Skip to main content Skip to secondary navigation

SQ1. What are some examples of pictures that reflect important progress in AI and its influences?

Main content start
SQ1 Photo
Caption: The GAN technology for generating images and the transformer technology for producing text can be integrated in various ways. These images were produced by OpenAI’s “DALL-E” given the prompt: “a stained glass window with an image of a blue strawberry.” A similar query to a web-based image search produces blue strawberries, blue stained-glass windows, or stained-glass windows with red strawberries, suggesting that the system is not merely retrieving relevant images but producing novel combinations of visual features. From:
RoboCup challenge
Ball control, passing strategy, and shooting accuracy have continued to improve over the quarter century the RoboCup competition has been held. While still dominated by human players, even in their researcher clothes, the best robot teams can occasionally score in the yearly human-robot match. Peter Stone, the AI-100 Standing Committee chair, is shown here taking a shot in the RoboCup 2019 match in Sydney, managed by ICMS Australasia. From:…
AI research on cooperation lags behind that of competition. Recently, the community has begun to invest more attention in cooperative games like Hanabi, shown here. Researchers at Facebook AI Research have shown that a combination of deep reinforcement learning and a “theory of mind”-like search procedure could achieve state-of-the-art performance in this cooperative game. It remains to be seen whether AI strategies learned from AI-AI cooperation transfer to AI-human scenarios. From:
Google AI Example
Widely available tools like Google Docs’ grammar checker uses transformer-based language models to propose alternative word choices in near-real time. While prior generations of tools could highlight non-words (“I gave thier dog a bone”), or even distinguish common word substitutions based on local context (“I gave there dog a bone”), the current generation can make recommendations based on much more distant or subtle cues. Here, the underlined word influences which word is flagged as a problem from 9 words away. Image credit: Michael Littman via
Portrait of person
Neural networks, trained on tens of thousands of portrait photographs of faces, can now generate novel high-resolution images that appear compellingly like pictures of real human faces. The technology behind this development, generative adversarial networks (GANs), has advanced rapidly since its introduction in 2014. Current versions still include telltale visual artifacts, like the strangely absent right shoulder in this image. Nonetheless, the previously unattainable level of realism raises concerns about the use of this technology to spread realistic disinformation. From: .
Two people side by side on social media site
Caption: Facial recognition technology, demonstrated here via Google Photos on a 2019 photo taken at an AI conference, can spot a wide range of individuals in photos and associate them with their names. Applying the same ideas to massive collections of imagery posted online makes it possible to spot and name strangers in public. The capability raises concerns about how AI can simplify mass intrusions into the privacy rights of citizens by governments and private companies all over the world. From: Michael Littman and
Corporate participation in academic conferences has been expanding. At flagship conferences like NeurIPS, nearly a third of all papers include a Fortune Global 500 affiliation. From:….
Neural-network language models called “transformers” consisting of billions of parameters trained on billions of words of text, can be used for grammar correction, creative writing, and generating realistic text. In this example, the transformer-based GPT-3 produces a natural sounding product description for a non-existent, and likely physically impossible, toy. From:…
An original image of low resolution and the resulting image of high resolution
Image-generation GANs can be used to perform other tasks like translating low-resolution images of faces into high resolution images of faces. Of course, such a transformation is not recovering missing information so much as it is confabulating details that are consistent with its input. As an example, the PULSE system tends to generate images with features that appear ethnically white, as seen in this input image of former US President Barack Obama. From:…
Percentage per year Employment-Population Ratio graph for ages 25-54 years
Data from the US Bureau of Labor Statistics shows that employment as a fraction of the population reached a 20-year high right before the pandemic, suggesting that the growth of AI is not yet producing large-scale unemployment. From:
Partially filled crossword puzzle from 2021 American Crossword Puzzle Tournament by Keven G. Der and Will Shortz.
The crossword-solving program Dr. Fill, spearheaded by Matt Ginsberg with contributions by Dan Klein and Nicholas Tomlin, combined generic deep learning approaches, classical AI, and some crossword-specific domain knowledge to best all comers in the 2021 American Crossword Puzzle Tournament for the first time. Combining machine learning and more traditional reasoning is likely to continue to be a winning recipe. Although the program made more mistakes than the human champions, its superior speed of solving put it at the top of the leaderboard. The puzzle shown here is a partial fill of one of the tournament puzzles. Image used with permission of the system authors.
COCO Val2017 Single Person Image Set
The use of dataset datasheets and model cards are two recent proposals for documenting the inputs and outputs of machine-learning systems so that they can be used responsibly in applications. This example model card comes from the MoveNet.SinglePose model that predicts body position from images. From:…
Person talks to the ElliQ device.
In-home sensors and robots are on the rise, offering new ways to provide support and care, but also raising concerns about the negative impacts of pervasive surveillance. The ElliQ robot is shown here. Image from

Cite This Report

Michael L. Littman, Ifeoma Ajunwa, Guy Berger, Craig Boutilier, Morgan Currie, Finale Doshi-Velez, Gillian Hadfield, Michael C. Horowitz, Charles Isbell, Hiroaki Kitano, Karen Levy, Terah Lyons, Melanie Mitchell, Julie Shah, Steven Sloman, Shannon Vallor, and Toby Walsh. "Gathering Strength, Gathering Storms: The One Hundred Year Study on Artificial Intelligence (AI100) 2021 Study Panel Report." Stanford University, Stanford, CA, September 2021. Doc: Accessed: September 16, 2021.

Report Authors

AI100 Standing Committee and Study Panel 


© 2021 by Stanford University. Gathering Strength, Gathering Storms: The One Hundred Year Study on Artificial Intelligence (AI100) 2021 Study Panel Report is made available under a Creative Commons Attribution-NoDerivatives 4.0 License (International):