Skip to Main Content

Artificial Intelligence

a guide to artificial intelligence in medicine and health sciences education

What we're reading right now

What is artificial intelligence?

Artificial Intelligence are “…algorithms which seek to create expert systems which make predictions or classifications based on input data. At its simplest form, artificial intelligence is a field, which combines computer science and robust datasets, to enable problem-solving.” (IBM 2023)

Artificial Narrow Intelligence, also known as "ANI, or weak AI, is the type of AI that has been achieved so far.” (UNESCO)​ ANI is "AI trained and focused to perform specific tasks." (IBM 2023)

Artificial General Intelligence, also known as Strong AI or AGI. “AGI, if ever reached, would be comparable to human intelligence.” (UNESCO) 

for a full glossary of AI-related terminologies, see the AIPRM's "Ultimate Generative AI Glossay" (and Thank you to the STEM Club folks of Fuller Library in London, UK, for suggesting this resource!)

 

Infographic with two large bubbles full of images. First bubble is labeled "We are here" and "Artificial narrow intelligence", and includes imaged of Amazon's Alexa, ChatGPT, Bard, and Med Palm-2 logos, a fitness tracker watch, and a roomba vacuum. Beneath these images in the first bubble are two categories of text. The first, labeled "Types of AI", includes machine learning, natural language processing, speech recognition, and facial recognition. The second category of text is labeled "Applications", and includes wearables, chatbots, curated product recommendations, and self-driving cars. The second bubble is labeled "Not here" and "Artificial general intelligence", and includes images of Data from Star Trek and Hal9000 from 2001, a Space Odyssey

Different kinds of Artificial Intelligence include:

  • Machine Learning: "Rather than being programmed with rules to produce answers, computers receive data and the answers expected from the data and, as a result, produce rules by identifying patterns between the two” (UNESCO)"
    • artificial neural networks: a kind of AI machine learning that uses "...a method of mimicking the way the human brain learns, through its connection of neurons, using a computer model.3 This form of AI can evaluate complex relationships between inputs and outputs through a hidden layer (or layers) of calculations" (CADATH 2018)
    • deep learning: "...a form of artificial neural network with many hidden layers between inputs and outputs that allow the program to analyze complex data of various structures.8 In health care, a common form of deep learning is the convolutional neural network" (CADATH 2018)
    • generative AI: deep learning models that use data to “generate statistically probable outputs when prompted" (IBM 2023) ChatGPT is an example of generative AI.
    • large language models: “… trained on large text datasets to learn to predict the next word in a sentence and, from that, generate coherent and compelling human-like output in response to a question or statement." (UNESCO)
      • examples: ChatGPT, Bard
    • Natural Language Processing​: "...(NLP) is a branch of AI concerned with understanding and interpreting human language.6,10,11 In health care, NLP could be used to analyze the content of electronic medical records or as an automated agent to respond to patient questions" (CADATH 2018)
      • example: Google search engine

Public AI awareness

2022 PEW Research Center figure titled "Half of American or more aware of common uses of AI, but fewer can identify AI's role in all six examples: % of U.S. adults who identify that the following use artificial intelligence in multiple choice questions", wearable fitness trackers that analyze exercise and sleeping patterns, 68%; a chatbot that immediately answers questions, 65%; product recommendations based on previous purchases, 64%; a security camera that sends an alert when there is an unrecognized person at the door, 62%; a music playlist recommendation, 57%; and the email service categorizing an email as spam, 51%. In the second part of the figure, "% of U.S. adults who correctly identify the following as using AI, 30% identified all of the previous categories as using AI, 38% identified 3 to 5 of the previous categories as using AI, and 31% identified 2 or fewer of the previous categories as using AI

How does ChatGPT work?

ChatGPT is "...a variant of the GPT (Generative Pre-training Transformer) language model, which was developed by OpenAI... trained to generate human-like text by predicting the next word in a sequence based on the words that come before it" (response from ChatGPT on 1/3/2023) (Montclair State 2023)

ChatGPT is a type of Artificial Intelligence, one of many programs called "large language models" which are “… trained on large text datasets to learn to predict the next word in a sentence and, from that, generate coherent and compelling human-like output in response to a question or statement. In the case of ChatGPT, 570gb of data representing 300 billion words have been supplied to the system and it has around 175 billion parameters.” (UNESCO)

 

a diagram depicting a form of machine learning called "reinforcement learning with human feedback" from Sasha Luccioni's April 12th 2023 ArsTechnica article, "The mounting human and environmental costs of generative AI". In the figure, images of wikipedia, common crawl, and Reddit logos are labeled as training data which are inputted into a labeled image of an initial language model, which in turn lead to model-generated text and combined with human-provided prompts, and screened by humans, and finally a series of rainbow bars completes the diagram and represents the AI output of text​

Above is an image depicting a form of machine learning called "reinforcement learning with human feedback" from Sasha Luccioni's April 12th 2023 ArsTechnica article, "The mounting human and environmental costs of generative AI"

This process of training LLMs on large text datasets uses a process called "machine learning", where, “Rather than being programmed with rules to produce answers, computers receive data and the answers expected from the data and, as a result, produce rules by identifying patterns between the two” (UNESCO)​​

Recommended Readings

Our choice for understanding how ChatGPT works:

 

Measured responses to AI drama:

  • John Warner, the author of the book Why They Can’t Write, has been railing against the five-paragraph essay for years and wrote a Twitter thread about how ChatGPT reflects this rules-based, standardized form of writing: “Students were essentially trained to produce imitations of writing,” he tweeted. The AI can generate credible writing, but only because writing, and our expectations for it, has become so unaspiring. (Bogost 2023)
  • Bogost, I. (2022, December 7). ChatGPT Is Dumber Than You Think. The Atlantic.
  • In New York Magazine, John Herrman analyzed the shifting nature of ChatGPT, which could be interpreted as a decline in the tool’s effectiveness, an impression perhaps confirmed by a Stanford research team’s finding that ChatGPT went from producing correct responses to math problems at a high of 90 percent to down to a rate of less than 3 percent in recent months.(Gannon 2023)

 

Ethical issues, environmental and human costs:

 

Baked-in Bias:

 

Why ChatGPT lies:

 

Being human in the age of AI:

 

Comprehensive Briefs:

 

Bibliographies: