that maps an input to an output based on example input-output pairs. The process of an algorithm learning from the training dataset can be through of as a teacher supervising the learning process. <examples of supervised learning>Unsupervised learning looks for previously undetected patterns in a data set with no pre-existing labels and with a minimum of human supervision. Algorithms are left to their own devises to discover and present the interesting structure in the data. Reinforcement learning is concerned with how software agents ought to take actions in an environment in order to maximize the notion of cumulative reward. The focus of reinforcement learning is on finding a balance between exploration (of uncharted territory) and exploration (of current knowledge). Note: will need to find license-free version of this image.