A look at how AI is being used in science, from probing the evolution of galaxies and calculating quantum wave functions to discovering new chemical compounds
Dan Falk / Quanta Magazine : Tweets: @royalsociety , @quantamagazine , @techticee , and @stevenstrogatz Tweets: @royalsociety : “The Square Kilometer Array, a radio telescope slated to switch on in the mid-2020s, will generate about as much data traffic each year as the entire internet.” Discover how #AI is being used to extend our understanding and improve our research http://www.quantamagazine.org/ ... Quanta / @quantamagazine : At the very least, new AI-based systems are “hardworking assistants” that can comb through data for hours on end without getting bored or complaining about the working conditions. http://www.quantamagazine.org/ ... @techticee : It's fascinating isn't it? You should look into @goodfellow_ian's paper on pinning adversarial neural nets against each other; one becoming better at detecting falsefied faces and one getting better creating them! The math is rather fascinating! This is the core of my research. http://twitter.com/... Steven Strogatz / @stevenstrogatz : Today I learned that “generative adversarial networks” can generate realistic-looking faces http://www.quantamagazine.org/ ... http://twitter.com/...
Context & Ripple Effects
Written when open-sourced frameworks were already letting hobbyists run machine learning outside big labs, this Quanta survey extends the same democratization argument into research science: Dan Falk catalogs AI as a "hardworking assistant" for galaxy evolution, quantum wave functions, and compound discovery, with the Square Kilometer Array's internet-scale data traffic as the forcing case for automation.
The piece reads differently now that its own six-year report card exists: a 2025 follow-up finds AI genuinely useful for experiment design and pattern-spotting in physics, yet still short of producing new physics discoveries — the exact boundary this 2019 article was probing.
First-order effects
- Researchers in data-saturated fields gain tireless combing power over datasets too large for human review, with the Square Kilometer Array's projected data load making algorithmic triage a prerequisite rather than an option.
- The named practitioners — astronomers, quantum physicists, chemists — shift effort from manual analysis toward curating and validating model outputs.
Second-order effects
- Instrument and experiment planners begin designing facilities like the SKA around machine-learning pipelines from day one, since the data volume forecloses any fallback to human-scale review.
- Competition for ML-literate scientists intensifies between academia and industry, as the same skill set commands value in both.
Third-order effects
- If the pattern holds, scientific institutions reorganize around AI-assisted workflows — but the later finding that AI has not yet yielded new physics discoveries marks the structural open question: whether these systems stay amplifiers of human hypothesis-making or eventually become generators of it.
The trend: AI in science is moving from tireless data-combing assistant toward potential discovery engine, with the gap between the two now explicitly measured rather than assumed.