The world might keep in mind 2022 as the 12 months of generative synthetic intelligence (AI): the 12 months that enormous language fashions (LLMs), akin to OpenAI’s GPT-3, and text-to-image fashions, akin to Secure Diffusion, marked a sea change within the potential for social media manipulation. LLMs which were optimized for dialog (akin to ChatGPT) can generate naturalistic, human-sounding textual content content material at scale, whereas open-source text-to-image fashions can generate photorealistic pictures of something (actual or imagined) and may accomplish that at scale. Utilizing present expertise, U.S. adversaries may construct digital infrastructure to fabricate reasonable however inauthentic (pretend) content material that would gasoline equally reasonable however inauthentic on-line human personae: accounts on Twitter, Reddit, or Fb that appear actual however are artificial constructs, fueled by generative AI and advancing narratives that serve the pursuits of these governments.
On this Perspective, the authors argue that the emergence of ubiquitous, highly effective generative AI poses a possible nationwide safety menace when it comes to the chance of misuse by U.S. adversaries (particularly, for social media manipulation) that the U.S. authorities and broader expertise and coverage neighborhood ought to proactively deal with now. Though the authors deal with China and its Folks’s Liberation Military as an illustrative instance of the potential menace, a wide range of actors may use generative AI for social media manipulation, together with technically subtle nonstate actors (home in addition to overseas). The capabilities and threats mentioned on this Perspective are possible additionally related to different actors, akin to Russia and Iran, which have already engaged in social media manipulation.
Funding for this analysis was made doable by the unbiased analysis and improvement provisions of RAND’s contracts for the operation of its U.S. Division of Protection federally funded analysis and improvement facilities. This analysis was performed inside the Worldwide Safety and Protection Coverage Program of the RAND RAND Nationwide Safety Analysis Division (NSRD).
This publication is a part of the RAND Company Perspective collection. RAND Views current skilled insights on well timed coverage points. All RAND Views endure peer evaluation to make sure excessive requirements for high quality and objectivity.
This doc and trademark(s) contained herein are protected by regulation. This illustration of RAND mental property is supplied for noncommercial use solely. Unauthorized posting of this publication on-line is prohibited; linking on to this product web page is inspired. Permission is required from RAND to breed, or reuse in one other type, any of its analysis paperwork for industrial functions. For data on reprint and reuse permissions, please go to www.rand.org/pubs/permissions.
The RAND Company is a nonprofit establishment that helps enhance coverage and decisionmaking by analysis and evaluation. RAND’s publications don’t essentially replicate the opinions of its analysis shoppers and sponsors.
#U.S #Adversaries #Generative #Social #Media #Manipulation