AI Basics

An overview of Artificial Intelligence

Introduction

Artificial intelligence (AI) has become ubiquitous; from product and movie recommendations on Amazon and Netflix to friend suggestions on Facebook or autofill in search on Google, AI has permeated our daily lives. But what exactly is AI, and how do we design for it?

AI typically refers to human-imitative intelligence; the simulation of human thought processes, such as learning or problem solving in a computerized model. AI systems are often used to recognize patterns, make predictions, and provide insights from large amounts of data.

Some common use cases of AI include: 

  • Computer Vision

    Extracting information from pictures

    Natural Language Processing

    Pulling insights and patterns out of written text

    Robotics

    Autonomously moving through spaces based on sensory input

    Machine Learning

    Looking for patterns in data

AI can be used to augment or automate workflows. Automation can refer to the full or partial replacement of a function previously carried out by a human operator, and as such, can be thought of as a continuum of levels, from the lowest level of fully manual performance to the highest level of full automation: 

Levels of Automation of Decision and Action Selection
  • High
  • 10

    The computer decides everything, acts autonomously, ignoring the human

  • 9

    The computer informs the human only if it, the computer, decides to

  • 8

    The computer informs the human only if asked

  • 7

    The computer executes automatically, then necessarily informs the human

  • 6

    The computer allows the human a restricted time to veto before automatic execution

  • 5

    The computer executes a suggestion if the human approves

  • 4

    The computer suggests one alternative

  • 3

    The computer narrows the selection down to a few

  • 2

    The computer offers a complete set of decision/action alternatives

  • 1

    The computer offers no assistance: human must take all decisions and actions

  • Low

It is important to understand (1) whether to automate or augment a task, and (2) if automating, how much to automate in order to design successful human-AI interactions.

How AI Works

There are various ways to implement AI, but at a high level, there are two approaches: 

  • 1. Rules-based

    The rules-based approach uses algorithms, a process or set of instructions that a computer uses for calculations or problem-solving. 

  • 2. Examples-based

    The examples-based approach uses data to create models. These models are the result of training an AI on data to find patterns.

Examples-based approaches are promising in areas wherein specifying a sequence of rigid rules is difficult, such as diagnosing a disease or recommending a video for someone to watch. This way of problem solving is largely made possible by machine learning. 

Machine Learning

Machine learning (ML) is a subset of AI that uses an examples-based approach to get an AI to accomplish tasks without being given specific instructions, which is especially beneficial when data has several different variables. 

ML uses different ways to teach a machine how to learn:

  • Supervised

    Supervised learning requires labeled data. In other words, it utilizes data grouped into samples that have been tagged with one or more labels. Supervised learning typically uses classification when we want to predict a thing or regression when we want to predict a number. It is effective when there is a clear output in mind (e.g., “is this plant safe or poisonous”).

  • Unsupervised

    Unsupervised learning finds commonalities and patterns in the input data on its own, without using labeled data. It typically uses clustering, which is the grouping of data by some set of characteristics or features. It is effective when there is an unclear output in mind.

  • Reinforcement

    Reinforcement learning rewards positive behavior and punishes negative behavior, which over time, can help the AI determine the optimal behavior for a particular environment or situation. It is effective when the AI is not provided with historical data and instead learns from the environment, collecting and learning from data real-time (e.g., self-driving cars).

AI-Assisted Decision Making

Decision support/augmentation is one of the most common AI applications. A closer look reveals the wide range of possible applications of AI-assisted decision making, many of which involve high-stakes scenarios and a need for explainability.

References

  • Adadi, A., & Berrada, M. (2018). Peeking Inside the Black-Box: A Survey on Explainable Artificial Intelligence (XAI). IEEE Access, 6, 52138-52160. 
  • Alto, V. (2019). Reinforcement Learning: beyond the supervised and unsupervised ways. 
  • IBM. (n.d.). Design for AI. Retrieved from https://www.ibm.com/design/ai/
  • Parasuraman, R., Sheridan, T.B., & Wickens, C.D. (2000). A model for types and levels of human interaction with automation. IEEE transactions on systems, man, and cybernetics. Part A, Systems and humans : a publication of the IEEE Systems, Man, and Cybernetics Society, 30 3, 286-97. 
  • Sheridan, T.B., & Verplank, W.L. (1978). Human and Computer Control of Undersea Teleoperators.
  • Sicular, S., & Aron, D. (2019). Leverage Augmented Intelligence to Win With AI. Gartner. 
  • Sokol, K., & Flach, P.A. (2020). Explainability fact sheets: a framework for systematic assessment of explainable approaches. Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency.
  • Wickens, C.D., Mavor, A., Parasuraman, R., & McGee, J. (1998). The future of air traffic control: human operators and automation.