AI is getting smarter every day —with thought processes already eerily similar to humans: study
Artificial intelligence doesn't just look and act human — it supposedly thinks like us as well.
Chinese researchers found the first-ever evidence that AI models like ChatGPT process information similarly to the human mind, detailing the dystopian-seeming discovery in the journal 'Nature Machine Intelligence.'
Advertisement
'This provides compelling evidence that the object representations in LLMs (large language models), although not identical to human ones, share fundamental similarities that reflect key aspects of human conceptual knowledge,' wrote the team behind the study, which was a collaboration between the Chinese Academy of Sciences and the South China University of Technology, the Independent reported.
3 The AI models displayed remarkable powers of categorization.
Malambo C/peopleimages.com – stock.adobe.com
The team reportedly wanted to see if LLM models can 'develop humanlike object representations from linguistic and multimodal data (data in different forms such as text, audio, etc).'
Advertisement
3 Prior to the study, many tech experts simply assumed that language models like ChatGPT mimicked human responses through pattern recognition.
AlexPhotoStock – stock.adobe.com
To discover whether the AI 'bot process' mirrors our cognition, researchers had OpenAI's ChatGPT-3.5 and Google's Gemini Pro Vision perform a series of 'odd-one-out' trials, in which they were given three items and tasked with selecting the one that doesn't fit, the South China Morning Post reported.
3 Researchers remain unsure if AI understands the significance or emotional value of cats and other entities on a human level.
AntonKhrupinArt – stock.adobe.com
Remarkably, the AI created 66 conceptual dimensions to sort the objects.
Advertisement
After comparing this cybernetic object sorting to human analysis of the same objects, they found striking similarities between the models' 'perception' and human cognition — particularly when it came to language grouping.
From this, researchers deduced that our psychological doppelgangers 'develop human-like conceptual representations of objects.'
'Further analysis showed strong alignment between model embeddings and neural activity patterns' in the region of the brain associated with memory and scene recognition.
Advertisement
Researchers noted that the language-based LLMs were a bit lacking with regard to categorizing visual aspects such as shape or spatial properties.
Meanwhile, research has shown that AI struggles with tasks that require deeper levels of human cognition, such as analogical thinking — drawing comparisons between different things to conclude — while it's unclear if they comprehend certain objects' significance or emotional value.
'Current AI can distinguish between cat and dog pictures, but the essential difference between this 'recognition' and human 'understanding' of cats and dogs remains to be revealed,' said He Huiguang, a professor at the Chinese Academy of Sciences' (CAS) Institute of Automation.
Nonetheless, the scientists hope that these findings will allow them to develop 'more human-like artificial cognitive systems' that can collaborate better with their flesh-and-blood brethren.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Business Insider
29 minutes ago
- Business Insider
Inside the battle over Microsoft's access to OpenAI's technology
The future of the most important partnership in technology depends on Microsoft 's access to the artificial intelligence powering ChatGPT. When OpenAI first demoed a breakthrough feature in May 2024, allowing users to talk to its AI just like a person, it looked as if the company did so in lock-step with its partner and investor, Microsoft. Soon after OpenAI demoed this voice capability for its new GPT-4o model, Microsoft CEO Satya Nadella included it in a keynote speech at the company's Build developer conference. Behind the scenes, Microsoft had little knowledge of the feature until days before the demo, people involved in the matter told Business Insider. While the agreement between the two companies gives Microsoft access to OpenAI's technology, exactly what OpenAI has to share — and when — is sometimes a gray area. In this case, Microsoft had access to frequent updates of the core model at the time, but not the voice technology OpenAI built on top of it. Microsoft found out about the demo, and pressured OpenAI executives, including then-technology boss Mira Murati, to get access to the code so Microsoft could do its own announcement, the people said. The company did not want to appear flat-footed to investors, to whom the company has to justify its $13 billion investment in OpenAI, they said. The example illustrates the ongoing complexity of Microsoft's relationship with OpenAI, and why access to the AI startup's technology is a core issue as the companies renegotiate their agreement. OpenAI needs Microsoft's blessing for a corporate restructuring. To get that, OpenAI may need to convince Microsoft to change or give up some pretty sweet terms: Microsoft has access to much of OpenAI's technology, exclusive rights to sell it on Azure, first right of refusal to provide computing resources, and a revenue-sharing agreement worth billions of dollars. Lots of thorny details about points of contention in the negotiations have made recent headlines — a looming, existential clause OpenAI could activate to cut off Microsoft's access, a "nuclear option" reportedly considered by OpenAI executives to accuse Microsoft of anticompetitive behavior, and a report that Microsoft was prepared to walk away from the renegotiations. In response to those reports, Microsoft and OpenAI released a joint statement saying, "Talks are ongoing and we are optimistic we will continue to build together for years to come." People close to Microsoft's side of the negotiating talks tell BI the software giant is unlikely to walk away because it is deeply reliant on OpenAI's intellectual property, and the negotiations are an opportunity to improve and expand its access to this technology. It's all about IP Microsoft has significantly benefited from its arrangement to access the rights to OpenAI's intellectual property, both by selling it to customers through the Azure OpenAI service and creating its own products using OpenAI's technology, like its AI assistant Copilot. There are limits, however, to what the companies consider "IP." For example, Microsoft gets access to important aspects of OpenAI's models, like model weights that help determine AI outputs and inference codes that instruct the models on how to use data, the people said. Some things are excluded from what the companies consider IP, like product and user interface information, they said, and the point at which OpenAI must share any technology is up to interpretation. One person with knowledge of OpenAI's operations said the company doesn't have to share what it's developing until it's finished, which can be subjective. "You can make sure you share something with Microsoft as late as possible, so they can still simultaneously announce, but make it really difficult to build the same product on top of it," the person said. Plus, having access to the IP isn't the same as knowing how to use it. This has been harder than expected, several people told BI. OpenAI has grown frustrated with Microsoft's request to spell out the technology to its employees, people with knowledge of both companies said. Sometimes, Microsoft doesn't know what questions to ask of OpenAI. Microsoft formed a new AI organization two years ago and hired Mustafa Suleyman, the former Inflection CEO who cofounded the AI pioneer DeepMind, to run it. The hiring was meant as a hedge against OpenAI, after Nadella received pressure from Microsoft's board to diversify following instability at its partner startup and CEO Sam Altman's ouster. However, little has come of that in terms of replacing Microsoft's need for its partnership with OpenAI, the people said. Suleyman has completely rebuilt Microsoft's Copilot app. That effort has yet to achieve much growth for Copilot. His team is focused on building smaller models and has seen success with post-training existing models for new purposes, one of the people said. Overall, Microsoft isn't trying to build frontier models like those that would compete with GPT, and is instead putting resources toward OpenAI, they said. Microsoft is less worried about AGI, antitrust, Windsurf, or SoftBank Other points of contention that have made recent headlines are of less concern to Microsoft, the people with knowledge of Microsoft's position said. Included in the agreement is a clause that would allow OpenAI to declare what's called artificial general intelligence or "AGI," and cut off Microsoft's access to OpenAI's IP and profits. OpenAI defines AGI as "a highly autonomous system that outperforms humans at most economically valuable works." While OpenAI could declare AGI, the concept is so open to interpretation that Microsoft could sue and easily tie the company up in a legal battle for years, the people said. There's another version of the clause, "sufficient AGI," that OpenAI could declare when it builds a technology capable of achieving a certain level of profits, but Microsoft has to sign off on that. Another apparent sticking point in the negotiations has centered on OpenAI's desire to acquire AI coding assistant startup Windsurf. Under the current agreement, that would give Microsoft access to Windsurf's technology. Microsoft has said it would agree to the acquisition, but Windsurf's CEO doesn't want the company's technology to be shared with Microsoft, one of the people said. While GitHub Copilot faces increasing competition from other AI coding assistants, access to Windsurf's technology is not a big desire for Microsoft, and the company might consider a carve-out of Windsurf IP in a new deal, the person said. OpenAI executives have reportedly considered accusing Microsoft of anticompetitive behavior. Microsoft is largely unconcerned about this, two of the people said. The existing deal was investigated by antitrust regulators, including the European Union and the United Kingdom's Competition and Markets Authority. OpenAI's desire to restructure is in part motivated by a deadline from investor SoftBank that risks a percentage of funding if OpenAI can't close such a deal by the end of the year. OpenAI has said it struggled to fundraise due to its peculiar structure. SoftBank has a reputation for taking risks with its funding, and its CEO is keen on OpenAI, so some observers doubt SoftBank will follow through on its deadline to revoke funding if a new deal isn't reached.
Bloomberg
an hour ago
- Bloomberg
Senators Came to Their Senses on AI Regulation Ban
Some sense has prevailed in the Senate — a 99-1 vote against a provision in its huge tax and spending bill that would have banned state-level artificial-intelligence laws for the next 10 years. It's been just 944 dizzying days since ChatGPT was launched into the world — imagine what might have happened over the next 3,653. A last-gasp effort to amend the bill, which included reducing 10 years to five, also failed. The new wording would have been more onerous than the original, decimating existing state laws on facial recognition and data privacy. New laws will need to tackle AI-triggered issues on discrimination, recruitment and mental health. The matter is simply too urgent to be left only in Washington's hands. Senators rightly saw through the moratorium as doing the bidding of big tech companies that want free rein to do as they please in the insatiable race to build and sell AI.
Yahoo
2 hours ago
- Yahoo
China's Huawei must face US criminal charges, judge rules
By Jonathan Stempel NEW YORK (Reuters) -A U.S. judge on Tuesday rejected Huawei Technologies' bid to dismiss most of a federal indictment accusing the Chinese telecommunications company of trying to steal technology secrets from U.S. rivals, and misleading banks about its work in Iran. In a 52-page decision, U.S. District Judge Ann Donnelly in Brooklyn found sufficient allegations in the 16-count indictment that Huawei engaged in racketeering to expand its brand, stole trade secrets from six companies, and committed bank fraud. The Iran accusations stemmed from Huawei's alleged control of Skycom, a Hong Kong company that did business in that country. Donnelly said prosecutors satisfactorily alleged Skycom "operated as Huawei's Iranian subsidiary and ultimately stood to benefit, in a roundabout way," from more than $100 million of money transfers through the U.S. financial system. Huawei has pleaded not guilty and had sought to dismiss 13 of the 16 counts, calling itself "a prosecutorial target in search of a crime." A trial is scheduled for May 4, 2026, and could last several months. Neither Huawei nor its lawyers immediately responded to requests for comment. A spokesperson for Interim U.S. Attorney Joseph Nocella in Brooklyn declined to comment. The criminal case began during U.S. President Donald Trump's first term in 2018, the same year the Department of Justice launched its China Initiative to address Beijing's alleged theft of intellectual property. Huawei Chief Financial Officer Meng Wanzhou, whose father founded the company, had been a defendant, and was detained in Canada for nearly three years before being allowed to return to China. Charges against her were dismissed in 2022. In 2022, President Joe Biden's administration scrapped the China Initiative, after critics said it amounted to racial profiling and caused fear that chilled scientific research. Based in Shenzhen, Huawei operates in more than 170 countries and has about 208,000 employees. The U.S. government has restricted Huawei's access to American technology since 2019, citing national security concerns. Huawei denies it is a threat. The case is U.S. v. Huawei Technologies Co et al, U.S. District Court, Eastern District of New York, No. 18-cr-00457.
