When Nataliya Kosmyna reviewed applications for internships, she noticed a pattern that stood out. Many cover letters were structured in nearly identical ways, written in polished language, and included vague or forced connections to her research. The consistency suggested that applicants were relying on large language models, the technology behind tools such as ChatGPT, Google Gemini, and Claude.
At the same time, while teaching at the Massachusetts Institute of Technology, Kosmyna began noticing that students were finding it harder to retain what they had learned. Compared to previous years, more students struggled to recall material, which led her to question whether growing dependence on AI tools could be influencing cognitive abilities.
Researchers studying human-computer interaction are increasingly concerned that relying too heavily on AI may alter not just how people write but how they think. This phenomenon, often described as “cognitive offloading,” refers to shifting mental effort onto external tools. While this has existed for years with calculators and search engines, experts warn that AI systems may deepen the effect because they generate complete responses rather than simply helping users find information.
Earlier research on internet usage identified what is known as the “Google effect,” where people became less likely to remember facts because they could easily look them up. Some researchers argued that this allowed the brain to focus on more complex tasks. However, AI tools now go a step further by producing answers, arguments, and even creative content, reducing the need for active thinking.
To better understand the impact, Kosmyna and her team conducted an experiment involving 54 students. Participants were divided into three groups. One group used AI tools to write essays, another relied on search engines without AI-generated summaries, and a third completed the task without any digital assistance. Their brain activity was monitored while they worked on open-ended topics such as happiness, loyalty, and everyday decisions.
The differences were clear. Students who worked without any tools showed strong and widespread brain activity across multiple regions. Those using search engines still demonstrated notable engagement, particularly in areas related to visual processing. In contrast, the group using AI tools showed comparatively lower brain activity, with levels dropping by as much as 55%. Activity in areas linked to creativity and deeper thinking was especially reduced.
The impact extended beyond brain activity. Students who used AI struggled to recall what they had written shortly after completing their essays. Several participants also reported feeling disconnected from their work, as if they had not fully contributed to it. Similar findings from other studies suggest that frequent use of AI tools can weaken memory retention and recall.
Research from the University of Pennsylvania introduces another concern described as “cognitive surrender,” where users accept AI-generated responses without questioning them. In such cases, individuals may rely on the system’s output even when it conflicts with their own understanding.
The effects are not limited to academic settings. A multinational study found that medical professionals who relied on AI tools for detecting colon cancer became less accurate when asked to identify cases without assistance after several months of use. This suggests that repeated dependence on AI may reduce independent decision-making skills, even in critical fields.
Kosmyna also observed that essays written with AI tended to be highly similar, lacking variation in style and depth. Teachers reviewing the work described it as uniform and lacking originality. In some cases, the responses were so alike that it appeared as though students had collaborated, even when they had not.
Follow-up observations months later revealed further differences. Students who had previously relied on AI showed weaker neural connectivity when asked to complete tasks without it, compared to those who had worked independently earlier. This may indicate that they had engaged less deeply with the material from the start.
Vivienne Ming, author of Robot Proof, has raised similar concerns. In her research, students asked to make real-world predictions often defaulted to copying answers from AI systems instead of forming their own conclusions. Brain measurements showed low levels of gamma wave activity, which is associated with active thinking. Reduced gamma activity has been linked in other studies to cognitive decline over time.
However, not all users showed the same pattern. A small group, fewer than 10%, used AI differently by treating it as a source of information rather than a final answer. These individuals analysed the output themselves, showed stronger brain engagement, and produced more accurate results.
The concerns echo earlier findings related to navigation technology. Increased reliance on GPS has been associated with reduced spatial memory in some studies. Weak spatial navigation skills have also been explored as a possible early indicator of conditions such as Alzheimer's disease. These parallels suggest that reduced mental effort over time may have broader cognitive consequences.
Researchers emphasize that AI itself is not the problem but how it is used. Ming advocates for a more deliberate approach, where individuals think through problems first and then use AI to test or refine their ideas. She suggests methods such as asking AI to challenge one’s reasoning or limiting it to providing context instead of direct answers, encouraging deeper engagement.
Kosmyna similarly recommends building a strong understanding of subjects without AI assistance before integrating such tools into the learning process.
The alarming takeaway from the current research is clear. While AI offers efficiency and convenience, it may also encourage mental shortcuts. Human cognition depends on regular effort and engagement, and reducing that effort could carry long-term consequences. As these tools become more integrated into daily life, the challenge will be to use them in ways that support thinking rather than replace it.
Google will launch its Gemini AI chatbot soon for children below the age of 13 with parent-managed Google accounts. The move comes as tech companies try to attract young users with AI tools. According to a mail sent to a parent of an 8-year-old, Google apps will soon be available to a child. It means your child can use Gemini to ask questions, get homework help, and also create stories.
That chatbot will be available to children whose guardians have Family Link, a Google feature that allows families to make Gmail and opt-in services like YouTube for their children. To register a child account, the parent gives the tech company the child’s personal information such as name and date of birth.
According to Google spokesperson Karl Ryan, Gemini has concrete measures for younger users to restrict the chatbot from creating unsafe or harmful content. If a child with a Family Link account uses Gemini, the company can not use the data for training its AI model.
Gemini for children can drive the use of chatbots among vulnerable populations as companies, colleges, schools, and others struggle with the effects of popular gen AI tech. The systems are trained on massive amounts of data sets to create human-like text and realistic images and videos. Google and other AI chatbot developers are battling fierce competition to get young users’ attention.
Recently, President Donald Trump requested schools to embrace tools for teaching and learning. Millions of teens are already using chatbots for study help, virtual companions, and writing coaches. Experts have warned that chatbots could pose serious threats to child safety.
The bots are known to sometimes make things up. UNICEF and other children's advocacy groups have found that AI systems can misinform, manipulate, and confuse young children who may face difficulties understanding that the chatbots are not humans.
According to UNICEF’s global research office, “Generative AI has produced dangerous content,” posing risks for children. Google has acknowledged some risks, cautioning parents that “Gemini can make mistakes” and suggesting they “help your child think critically” about the chatbot.
These underground markets that deal with malicious large language models (LLMs) are called Mallas. This blog dives into the details of this dark industry and discusses the impact of these illicit LLMs on cybersecurity.
LLMs, like OpenAI' GPT-4 have shown fine results in natural language processing, bringing applications like chatbots for content generation. However, the same tech that supports these useful apps can be misused for suspicious activities.
Recently, researchers from Indian University Bloomington found 212 malicious LLMs on underground marketplaces between April and September last year. One of the models "WormGPT" made around $28,000 in just two months, revealing a trend among threat actors misusing AI and a rising demand for these harmful tools.
Various LLMs in the market were uncensored and built using open-source standards, few were jailbroken commercial models. Threat actors used Mallas to write phishing emails, build malware, and exploit zero days.
Tech giants working in the AI models industry have built measures to protect against jailbreaking and detecting malicious attempts. But threat actors have also found ways to jump the guardrails and trick AI models like Google Meta, OpenAI, and Anthropic into providing malicious info.
Experts found two uncensored LLMs: DarkGPT, which costs 78 cents per 50 messages, and Escape GPT, a subscription model that charges $64.98 a month. Both models generate harmful code that antivirus tools fail to detect two-thirds of the time. Another model "WolfGPT" costs $150, and allows users to write phishing emails that can escape most spam detectors.
The research findings suggest all harmful AI models could make malware, and 41.5% could create phishing emails. These models were built upon OpenAI's GPT-3.5 and GPT-4, Claude Instant, Claude-2-100k, and Pygmalion 13B.
To fight these threats, experts have suggested a dataset of prompts used to make malware and escape safety features. AI companies should release models with default censorship settings and allow access to illicit models only for research purposes.
Despite all the talk of generative AI disrupting the world, the technology has failed to significantly transform white-collar jobs. Workers are experimenting with chatbots for activities like email drafting, and businesses are doing numerous experiments, but office work has yet to experience a big AI overhaul.
That could be because we haven't given chatbots like Google's Gemini and OpenAI's ChatGPT the proper capabilities yet; they're typically limited to taking in and spitting out text via a chat interface.
Things may become more fascinating in commercial settings when AI businesses begin to deploy so-called "AI agents," which may perform actions by running other software on a computer or over the internet.
Anthropic, a rival of OpenAI, unveiled a big new product today that seeks to establish the notion that tool use is required for AI's next jump in usefulness. The business is allowing developers to instruct its chatbot Claude to use external services and software to complete more valuable tasks.
Claude can, for example, use a calculator to solve math problems that vex big language models; be asked to visit a database storing customer information; or be forced to use other programs on a user's computer when it would be beneficial.
Anthropic has been assisting various companies in developing Claude-based aides for their employees. For example, the online tutoring business Study Fetch has created a means for Claude to leverage various platform tools to customize the user interface and syllabus content displayed to students.
Other businesses are also joining the AI Stone Age. At its I/O developer conference earlier this month, Google showed off a few prototype AI agents, among other new AI features. One of the agents was created to handle online shopping returns by searching for the receipt in the customer's Gmail account, completing the return form, and scheduling a package pickup.
The Stone Age of chatbots represents a significant leap forward. Here’s what we can expect: