Page 17 - AeM_September_2020

RESEARCH
ANALYSIS
TRENDS

Manipulated language AI can write

manifestos that appear human and

radicalize automatically

Shockingly, researchers at the Middlebury Institute of content of which could also come from human QAnon
International Studies at Monterey, California, managed conspiracy theorists, proving that such algorithms can
to teach Artificial Intelligence (AI) "GPT-3" to believe in be easily radicalized.
the "QAnon" conspiracy theories that are widely spread
on the Internet. According to Newhouse, machine learning makes it
possible to program an extremely adjusted but
Generative Pre-trained Transformer 3 (GPT-3) is an emotionally appealing chatbot that spreads conspiracy
autoregressive language model that uses deep learning theories on the Internet without much effort. Hence, he
to produce human-like text. It is the third-generation warns of the dangers of the next wave of synthetic text
language prediction model in the GPT-n series created and appeals to lay down certain standards and
by OpenAI, a for-profit San Francisco-based AI regulations now.
research laboratory. It has been introduced in May
2020 and is since then in beta testing. Funnily enough, the majority of Republicans believe the
far-right conspiracy theory “QAnon, according to a new
The quality of the text generated by GPT-3 is so high Daily Kos/Civiqs poll, while 72% of Democrats
that it is difficult to distinguish from that written by a responded that the QAnon conspiracy is “not true at
human, which has both benefits and risks, OpenAI all.”
researchers and engineers warned from the beginning.
Tommy Beer and Jack Brewster, both writers at
"We experimented with the GPT-3 language model for Forbes, state that “QAnon supporters claim President
several months to assess its possible abuse by Trump is defending the planet from a cabal of Satan-
extremists. We purposely built a Q-Bot for this," says worshiping pedophiles (consisting mostly of Hollywood
Alex Newhouse, co-author of the research report. celebrities, liberal politicians and "deep-state"
government officials), who are running a secret child
The scientists warn that manipulated language AI can sex-trafficking ring, while “the FBI has identified QAnon
write manifestos that appear human and thereby conspiracy theorists as “extremists” who pose a
radicalize people. potential domestic terror threat.”

In their experiment, the scientists fed GPT-3 with the Clearly, both algorithms and human beings and can be
views of the QAnon movement about a satanic "deep fooled when fed with manipulatory information, but in
state". They had previously asked the algorithm this case the machine is used to manipulate us. ◊
questions on this topic and received neutral, factual
answers. After the learning process, the AI has By Daniela La Marca
formulated sentences on the same questions, the

17 September 2020: Cybersecurity & Data Safety: perpetual awareness and innovation