Adam Mosseri, head of Instagram, stated that content generated by artificial intelligence should not be shown to users who do not show interest in it. The executive argues that the platform's algorithm must be adapted to respect this preference, although he dismisses the possibility of simply blocking all AI content on the social network.
With the proliferation of AI-created materials in recent years, there is currently no simple mechanism for users to hide this type of content. Mosseri explained in an interview with Lenny Rachitsky's podcast that Meta's goal is to improve the ability to identify AI-generated videos and images, allowing recommendation systems themselves to understand the user's level of interest.
Mosseri had previously admitted publicly that the Instagram feed had stopped reflecting users' real interests due to the growing power of the algorithm. This observation arose after the expansion of the 'Your Algorithm' feature, which shows users the types of posts the technology deems relevant and offers the option to decrease the visibility of certain topics, although it does not completely prevent the delivery of specific content.
According to Mosseri, detecting AI content is becoming progressively more complex as language models evolve. For this reason, he proposed that Instagram could adopt less definitive responses regarding the origin of media. Instead of rigidly categorizing whether something was or was not made by AI, the network could indicate confidence levels, such as 'probably yes,' 'we are not sure,' or 'definitely no.'
Currently, the platform has a flagging system for the use of generative tools, as do competitors like YouTube and TikTok. These posts may display an 'AI Information' notification in the feed, detailing that the content 'may have been created by AI.' The pop-up clarifies that this flag generally applies to content entirely generated by AI, and not just modified with AI assistance.
During the podcast, the executive presented two main ideas. The first is to create a dedicated space to segregate content entirely generated by AI—including spam, fraudulent montages, and other 'AI junk'—from posts that used creative editing or content remixing tools.
Meta has already experimented with similar concepts in the Meta AI app, presenting feeds of images generated by the company's models and even a Reels format with synthetic videos. The second proposal, advocated by Mosseri since December of last year, is the implementation of certification for authentic content. The argument is that it would be simpler to attest to the provenance of a real photograph than to prove the use of AI.
Large AI corporations developing image and video models, such as OpenAI, Google, Microsoft, and Meta itself, are using digital certificates, such as C2PA, and invisible watermarks in content, such as SynthID. These marks are generated by the tools themselves to aid identification. In addition to the declaration made by the user at the time of posting, Instagram can read this information to alert about possible AI use. However, this practice is not yet standardized and these identifiers are not impossible to bypass or remove.
Artificial intelligence (AI) tools have demonstrated great potential to increase productivity in various areas. However, recent research suggests that indiscriminate use of these technologies can lead to the weakening of individuals' cognitive abilities.
Studies analyzed indicate that excessive dependence on these systems can diminish users' aptitude for problem-solving without technological support. This warning arose from investigations conducted with students and professionals, where participants who used AI showed immediate improvements in performance but faced difficulties when performing the same tasks autonomously.
Experiments with workers and students confirmed that AI can optimize results when applied to activities aligned with their competencies. In a study involving hundreds of consultants from the Boston Consulting Group, researchers from the Wharton School noted an increase in the number of completed tasks and a decrease in time spent by those who had access to the tool.
This survey, published in 2026 in the journal Organization Science, revealed that employees assisted by AI produced higher quality work in functions where the technology was most competent. The most significant progress was observed among professionals who had a more modest initial performance.
A similar finding was found in a study conducted by Grace Liu of Carnegie Mellon University, focusing on mathematical problem-solving. By comparing students with and without access to AI, the study identified superior performance among those who could use the resource during exercises.
Despite the instant gains, researchers detected adverse effects after the removal of technological assistance. Individuals accustomed to AI began to show lower performance than those who had never used the resource and demonstrated less resilience in the face of difficulties.
Additionally, another study investigated how excessive trust in responses generated by AI systems affects decision-making. Steven Shaw and Gideon Nave evaluated over 1,300 participants and identified a phenomenon called 'cognitive surrender,' characterized by the user abandoning their own evaluation in favor of the conclusion provided by the machine.
Researchers argue that AI can operate as a third cognitive mechanism, complementing the traditional methods of fast thinking and detailed analysis described by Daniel Kahneman. The problem arises when this tool ceases to be a complement to human reasoning and begins to replace it entirely.
Experts emphasize that the crucial skill in the age of AI will be discerning which tasks should remain under human control and which can be delegated to automated systems. Collaboration is most effective when the user understands the limits of the technology and can judge its outputs.
An analysis from 2024, published in the journal Nature Human Behavior and based on 106 AI experiments, showed that joint performance between humans and machines is optimized when each party acts in its area of greatest advantage. However, when the system demonstrates superiority, the human difficulty in knowing whether to trust or contest the tool can compromise the final result.
The interviewed experts highlighted that processes such as initial idea conception, text writing, and knowledge generation require active human participation. For them, AI is most valuable in phases of review, questioning arguments, and refining existing work.
The concern extends to the educational environment. The cited studies indicated that students may learn less by using AI solely to speed up school assignments. On the other hand, when the tool is used to gain clarification, formulate questions, and assimilate concepts, the damage to the learning process is minimized.
Judy Hanwen Shen and Alex Tamkin, from Anthropic, reported in their research with developers learning a new programming library conceptual difficulties, code reading, and debugging when AI was used as a shortcut to obtain ready-made solutions. The researchers' recommendation is to reconfigure AI as a tool for deepening understanding, stimulating questioning and expanding analytical capacity, rather than simply eliminating the mental effort required to learn.
