# MIT Technology Review asks what babies can teach AI

[Canonical reading](https://escapemeridian.com/readings/mit-technology-review-asks-what-babies-can-teach-ai-873e0e1b)

Published February 6, 2024, MIT Technology Review’s The Algorithm describes research inspired by how babies learn, and argues that studying their learning could help build AI with more common sense. New York University researchers trained a neural network on sights and sounds recorded by one child, Sam, who wore a camera off and on from six months old until a little after his second birthday. The researchers used this material to teach the system to match words to objects. The article also describes Google DeepMind work on “intuitive physics”: a model learned how objects move from videos and showed “surprise” when an object moved differently than expected. These examples are presented as steps toward AI that learns more like people, not as systems that already have human common sense. The article’s broader point is that AI excels at narrow tasks but remains brittle in a messy world; studying how babies learn could help address that gap. Whether these approaches will produce AI that reasons or operates more reliably in unfamiliar situations remains unknown.

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

MIT Technology Review · Article

Original source date: Feb 6, 2024 (2024-02-06T10:24:15Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-21T16:48:12.279Z. First observed: 2026-11-02T02:03:50.830Z. Retrieved: 2026-11-02T02:07:22.955Z.

- [What babies can teach AI](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Limits

Text as retrieved on November 2, 2026; originally published February 6, 2024. The article describes research examples and a possible direction for AI; it does not establish that the approaches produce human-like common sense or reliable performance in unfamiliar situations. The Sam study is based on one child's recorded experience.


## Open question

What abilities, if any, will systems trained on children’s experiences gain when they encounter unfamiliar situations?


## Source-stated

MIT Technology Review’s The Algorithm says New York University researchers trained a neural network using sights and sounds recorded by one child, Sam, to match words to objects.

> The material he collected allowed the researchers to teach a neural network to match words to the objects they represent

- [Supporting source](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Source-stated

The article says Google DeepMind researchers trained a model on videos to learn how objects move, and that it showed “surprise” when an object moved differently from the way it had learned objects move.

> The researchers at Google DeepMind managed to get their AI system, too, to show “surprise” when an object moved differently from the way it had learned that objects move.

- [Supporting source](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Source-stated

The article argues that studying how babies learn could help AI develop common sense and operate more effectively in a messy world; it does not say these abilities have been achieved.

> Studying how babies learn could help us unlock those abilities.

- [Supporting source](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Escape Meridian interpretation

Escape Meridian reads the examples as research steps toward the article’s broader goal, not evidence that current AI has human common sense.

> Even the most powerful AI systems we have today lack those abilities.

- [Supporting source](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Source and contributor context

MIT Technology Review · 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.


## Neural network

A system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.

- [Definition reference](https://developers.google.com/machine-learning/glossary#neural-network)

## Google DeepMind

Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.

- [Explore this context](/entities/google-deepmind)
- [Reference](https://deepmind.google/blog/)
- [Reference](https://deepmind.google/science/alphafold/)

## Editorial provenance

Version 1. Publication ID: 275715c5-da92-ab23-fd2c-f35e50e733ed.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-21T16:43:56.378Z. 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: 2a1c8ef7dbca2f50fb41e3551880ee1ab2cf18bfe5d0c893408afc1c1b2daf9c. Source text hash: 8a52353e2b541ff541ee8183bef1801e0cc8a0451c410293e3be6bbea713fa6c.


## Revision and source-check history

### Version 1 — 2026-11-21T16:48:12.279Z

Published February 6, 2024, MIT Technology Review’s The Algorithm describes research inspired by how babies learn, and argues that studying their learning could help build AI with more common sense. New York University researchers trained a neural network on sights and sounds recorded by one child, Sam, who wore a camera off and on from six months old until a little after his second birthday. The researchers used this material to teach the system to match words to objects. The article also describes Google DeepMind work on “intuitive physics”: a model learned how objects move from videos and showed “surprise” when an object moved differently than expected. These examples are presented as steps toward AI that learns more like people, not as systems that already have human common sense. The article’s broader point is that AI excels at narrow tasks but remains brittle in a messy world; studying how babies learn could help address that gap. Whether these approaches will produce AI that reasons or operates more reliably in unfamiliar situations remains unknown.

## Original source

MIT Technology Review · Article

Original source date: Feb 6, 2024 (2024-02-06T10:24:15Z).

Date basis: Date from the publisher’s own records, through its API.

Added to Escape Meridian: 2026-11-21T16:48:12.279Z. First observed: 2026-11-02T02:03:50.830Z. Retrieved: 2026-11-02T02:07:22.955Z.

- [What babies can teach AI](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Limits

Text as retrieved on November 2, 2026; originally published February 6, 2024. The article describes research examples and a possible direction for AI; it does not establish that the approaches produce human-like common sense or reliable performance in unfamiliar situations. The Sam study is based on one child's recorded experience.


## Open question

What abilities, if any, will systems trained on children’s experiences gain when they encounter unfamiliar situations?


## Source-stated

MIT Technology Review’s The Algorithm says New York University researchers trained a neural network using sights and sounds recorded by one child, Sam, to match words to objects.

> The material he collected allowed the researchers to teach a neural network to match words to the objects they represent

- [Supporting source](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Source-stated

The article says Google DeepMind researchers trained a model on videos to learn how objects move, and that it showed “surprise” when an object moved differently from the way it had learned objects move.

> The researchers at Google DeepMind managed to get their AI system, too, to show “surprise” when an object moved differently from the way it had learned that objects move.

- [Supporting source](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Source-stated

The article argues that studying how babies learn could help AI develop common sense and operate more effectively in a messy world; it does not say these abilities have been achieved.

> Studying how babies learn could help us unlock those abilities.

- [Supporting source](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Escape Meridian interpretation

Escape Meridian reads the examples as research steps toward the article’s broader goal, not evidence that current AI has human common sense.

> Even the most powerful AI systems we have today lack those abilities.

- [Supporting source](https://www.technologyreview.com/2024/02/06/1087793/what-babies-can-teach-ai/)

## Source and contributor context

MIT Technology Review · 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.


## Neural network

A system of connected simple calculations, loosely inspired by the brain, that learns patterns from data.

- [Definition reference](https://developers.google.com/machine-learning/glossary#neural-network)

## Google DeepMind

Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.

- [Explore this context](/entities/google-deepmind)
- [Reference](https://deepmind.google/blog/)
- [Reference](https://deepmind.google/science/alphafold/)

## Editorial provenance

Version 1. Publication ID: 275715c5-da92-ab23-fd2c-f35e50e733ed.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-21T16:43:56.378Z. 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: 2a1c8ef7dbca2f50fb41e3551880ee1ab2cf18bfe5d0c893408afc1c1b2daf9c. Source text hash: 8a52353e2b541ff541ee8183bef1801e0cc8a0451c410293e3be6bbea713fa6c.


### passed — 2026-11-21T16:43:56.378Z — anthropic/claude-opus-5.5

Publication: 275715c5-da92-ab23-fd2c-f35e50e733ed

attribution — pass: The reading attributes the piece to MIT Technology Review's The Algorithm newsletter. The source text says it originally appeared there. It attributes the NYU and Google DeepMind research to those researchers as the article describes them. The newsletter issue comments on its items in its own words, so it is one piece, not a listing.

> This story originally appeared in The Algorithm, our weekly newsletter on AI.

summary — pass: The reading leads with the article's argument and explains both examples in plain language: the single-child camera study and DeepMind's video model showing 'surprise'. It explains 'intuitive physics' through the surprise mechanism, and 'neural network' is defined on the page. It keeps the article's thesis that AI is brittle and could gain common sense from studying babies. Omitting the LeCun passage and the unrelated newsletter items does not distort the meaning. The question follows from the evidence and is not leading.

> But compared with the human brain—the most powerful machine we know of—these systems are brittle.

date — pass: February 6, 2024 is the UTC calendar day of originalPublished (2024-02-06T10:24:15Z) and matches the URL path. Sidebar items describing later events, such as the 'Ten years after AlphaGo' blurb, are probably later page additions. The reading neither uses nor relies on them. It frames the article's points as an argued possible direction, not as outcomes. This is a publisher page, so no archive disclosure is needed. dateModified falls one second after publication.

support — pass: Each claim matches its excerpt: the NYU word-object matching, DeepMind's 'surprise' result and the article's conditional 'could help' argument. The interpretation claim is labeled as Escape Meridian's and is supported by the article's statement that current AI lacks these abilities. The limitations correctly note that the study used one child and that no reliability in unfamiliar settings is shown. Nothing invents benefits or costs.

> Studying how babies learn could help us unlock those abilities.

media — pass: No media is selected and no verified media candidates were supplied, so there is no identity or relevance issue.

## Recorded references

Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter.

No public references recorded here yet.

[Reference data (JSON)](/readings/873e0e1b6dd2933acd25d82c9da9f70573b5a90f7c0f67d6e8e55237bc44f882/references.json)

## Corrections and challenges

[Send a correction or an opposing source](https://escapemeridian.com/corrections/)

[JSON record](https://escapemeridian.com/readings/mit-technology-review-asks-what-babies-can-teach-ai-873e0e1b.json)
