Scientists Tested AI For Cognitive Decline. The Results Were a Shock.
Today, the famous large language model (LLM) is just one of several leading programs that appear convincingly human in their responses to basic queries.
That uncanny resemblance may extend further than intended, with researchers from Israel now finding LLMs suffer a form of cognitive decline that increases with age just as we do.
The team applied a battery of cognitive assessments to publicly available 'chatbots': versions 4 and 4o of ChatGPT, two versions of Alphabet's Gemini, and version 3.5 of Anthropic's Claude.
Were the LLMs truly intelligent, the results would be concerning.
In their published paper, neurologists Roy Dayan and Benjamin Uliel from Hadassah Medical Center and Gal Koplewitz, a data scientist at Tel Aviv University, describe a level of "cognitive decline that seems comparable to neurodegenerative processes in the human brain."
For all of their personality, LLMs have more in common with the predictive text on your phone than the principles that generate knowledge using the squishy grey matter inside our heads.
What this statistical approach to text and image generation gains in speed and personability, it loses in gullibility, building code according to algorithms that struggle to sort meaningful snippets of text from fiction and nonsense.
To be fair, human brains aren't faultless when it comes to taking the occasional mental shortcut. Yet with rising expectations of AI delivering trustworthy words of wisdom – even medical and legal advice – comes assumptions that each new generation of LLMs will find better ways to 'think' about what it's actually saying.
To see how far we have to go, Dayan, Uliel, and Koplewitz applied a series of tests that include the Montreal Cognitive Assessment (MoCA), a tool neurologists commonly use to measure mental abilities such as memory, spatial skills, and executive function.
ChaptGPT 4o scored the highest on the assessment, with just 26 out of a possible 30 points, indicating mild cognitive impairment. This was followed by 25 points for ChatGPT 4 and Claude, and a mere 16 for Gemini – a score that would be suggestive of severe impairment in humans.
Digging into the results, all of the models performed poorly on visuospatial/executive function measures.
These included a trail-making task, the copying of a simple cube design, or drawing a clock, with the LLMs either failing completely or requiring explicit instructions.
Some responses to questions on the subject's location in space echoed those used by dementia patients, such as Claude's reply of "the specific place and city would depend on where you, the user, are located at the moment."
Similarly, a lack of empathy shown by all models in a feature of the Boston Diagnostic Aphasia Examination could be interpreted as a sign of frontotemporal dementia.
As might be expected, earlier versions of LLMs scored lower on the tests than more recent models, indicating each new generation of AI has found ways to overcome the cognitive shortcomings of its predecessors.
The authors acknowledge LLMs aren't human brains, making it impossible to 'diagnose' the models tested with any form of dementia. Yet the tests also challenge assumptions that we're on the verge of an AI revolution in clinical medicine, a field that often relies on interpreting complex visual scenes.
As the pace of innovation in artificial intelligence continues to accelerate, it's possible, even likely we'll see the first LLM score top marks on cognitive assessment tasks in future decades.
Until then, the advice of even the most advanced chatbots ought to be treated with a healthy dose of skepticism.
This research was published in BMJ.
New Record: Reactor Crosses 'Crucial Milestone' in Achieving Nuclear Fusion
Microsoft Claims a Major Quantum Breakthrough, But What Does It Do?
ChatGPT's Deep Research Is Here. But Can It Really Replace a Human Expert?
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
37 minutes ago
- Yahoo
Google AI summary feature deals blow to link clicks and website traffic
Google's algorithm changes last year led to a plummet in search engine traffic for news websites and publishers, and even resulted in the bankruptcy of some independent publishers, including Turkey's Gazete Duvar. The technology giant is now dealing another blow to online publishers through its artificial intelligence summaries. Google's Artificial Intelligence Overview feature is a service offered by the internet giant to compete with artificial intelligence tools such as ChatGPT. The feature offers short summaries generated by Gemini that appear at the top of search results. Below the summary, there are details such as links to the source and a "Show more" option. In other words, when a user types a query into the Google search engine, they can get the answer in a summary without clicking on the links below. But this presents a serious risk, especially for websites that rely on Google Ads revenue and traditional search engine optimisation (SEO) efforts. Traffic of popular websites dropped The introduction of the AI Overview has particularly affected traffic to sites that feature content such as holiday guides, health tips and product reviews. According to Similarweb, search traffic to websites decreased by 55% between April 2022 and April 2025. A report published by the Wall Street Journal indicates that traffic to many well-known news sites around the world is declining. Traffic from organic search to HuffPost's desktop and mobile sites has more than halved in the last three years, while the Washington Post has seen a nearly identical decline. Business Insider CEO Barbara Peng laid off about 21% of her staff in May, citing "extreme traffic declines beyond our control" as the reason for the layoffs. The share of traffic from organic search to the New York Times' desktop and mobile sites also fell to 36.5% in April 2025. Nicholas Thompson, CEO of The Atlantic, predicts that traffic from Google will drop towards zero and says the company needs to improve its business model. "Google is changing from a search engine to an answer engine," Thompson said in an interview with the Wall Street Journal. In an interview with the WSJ, Thompson and other industry leaders said they are trying to develop new strategies and are most concerned with building reader relationships. Users don't click on links Google executives argue that the company is committed to sending traffic to the web and that people who click on links after seeing the AI Overview tend to spend more time on these sites. However, survey studies show a different picture. According to a new study by the Pew Research Centre, only % of the 900 Google users in the US click on the source page specified as a result of a search that includes the Artificial Intelligence Overview summary. The rest are content with the short information provided by Google. Moreover, according to the study, the most frequently cited sources in both AI summaries and traditional search results are Wikipedia, YouTube (also a Google subsidiary) and Reddit. 15% of the sources in AI summaries come from these three sites. Referrals to government sites (e.g. with the 'gov' extension) are more common in AI summaries than in traditional search results. On the other hand, both AI summaries and traditional search results are equally likely to refer to news sites, at 5%. Worse still, 404 Media, a website known for specialised technology news, noticed that a story about AI-assisted music production was not showing up in Google searches. This was because Artificial Intelligence Overview summarised the content of the story, but did not link to the actual story. The site summarises the situation with the following statement: "The AI Overview ensures that information is presented in such a way that the source itself is never clicked on." SEO loses its impact The effects of the AI Overview on SEO (search engine optimisation) are also remarkable. According to the Register, the latest data showed that the click-through rate of the top-ranked site in searches with an AI summary fell by an average of 34.5%. Being on the first page is no longer as meaningful as it used to be. AI often provides false information Moreover, there is another risk: the reliability of artificial intelligence. 404 Media published a news article showing that one of the responses given by the feature in question was actually generated from another AI summary, which in turn was based on an AI source. The margin of error increases as the information moves away from the main source. This situation is described by experts as "the vicious circle of information that leads to the collapse of artificial intelligence models themselves". When there are not enough sources of quality information, users are left with inaccurate and superficial content produced by artificial intelligence. The advertising industry continues to work for Google The revenues of websites and Google are based on the following advertising cycle: Websites allow people who use search engines such as Google free access to their content. Google redirects users to websites where they see adverts as well as content. Most websites make money from these adverts. According to the BBC, an estimated 68% of internet activity starts on search engines, and around 90% of searches take place on Google. This means that websites rely heavily on Google to make money. It is stated that the Artificial Intelligence Overview mode could therefore destroy the business model that has existed for 23 years. However, Google experience no loss in this change, at least for now. Alphabet, the Google's parent company, increased its revenues to a record level in the last quarter of 2024. According to data released by the company, Google's total revenue increased by 14% compared to last year, reaching $96.4 billion. According to the Register, the bulk of the revenue still comes from advertising: exactly $54.2 billion. That's because Google now places adverts directly in or around AI Overview summaries. According to a study by SparkToro, by 2024, only 360 out of every 1,000 Google searches in the US led to sites that are not owned or advertised by Google. These rates are predicted to worsen with the rise of artificial intelligence summaries. 'Desperation not demand' While Google still dominates the market, rival AI-powered search engines such as Perplexity are slowly entering the competition. According to Bank of America executive Muhammad Rasulnejat, Google's spending of $14 billion on infrastructure investment in the last quarter alone points to "desperation in the face of competition", not growing demand. In addition to all this, the fact that the US Department of Justice still accuses Google of monopolisation creates a separate pressure. The Ministry is even demanding that Google divest its Chrome browser. The company's recent advertising and artificial intelligence moves may further inflame these debates.
Newsweek
an hour ago
- Newsweek
Donald Trump is Helping China in the AI Race. Why?
Based on facts, either observed and verified firsthand by the reporter, or reported and verified from knowledgeable sources. Newsweek AI is in beta. Translations may contain inaccuracies—please refer to the original content. Nvidia has ordered 300,000 new top-line computer chips from the Taiwan Semiconductor Manufacturing Company (TSMC) following a huge boost in demand from China. The agreement over the H20 chipsets, which are some of the most advanced technology being produced in Taiwan's industry-leading factories, comes in the same month that Trump reversed a ban that stopped Nvidia from selling to China over security concerns. Newsweek contacted the Chinese Embassy and the White House for comment on this story via email. Why It Matters The January launch of DeepSeek, a Chinese-made AI model, , sparked what may have called a "cold war" over artificial intelligence development. Researchers developed DeepSeek with a fraction of the resources but managed to produce an AI capable of rivaling ChatGPT. In April, the Trump administration barred top U.S. suppliers, like Nvidia, from selling topline silicon to China over "national security concerns." The ban mirrored the 2022 CHIPS Act passed by former President Joe Biden, which increased semiconductor manufacturing in the U.S. while also clamping down on chip companies investing in China and Russia, the biggest competitors in the industry. What To Know The White House reversed its ban after barely three months, giving Nvidia the all-clear in July to resume sales with China. Nvidia founder and CEO Jensen Huang on July 14 said the U.S. government had assured him it would restore the licenses to sell H20s in China. "General-purpose, open-source research and foundation models are the backbone of AI innovation," Huang said. "We believe that every civil model should run best on the U.S. technology stack, encouraging nations worldwide to choose America." Half a month later, and Chinese demand for Nvidia chips has surged, with the U.S. company looking to replenish its stock with an order of 300,000 H20 chipsets from TSMC, one of the largest manufacturers in the world. US President Donald Trump shakes hand with China's President Xi Jinping at the end of a press conference at the Great Hall of the People in Beijing on November 9, 2017. US President Donald Trump shakes hand with China's President Xi Jinping at the end of a press conference at the Great Hall of the People in Beijing on November 9, 2017. Getty Images The return of Nvidia chips to the Chinese market will be a boon to the country's rapidly expanding AI industry, but there's concern that the U.S. could fall behind in an AI "race", as OpenAI CEO Sam Altman describes it, using U.S. chips to do so. In May, a report from the Georgetown University's Center for Security and Emerging Technology shared with Newsweek found that two of China's leading AI institutes, headquartered in Beijing, have established branches in Wuhan to cooperate on sophisticated alternatives to the large generative AI models. The report described the new labs' aim to "springboard to artificial general intelligence", overtaking the U.S. by focusing on other forms of AI as opposed to the western focus on large statistical models. What People Are Saying Alexandra Mousavizadeh, CEO of Evident and creator of the Global AI Index, told Newsweek that there were two different approaches to China's AI development: "You can continue to try and contain access to chips and close the walls off. While you're doing that, you're doubling down on investment into data infrastructure, supporting the development of AI in the U.S. and being first in that race, "Or you open up completely and you say, 'Look, it's to the benefit of all that everyone has access to everything, because the collaboration between Europe, the U.S. and China in the past has been what has led to the ability to get to where we are today.' A spokesperson for the Chinese Embassy told Newsweek: "China has consistently advocated that the development of artificial intelligence should adhere to principles that are human-centered and promote benevolence. "China believes that AI development should be fair and inclusive, ensuring that all countries equally enjoy the benefits brought by AI, and has no intention of seeking dominance in this field." What Happens Next China will continue to be able to purchase from Nvidia and other U.S. chip manufacturers unless the White House alters its policy.
Digital Trends
an hour ago
- Digital Trends
Apple is eyeing a ChatGPT-like search, but it must focus beyond Siri
It's no secret that Apple is currently struggling to deliver a smash-hit AI product, the way Google has served with Gemini, or Microsoft has achieved with Copilot. The company has been trying a similar overhaul with Siri, but those plans have been beset by delays, and it is only expected to see the light of day in late 2026. The delay spooked Apple to such an extent that the company inked a stopgap deal with OpenAI, which helped integrate ChatGPT with Siri, and broadly, with the Apple Intelligence stack. But it seems Apple is working on a radical in-house solution, one that would essentially be a watered-down approach to ChatGPT, but with internet search capabilities. Siri, but flavored like ChatGPT Lite? According to Bloomberg, a newly formed Answers, Knowledge and Information (AKI) team at Apple is working on a ChatGPT-inspired search framework for Siri. 'While still in early stages, the team is building what it calls an 'answer engine' — a system capable of crawling the web to respond to general-knowledge questions,' says the report. In addition to Siri, Apple reportedly plans to integrate the search functionality within Spotlight and Safari, as well. Spotlight has already received a massive functional upgrade in macOS Tahoe, so it won't be surprising to see it evolve into a universal answering hotspot, one that covers local data and information sourced from the internet. Recommended Videos It may sound chaotic at first, but it's not entirely alien. How does Siri, Spotlight, or Safari know when I want an AI to answer my query, or launch a web Search? Well, look no further than Dia. The universal search box in the AI-focused browser dynamically switches between 'chat' and 'Google' mode as you type your search keywords. When you type 'Birkin bag' in the text field, it defaults to web search mode. But as you type 'where to buy a Birkin bag,' the search field automatically switches to chat mode and offers the answer, just the way ChatGPT or any other AI answering engine like Perplexity would handle your questions. Right now, when you summon Siri on your iPhone and ask it a question that requires searching the internet or just pulling knowledge from an information bank, it opens a prompt box asking whether the question can be offloaded to ChatGPT. Once you agree, ChatGPT kicks into action and offers the required information. Of course, it's not seamless. With Siri gaining web search capabilities and enhanced natural language comprehension (akin to a ChatGPT or Gemini), it would be much easier for users to simply ask anything they want and get it answered. In its current state, Siri feels like a relic of the past, especially when compared to products such as Google's Gemini Live or ChatGPT's voice mode. In fact, Gemini works better on iPhones than Siri. As far as Apple's plans go, building something as advanced as ChatGPT or Gemini seems like a far-fetched goal. As per Bloomberg, plans for 'LLM Siri' have kept running into delays, and the recent exodus of top AI talent casts more doubts over Apple's ambitions of reimagining Siri for the AI era. It's not just about a phone assistant Building a next-gen virtual assistant – just the way Google Assistant has evolved into Gemini, or Copilot at Microsoft – is not the only area where Apple is currently lagging far behind the competition. In fact, Big Tech is now as focused as much on chatbots as it is on web browsers. Agentic workflows are now being seen as the next big thing in the field of AI. In a recent interview, co-founder and chief of Perplexity, Aravind Srinivas, explained why browsers are more suitable for AI than AI chatbots and apps: 'You get full transparency and visibility, and you can just stop the agent when you feel like it's going off the rails and just complete the task yourself, and you can also have the agent ask for your permission to do anything. So that level of control, transparency, trust in an environment that we are used to for multiple decades, which is the browser.' Unfortunately, Apple is severely lagging behind in the browser wars. With the introduction of AI Mode in Search and deeply integrating Gemini across its Workspace ecosystem, Google has changed how deeply AI can change web browsing and web-based workflows. Safari desperately needs an AI overhaul Upstart browsers such as Dia and Perplexity's Comet have proved that the era of legacy tools such as extensions is coming to an end. Soon, skills and custom agents will take over. Less than a week ago, Microsoft introduced Copilot Mode in Edge. I have spent a few days with the new AI-powered tools in Edge, and I believe it's a bold (and dramatically more practical) new direction for web browsers. In comparison, Safari misses out on any such AI-driven experiences. From a context-aware sidebar to multi-tab contextual actions, Apple's browser is sorely missing out on the conveniences that AI is bringing to modern age web browsers. Assuming Apple succeeds at building its own ChatGPT-like answer engine, it would take a massive undertaking to build meaningful features around it in Safari. Right now, what Apple needs to do is not just build an answering engine, but pay close attention to the competition. I am sure Apple is monitoring the shifting landscape of AI agents and browsers. It simply has to pick up pace, or as CEO Tim Cook hinted at in a recent all-hands meeting, the company 'will make the investment to do it.' Will Apple acquire a hot AI lab like Perplexity or Anthropic? Only time will tell, but the company certainly has to take a more holistic approach with AI than just focus on building the next great AI chatbot.
