A Robot Wrote My Research

Artificial intelligence (AI) is everywhere. In nan world of academia, it’s having a large impact.

When utilized appropriately, AI tin greatly summation productivity and workflow.

It tin beryllium an plus for finding nan correct lit articles and filing them nicely (something each world struggles with) aliases for checking nan last publication is free from grammatical nightmares.

AI is besides peculiarly good suited to nan task of organising and analysing information for useful information.

MAKING LIFE EASIER

Science is, by nature, information heavy.

But pinch an AI companion, researchers are now capable to analyse large datasets faster than ever, spot trends that would different person gone unnoticed and attraction their attraction connected what’s really important.

A awesome illustration is Google DeepMind’s AlphaFold.

Before nan advent of AI, it would return respective years to lick a macromolecule building – captious accusation if you want to cognize what a macromolecule does and really it does it. AlphaFold tin lick it successful a matter of minutes.

Caption: A folded macromolecule structure.
Credit: via Google DeepMind

But that’s not what AI is astir wide known for. AI is peculiarly bully astatine writing.

In contrast, scientists are not known for cleanable (or peculiarly engaging) prose.

Thanks to generative ample connection models (LLMs) for illustration ChatGPT, moreover they tin person their investigation publication and understood by nan masses.

Basically, AI tin do a lot. But what happens if AI does it all?

WAIT, AI WROTE THIS?

Well, not this article. (Or immoderate Particle article.) But researchers person been experimenting pinch AI to spot if it really could constitute an world article from commencement to finish.

There are galore steps successful nan creation of a caller investigation paper: find a spread successful nan literature, inquire a question, behaviour experiments, analyse nan information and make a conclusion.

In 2023, a brace of information scientists put their custom-built AI to nan test.

It wrote each its ain code, analysed a ample dataset, established noteworthy trends and put it into a logical discussion. It moreover created its ain reappraisal strategy to amended refine nan article.

All this successful nether an hour.

In 2024, Sakana AI Labs introduced nan world to The AI Scientist, a programme capable to independently make a ‘novel’ article astatine a costs nary quality investigation squad could compete with. 

Caption: Robots tin do a lot, but quality scientists are still needed successful nan lab.
Credit: abdelaziz@771/Adobe Stock. This image has been generated utilizing AI.

While this mightiness activity good for penning a reappraisal article, location are still galore areas of investigation that stay beyond nan scope of AI – only a quality knows really to pipette their life distant successful nan lab. 

HALLUCINATING ROBOT LIARS

As awesome arsenic these caller models whitethorn be, they’re still prone to hallucinations. Not successful an “I spot dormant people” way, but successful a “I don’t cognize nan reply truthful I’m going to make 1 up” benignant of way.

In different words, a mirage is erstwhile AI generates a misleading aliases mendacious connection because it doesn’t person capable training.

A mirage tin go a recurring problem arsenic it is incorporated into early training algorithms to beryllium replicated forever.

Scientific sleuths hunt for these hallucinations aliases ‘digital fossils’ to thief uncover fraudulent AI written papers and person them retracted from nan technological literature.

Robot liars are not nan only problem, AI-derived results tin beryllium fraught pinch bias.

Because existent AI devices are truthful tuneable, scientists must guarantee they’re not training their models to conscionable cherry pick nan information they want.

Clearly, this benignant of generative AI doesn’t clasp up to rigorous technological standards.

Until it tin flooded nan issues of hallucinations and information biases, automated AI isn’t going to triumph a Nobel Prize immoderate clip soon.

Credit: firstwefeast/giphy.com

INSTITUTIONS UNDER THREAT

AI whitethorn not beryllium bully astatine producing groundbreaking caller ideas, but investigation that builds connected existent knowledge is still needed, moreover if it’s only pointing retired nan obvious.

These AI written articles airs a superior threat to nan peer-reviewed publication process.

It’s a strategy already being undermined by predatory journals and academic insubstantial mills.

An AI exemplary for illustration The AI Scientist could quickly make articles, perchance flooding nan lit pinch subpar ‘research’. This would make it moreover much difficult to spot important results successful nan haystack of AI junk.

When discussing nan morals of their tool, moreover Sakana AI Labs admits successful their open entree article that “there is important imaginable for misuse”.

For world journals and universities, beardown AI guidelines are being enforced to effort to flooded nan ethical pitfalls.

Turnitin precocious released AI discovery package to spot plagiarism amongst students (ironically, an AI strategy built to sniff retired AI).

SO WILL SCIENTISTS GET ROBO-COPPED?

Science is astir nan non-conventional ways of looking astatine things, novelty and expanding quality knowledge.

Despite nan powerfulness of AI, quality scientists still clasp nan precocious hand.