AIin astrophotography: friend or foe?

It makes processing a breeze, but creates very convincing fakes too. Pete Lawrence looks at how AI is changing astro imaging

- Pete Lawrence

Artificial intelligence or AI is a term heard everywhere these days. Its application is exciting, confusing and very often misunderstood, so let's start by defining exactly what AI is and what it does.

AI is essentially a computer-based tool that has the ability to learn and apply results to data processing. Unlike a conventional computer program, AI can - when given the right guidance - make more considered choices, emulating how a human would approach a task.

But the 'intelligence' in AI doesn't mean that AI systems are self-aware or intelligent in the way we are. An AI system can't operate without direction and this is provided in the form of models or neural networks trained using a dataset relevant to the subject. These models are typically produced by processes known as machine learning or deep learning. The source information is often real-world data pre-categorised by humans, but it can also be virtual data generated by a computer applying known processes to a set of base information.

Terms such as 'artificial intelligence', 'machine learning' and 'deep learning' can sound provocative but are merely references to what are essentially computer tools and systems. What's important to understand is that Al's abilities depend on how well-trained its models are. Poor or inappropriate model data will simply give poor results; the phrase 'garbage in, garbage out' is highly fitting to Al. Despite what Hollywood implies, Al is currently still dependent on humans to feed it the correct information in order for it to work efficiently.

Let's now look at some examples of how Al is being applied in astrophotography. On the internet, the inducement of social media 'Likes' has inevitably led to a rise in fake astronomical images. Many early attempts were poor manually edited examples, but skill levels have now improved, making some fakes hard to call out unless you're a skilled expert.

When algorithms make art