GPT-Driven Spam Bots Challenge Online Platforms
GPT-driven spam bots have become a significant issue on platforms like Twitter and Telegram, targeting unsolicited promotions.
These sophisticated AI programs can analyze and replicate the context of a post, making their interference appear more organic and harder to identify.
A new type of disturbance has emerged in the form of GPT-driven spam bots. These sophisticated AI programs have turned a fresh page in the spam playbook, frequently targeting posts on platforms such as Twitter and Telegram with unsolicited promotions.
These GPT-spam bots can analyse and replicate the context of a post, which makes their interference seem more natural and harder to spot than the plain spam of the past. Traditional protective measures are challenged by this because many of them are ineffective. Currently, these bots can only be identified by their quick response times, which enables manual intervention and removal.
The constant barrage of spam wears on creators and administrators. Many channel operators on platforms like Telegram have stated their desire for a specialised service that can recognise and delete these sophisticated spam comments in response to this growing need. These operators, who envision themselves as “moderators as a service,” are prepared to invest in a solution and have expressed a willingness to pay $20–30 per month or to use a usage-based billing model based on the quantity of posts or messages that are being watched.
The difficulty does not, however, end here. There is an impending wave of GPT-spammers that are anticipated to become even more skilled as technology advances, possibly using strategies like response delays or using various AI personalities that interact with one another. Differentiating between human users and bots in such circumstances becomes a challenging task.
Even tech giants are grappling with the issue. OpenAI took a step towards resolving this with the development of a text detector designed to identify AI-generated content. Unfortunately, their efforts faced a setback as the project was shelved due to the detector’s low accuracy, as reported by TechCrunch in July 2023.
Platform administrators are not the only ones concerned by the increase in GPT-powered spam bots. The challenge of separating authentic content from AI-generated submissions now faces even social media managers and startups. The circumstance highlights an urgent need and presents a chance for new initiatives and projects that can create efficient solutions to counter the advanced spamming methods of the modern era.
Advancements in Language Models and Implications for Online Misinformation
The practicality and near-human conversational aptitude of GPT have been noted by users. However, the same capabilities that have won it admiration also bring forth concerns about its potential misuse.
Given the AI’s proficiency in mimicking human-like responses, there are apprehensions surrounding its deployment for malicious intents. Experts across academia, cybersecurity, and AI sectors emphasize the potential use of GPT by ill-intentioned individuals to disseminate propaganda or foster unrest on digital platforms.
Historically, propagating misinformation demanded significant human intervention. The introduction of refined language processing systems can magnify the scale and reach of influence operations targeting social media, resulting in more tailored, and therefore, potentially more convincing campaigns.
Social media platforms have, in previous instances, witnessed coordinated efforts to spread misinformation. For instance, during the lead-up to the 2016 US election, the Internet Research Agency, based in St Petersburg, launched an expansive campaign. Their objective, as deduced by the Senate Intelligence Committee in 2019, was to impact the perception of the electorate towards the presidential nominees.
The January report highlighted that the emergence of AI-driven language models could augment the dissemination of misleading content. The content could not only increase in volume but also improve in persuasive quality, making it a challenge for average internet users to discern its authenticity.
Josh Goldstein, affiliated with Georgetown’s Center for Security and Emerging Technology and a contributor to the study, mentioned the ability of generative models to churn out large volumes of unique content. Such capability could allow individuals with malicious intent to circulate varied narratives without resorting to repetitive content.
Despite the efforts of platforms like Telegram, Twitter and Facebook to counter fake accounts, the evolution of language models threatens to saturate these platforms with more deceptive profiles. Vincent Conitzer, a computer science professor at Carnegie Mellon University, noted that advanced technologies, such as ChatGPT, could significantly boost the proliferation of counterfeit profiles, further blurring the lines between genuine users and automated accounts.
Recent studies, including Mr. Goldstein’s paper and a report by security firm WithSecure Intelligence, have highlighted the proficiency of generative language models in crafting deceptive news articles. These false narratives, when circulated on social platforms, could influence public opinion, especially during crucial electoral periods.
The rise of misinformation facilitated by advanced AI systems like Chat-GPT prompts the question: Should online platforms take more proactive measures? While some argue that platforms should rigorously flag dubious content, challenges persist. Luís A Nunes Amaral, associated with the Northwestern Institute on Complex Systems, commented on the platforms’ struggles, citing both the cost of monitoring each post and the inadvertent engagement boost such divisive posts bring.
Read more about AI:
Any data, text, or other content on this page is provided as general market information and not as investment advice. Past performance is not necessarily an indicator of future results.
The Trust Project is a worldwide group of news organizations working to establish transparency standards.