This Research Does Not Exist: AI Generated Science Papers that Read Like the Real Thing

Getting your Trinity Audio player ready...

Of increasing concern in this age of misinformation is the emergence of deepfakes. Using the power of machine learning and artificial intelligence, an image or video can be altered to seamlessly replace one person’s face with another, or to change the words that they are speaking.

AI is not limited to faking visual and audio content, however. Natural language processing (NLP) is an application of machine learning that concerns the interactions between human language and computers. After analysing large quantities of language data, NLP can be used to generate entirely new, coherent sentences. You can test this for yourself using sites like Talk to Transformer, which runs an NLP model called GPT-2.

NLP is already used in text auto-completion and speech recognition software, but as this technology improves, we must also consider the associated dangers. To demonstrate how easily scientific research might be faked using AI, This Research Does Not Exist used NLP to generate fake research paper abstracts. Here are our picks for the three most convincing/entertaining fake research papers.

Caffeine: a Potent Supplement for Muscle Growth

To measure the impact of caffeine consumption on muscle growth, we determined the amino acid content of different muscle groups from the masseter, gastrocnemius, triceps brachii, biceps brachii, and quadriceps femoris in 20 men after 3 and 5 months of treatment with either 400 mg/d of caffeine or a placebo. Our results indicated that protein synthesis rates were significantly greater after supplementation with caffeine than after placebo, with a time effect between treatments and between placebo and caffeine groups. These findings confirmed that caffeine is a potent supplement for muscle building.

What do Bipolar Disorder and Prostate Cancer Have In Common?

For this study, we selected a large double-blind, randomised, placebo-controlled trial of oxcarbazepine versus placebo in recurrent prostate cancer patients. Patients were randomly assigned to receive four infusions of oxcarbazepine (50 mg/day) or placebo for a median of 3 months and then followed for a median of 6 months. Patients were followed for disease progression (stage) and overall survival (OS) as well as survival at 12 months and overall survival. Oxcarbazepine treatment significantly slowed disease progression and improved survival by 38% relative to the placebo group.

High dietary fat enhances TGF-β signalling and intestinal permeability.

In this study of the effects of dietary fat on intestinal permeability, rats were fed diets varying in their fat content. The compositions of the fatty acids of these diets were as follows: MUFAs (0.44% — 0.59%, non-pooled), total fatty acids (0.54% — 0.79%, non-pooled).

Increased fat intake was associated with increased intestinal permeability as measured by magnetic resonance spectroscopy (MRS). In addition, mucosal mRNA and protein levels were measured using the reverse transcriptase–polymerase chain reaction. Our results showed that the expression of Akt and GSK3β was significantly increased in medium and high fat-fed rats (4.78-fold; P < 0.05). These findings indicate that fatty acids directly activate TGF-β signaling, which is an important step in the pathogenesis of type 2 diabetes and its association with obesity.

And there you have it: three research abstracts written by a machine. Would you have realised that these studies weren’t real? See what kind of research you can generate using the various NLP models out there, and share them with us at info@gowinglife.co.uk.