Just like many of you, I am also trying to get a grasp of what is Artificial Intelligence (AI), its use cases in the real world and most importantly how reliable it is.
There is no dearth of information however, finding answers is not so easy as it seems. As I embark on this journey to explore more about AI and its various facets, a series of small articles can provide you with some of these answers and a path to explore further along with me.
In the first part of this series, we will go through the basics of Artificial Intelligence and prepare the foundation. In the second part, we will discuss about Explainable AI (XAI), its use cases, XAI methods and why it is important. In the final part of this series, we will explore how to design intelligent applications with Explainable AI to make them trustworthy.
First things first, what is Intelligence anyways? You will find numerous definitions on the Internet but one that stuck with me is by Dr. Jeff Hawkins. It is the ability to see patterns and predict outcomes based on previous experiences. Such ability exhibited by machines is Artificial Intelligence (AI).
One of the subsets of AI is Symbolic AI that is rule-based mainly used for problem-solving, logical reasoning and decision-making. For example, a chatbot that can be programmed to respond based on set of rules.
Machine learning (ML) on the other hand, learn from the data it is trained with complex algorithms and generate models that can predict outcomes without being explicitly programmed.
ML Algorithms are these procedures that are implemented in code and is executed on training data or historical data.
Models are outcomes of ML algorithms on the data. It is a predication algorithm that can reasonably predict outcomes on new set of data.
Figure 1 Machine Learning Algorithms and Models
There are broadly three statistical approaches used in Machine Learning. We will cover them briefly so that we have a basic understanding of them when we later discuss Explainable AI.
Supervised Learning are ML algorithms that trains on labelled data. The training happens on input-output pair where input is the data and output is the label. Generally, the aim is to learn a mapping function that can predict the correct output for new inputs. The training. A common example is to define a model that can predict if an incoming email is a spam or not.
Unsupervised Learning is a type of learning in which the algorithm tries to find patterns and relationships in data without any explicit labels as outcomes. Unsupervised learning can be applied, for example, in designing marketing strategy based on the input data of customers with various characteristics (age, gender, income, location, spending habits etc.). The algorithm can identify patterns and cluster data, for example, to create a targeted marketing plan.
Reinforcement Learning involves an agent learning how to make decisions in an environment to maximize a reward or minimize penalty that is inspired by how humans learn through trial and error.
Figure 2: Representation of neural network at the core of AlphaGo
Image source: https://medium.com/hackernoon/the-3-tricks-that-made-alphago-zero-work-f3d47b6686ef
AlphaGo is an intriguing example of reinforcement learning. AlphaGo, an artificial intelligence program developed by DeepMind took the world by surprise when it outplayed the world champion Lee Sedol in the game of Go (an ancient board game originated in China). Go is known for its simple rules, complex strategic depth. It is often considered one of the most challenging board games to master. It is said that the game moves have more possibilities than the atoms in the Universe. AlphaGo, learned by observing hundred thousand of Go matches and learned itself the best possible moves in any situation by the concept of rewards.
Resources and further reads:
Generative AI at SAP – an OpenSAP course
SAP Fiori Design Guidelines for Web
UXAI – A visual introduction to Explainable AI for Designers.
Introduction to Explainable AI: Techniques and Design (By Vera Liao)
Building XAI applications with question-driven user-centered design (Blog on Medium by Vera Liao)
Trustworthy AI: How to make artificial intelligence understandable
People+AI Research (PAIR) Guidebook: AI Design Guideline from Google
AlphaGo documentary on YouTube
- SAP Managed Tags:
