To the line
Directions · AUTOMATED EXPERIMENT

AI in scientific research

In one viewAI already predicts protein and material structures, proposes candidates, and can guide robotic systems through narrow series of experiments while choosing the next step.

This is not yet an independent scientist, but a chain of specialised models, instruments and algorithms. The main frontier is a complete loop: hypothesis → experiment → verification → new hypothesis.

Status
TypeDirections
Marker
Events in dossier9
Development chronology

Researched

2021-07

AlphaFold 2

About this eventPredicting a protein's 3D structure from its amino-acid sequence approached experimental accuracy for a broad class of problems.

AlphaFold 2 did not replace structural biology, but it sharply shortened the route from a protein sequence to a plausible spatial hypothesis. Researchers gained a tool for prioritising experiments and mechanisms, not automatic proof of function or therapeutic effect.

Source: Nature
2022

200 million protein structures

About this eventAlphaFold DB published predictions for nearly every catalogued protein known to science. It accelerates research but does not replace experimental validation.

AlphaFold DB turned structure prediction from a specialised compute project into an accessible lookup layer for biologists. The database sharply narrows the hypothesis space, but a predicted structure does not establish function, interaction or therapeutic effect; those conclusions still require experiments.

Source: Google DeepMind
2023

2.2M predicted crystals

About this eventGNoME proposed 2.2 million new crystal structures, including 380,000 potentially stable candidates. These are computational candidates, not finished materials.

Source: Google DeepMind
2023

36 materials in 17 days

About this eventThe robotic A-Lab planned syntheses, ran experiments and revised recipes, producing 36 of 57 target compounds.

Source: Nature
2024

AlphaFold 3: molecular interactions

About this eventThe model began predicting joint structures of proteins, DNA, RNA, small molecules and ions — a step from protein shape toward interaction mechanisms.

Source: Nature

In progress

сейчас

The prediction-to-experiment gap

About this eventA model can propose thousands of candidates, but making a sample, measuring its properties and reproducing the result remain slow and expensive.

сейчас

Narrow closed loops

About this eventSelf-driving systems already optimise specific reactions and materials, but stay inside a predefined domain and still rely on people to set the objective.

Planned

в планах

A shared language for laboratories

About this eventStandard data formats and instrument interfaces are needed so models can transfer experience between setups and results can be reproduced.

Distant horizons

впереди

Verifiable machine discovery

About this eventAI formulates a novel hypothesis, chooses the decisive experiment and produces a result independently confirmed by other laboratories.

Sources and research

Primary material behind this dossier: papers, lab publications and official reports.

Enabled by