TechCrunch · Feb 5, 2019
TechCrunch reports on Iconary, an AI drawing-and-guessing game
Markdown · JSON · Original source
On February 5, 2019, TechCrunch’s Devin Coldewey described Iconary, a game in which a person and an AI take turns arranging or guessing at icons to communicate phrases. Coldewey argues that the challenge is not just recognizing objects, but interpreting relationships between them: a group of people beside a soccer ball might suggest a team, while the same figures beside a blackboard might suggest a class. The article says Iconary’s creators at the Allen Institute for AI trained the system on human drawings, while the phrases used in the game were new to the model. Coldewey reports that the researchers described early games in which the AI could depict and interpret phrases with a human partner, and he characterizes its results as sometimes surprisingly human but still far from human ability. Coldewey presents this collaborative task as a small step toward AI that can interpret context, a capacity he says could matter for robots and self-driving cars. Those examples are possibilities, not demonstrated real-world uses in the article. He also notes a left-to-right bias in the agent’s images and says its attempts could be too abstract. The account leaves open how well the system performs beyond these games and whether its way of interpreting scenes transfers to real-world settings.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
TechCrunch · Article
Source details
Original source date: Feb 5, 2019 (2019-02-05).
Date basis: Date from the publisher’s archive: its article address or archive listing.
Added to Escape Meridian: 2026-10-29T00:33:07.461Z. First observed: 2026-10-28T18:26:18.475Z. Retrieved: 2026-10-28T18:27:52.276Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on October 28, 2026; originally published February 5, 2019. The retrieved page is marked as edited after publication. Later additions may be present; this reading relies on the article’s account of the game and its stated early results. The article does not provide performance measures or evidence that Iconary’s approach transfers to robots, self-driving cars, or other real-world settings. The article notes a left-to-right bias and that some of the agent’s representations can be too abstract.
Open question
How well would Iconary’s approach to interpreting relationships between objects work beyond the game, including in real-world settings?
Source-stated
Iconary has players arrange icons to express phrases or guess the computer’s arrangements.
Iconary, as the game’s creators at the Allen Institute for AI decided to call it to avoid lawsuits from Mattel, has you drawing and arranging icons to form phrases, or guessing at the arrangements of the computer player.
Source-stated
Coldewey says the system was trained to recognize relationships between objects, including their type and position.
The AI behind Iconary, then, isn’t a drawing recognition engine at all but one that has been trained to recognize relationships between objects, informed by their type, position, number and everything else.
Source-stated
The researchers said the system had not previously encountered the phrases used in Iconary and described promising early games with human partners.
AllenAI has never before encountered the unique phrases in Iconary, yet our preliminary games have shown that our AI system is able to both successfully depict and understand phrases with a human partner with an often surprising deftness and nuance.
Source-stated
Coldewey says the agent’s images are biased toward a left-to-right arrangement.
Images “read” from left to right, as people tend to draw them, since we also read in that direction, so keep that in mind.
Escape Meridian interpretation
Escape Meridian reads Coldewey’s robot and self-driving-car examples as possible applications, not demonstrated uses of Iconary.
Robots and self-driving cars do need to know how to exceed human capacity in some cases, but it’s also massively important to be able to understand the world around them in the same way people do.
Source and contributor context
TechCrunch · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Ai2
Ai2 develops AI research tools and models, including the openly documented Olmo family.
Editorial provenance
Version 1. Publication ID: caf6c481-7e2a-333c-de96-2bc4eb8b534c.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T00:32:20.222Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: 9d634bcb8bbb37677da3122d495aa3d2e205ea6e8c2edeae4aa52125bb945550. Source text hash: 0cf80afc8117f8d953ddb8da52c260536cf991e0bd53f0d19e4126a21fef4cca.
Revision and source-check history
Version 1 · 2026-10-29T00:33:07.461Z
TechCrunch reports on Iconary, an AI drawing-and-guessing game
On February 5, 2019, TechCrunch’s Devin Coldewey described Iconary, a game in which a person and an AI take turns arranging or guessing at icons to communicate phrases. Coldewey argues that the challenge is not just recognizing objects, but interpreting relationships between them: a group of people beside a soccer ball might suggest a team, while the same figures beside a blackboard might suggest a class. The article says Iconary’s creators at the Allen Institute for AI trained the system on human drawings, while the phrases used in the game were new to the model. Coldewey reports that the researchers described early games in which the AI could depict and interpret phrases with a human partner, and he characterizes its results as sometimes surprisingly human but still far from human ability. Coldewey presents this collaborative task as a small step toward AI that can interpret context, a capacity he says could matter for robots and self-driving cars. Those examples are possibilities, not demonstrated real-world uses in the article. He also notes a left-to-right bias in the agent’s images and says its attempts could be too abstract. The account leaves open how well the system performs beyond these games and whether its way of interpreting scenes transfers to real-world settings.
Original source
TechCrunch · Article
Source details
Original source date: Feb 5, 2019 (2019-02-05).
Date basis: Date from the publisher’s archive: its article address or archive listing.
Added to Escape Meridian: 2026-10-29T00:33:07.461Z. First observed: 2026-10-28T18:26:18.475Z. Retrieved: 2026-10-28T18:27:52.276Z.
Limits
Text as retrieved on October 28, 2026; originally published February 5, 2019. The retrieved page is marked as edited after publication. Later additions may be present; this reading relies on the article’s account of the game and its stated early results. The article does not provide performance measures or evidence that Iconary’s approach transfers to robots, self-driving cars, or other real-world settings. The article notes a left-to-right bias and that some of the agent’s representations can be too abstract.
Open question
How well would Iconary’s approach to interpreting relationships between objects work beyond the game, including in real-world settings?
Source-stated
Iconary has players arrange icons to express phrases or guess the computer’s arrangements.
Iconary, as the game’s creators at the Allen Institute for AI decided to call it to avoid lawsuits from Mattel, has you drawing and arranging icons to form phrases, or guessing at the arrangements of the computer player.
Source-stated
Coldewey says the system was trained to recognize relationships between objects, including their type and position.
The AI behind Iconary, then, isn’t a drawing recognition engine at all but one that has been trained to recognize relationships between objects, informed by their type, position, number and everything else.
Source-stated
The researchers said the system had not previously encountered the phrases used in Iconary and described promising early games with human partners.
AllenAI has never before encountered the unique phrases in Iconary, yet our preliminary games have shown that our AI system is able to both successfully depict and understand phrases with a human partner with an often surprising deftness and nuance.
Source-stated
Coldewey says the agent’s images are biased toward a left-to-right arrangement.
Images “read” from left to right, as people tend to draw them, since we also read in that direction, so keep that in mind.
Escape Meridian interpretation
Escape Meridian reads Coldewey’s robot and self-driving-car examples as possible applications, not demonstrated uses of Iconary.
Robots and self-driving cars do need to know how to exceed human capacity in some cases, but it’s also massively important to be able to understand the world around them in the same way people do.
Source and contributor context
TechCrunch · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Ai2
Ai2 develops AI research tools and models, including the openly documented Olmo family.
Editorial provenance
Version 1. Publication ID: caf6c481-7e2a-333c-de96-2bc4eb8b534c.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-10-29T00:32:20.222Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: 9d634bcb8bbb37677da3122d495aa3d2e205ea6e8c2edeae4aa52125bb945550. Source text hash: 0cf80afc8117f8d953ddb8da52c260536cf991e0bd53f0d19e4126a21fef4cca.
passed · 2026-10-29T00:32:20.222Z · anthropic/claude-opus-5.5
Publication: caf6c481-7e2a-333c-de96-2bc4eb8b534c
attribution — pass: The reading attributes the article to TechCrunch and Devin Coldewey, whose byline appears in the text. It attributes the game to the Allen Institute for AI and the early-results quote to the researchers, matching the source.
Iconary, as the game’s creators at the Allen Institute for AI decided to call it to avoid lawsuits from Mattel
summary — pass: The summary explains the game, the relationship-reasoning idea and the team versus class example, and covers training on human drawings, phrases new to the model, the far-from-human caveat, the robot and car applications, and the noted bias. Terms are explained plainly, and the title and question fit the evidence.
It isn’t superhuman — in fact, it’s not even close to human.
date — pass: The stated date of February 5, 2019 matches originalPublished at day precision and the URL path. Later sidebar items, such as the Most Popular list, are neither used nor relied on. The page was edited in March 2019, but the passages the reading uses are core article body, and the limitations disclose the edit. No hindsight framing appears, and the promised paper is not presented as an outcome.
The team behind the system will be publishing a paper on it later this year.
support — pass: Each claim matches its excerpt. The robot and self-driving-car point is labeled as an interpretation and kept as a possibility, preserving the gap. The question and limitations do not add unsupported benefits. The cited passages appear to be original article body rather than later edits.
This kind of learning is also more broadly applicable in the real world.
media — pass: No media was selected, and none was supplied as a verified candidate, so there is no identity or relevance issue.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.