Science & technology | Digital poisons

It doesn’t take much to make machine-learning algorithms go awry

The rise of large-language models could make the problem worse

2CYCJE0 A visitor looks at Ceci n'est pas une pomme (This is not an Apple) from 1964 by Rene Magritte at Christie's in London February 1, 2013. The artwork is expected to sell for £1M-£1.5M (US $1.6M-2.3M) when it is auctioned at Christie's The Art of the Surreal Evening Sale on February 6, 2013 REUTERS/Suzanne Plunkett (BRITAIN - Tags: ENTERTAINMENT)
This is not an appleImage: Alamy

The algorithms that underlie modern artificial-intelligence (AI) systems need lots of data on which to train. Much of that data comes from the open web which, unfortunately, makes the AIs susceptible to a type of cyber-attack known as “data poisoning”. This means modifying or adding extraneous information to a training data set so that an algorithm learns harmful or undesirable behaviours. Like a real poison, poisoned data could go unnoticed until after the damage has been done.

This article appeared in the Science & technology section of the print edition under the headline “Digital poisons”

From the April 8th 2023 edition

Discover stories from this section and more in the list of contents

Explore the edition
A string of SpaceX StarLink satellites passes over an old stone house

Satellites are polluting the stratosphere

And forthcoming mega-constellations will exacerbate the problem

An illustration of hands and arms going through a nest of lines.

AI models are dreaming up the materials of the future

Better batteries, cleaner bioplastics and more powerful semiconductors await


Colossal woolly mouse

Mice have been genetically engineered to look like mammoths

They are small and tuskless, but extremely fluffy


Is posh moisturiser worth the money?

Don’t break the bank

How artificial intelligence can make board games better

It can iron out glitches in the rules before they go on the market

The skyrocketing demand for minerals will require new technologies

Flexible drills, distributed power systems and, of course, artificial intelligence