News analysis | Beware of "AI dependence" and how to maintain independent thinking in the era of human-machine symbiosis

Recently, researchers from Stanford University in the United States planned the world's first online open conference with artificial intelligence (AI) as the core author and paper reviewer. The organizer stated that the purpose of this conference is to create a "relatively safe testing ground" to explore whether and how AI can independently generate novel scientific insights, hypotheses, and methodologies.

Nowadays, AI has become a great helper in our work and life, but it also provides a "cognitive shortcut" that allows people to get the answers they want without having to think hard. At the beginning, we only searched for information and translated text, but gradually began to use AI to assist in writing emails, work summaries, research papers, and even making important decisions... Over reliance on AI and using it for "thinking outsourcing" may gradually lead us to lose the dominance of our thinking.

In the era of human-machine symbiosis, how can humans maintain independent thinking?

AI 'thinking outsourcing' hides hidden concerns

As discussed in the above conference, AI applications such as big models and intelligent agents have widely empowered human production and life in multiple fields, including scientific research, promoting academic research and improving the quality and efficiency of production and life. However, there are also hidden concerns under the technological dividends they bring. Some new studies suggest that excessive reliance on AI models may trigger "AI fog" and weaken work motivation.

A study by the Massachusetts Institute of Technology in the United States shows that long-term use of AI can lead to cognitive decline. After conducting EEG scans on participants, researchers found that compared to those who did not use AI tools, those who used AI language models for a long time had a decrease in the number of neural connections in their brains, which affected language and behavior levels. Similar findings from research by Microsoft and Carnegie Mellon University suggest that generative AI can weaken critical thinking skills, lead to excessive dependence, and potentially weaken people's ability to solve problems independently.

Researchers believe that developers should support users in enhancing their awareness of critical thinking when designing generative AI tools, such as prompting scenarios that need to be validated. Especially when designing high-risk tasks such as legal documents and medical advice, verification reminders (such as "please check the regulatory terms referenced by AI") should be mandatory to indicate potential risks and break the cognitive bias of users towards AI being "omniscient and omnipotent".

The research published by the team led by Wu Suqing from the School of Management at Zhejiang University in the British journal Scientific Reports this year also shows that the application of AI brings about a "hollowing out" of cognitive needs. When AI completes the "brain burning" part of the work, humans are left with only mechanical execution, and the work becomes a "passive fill in the blank", which may lead to psychological alienation and loss of work motivation. This requires companies, government departments, and others to slow down when deploying AI. They can first pilot it on a small scale and promote it in a controllable manner, while observing its impact on employee motivation and creativity.

Why do people trust AI so easily

The reason why people are increasingly relying on AI stems from the leap in AI's specialization and ease of interaction in recent years: the degree of specialization in its generated content makes many people regard it as an expert, while the conversational interaction makes it seem like a knowledgeable and talkative "friend", which makes it easier for people to form psychological dependence and trust.

Nowadays, the convenience of using AI big models out of the box has further lowered the threshold for use. Due to the lack of basic AI literacy education, users have insufficient awareness of the technical deficiencies of AI itself and are more likely to "trust" it. The UK High Court required lawyers to take urgent action in June this year to prevent the misuse of AI, as several false case citations that may have been generated by AI were submitted to the court. And this year, the Children's Chronic Disease Report released by the "Make America Healthy Again" committee led by the US Department of Health and Human Services also had significant citation errors due to the use of generative AI content.

James Pearson, a researcher at the University of Amsterdam in the Netherlands, said that AI has significantly driven the scientific research process, such as being used to map protein structures, develop new antibiotics, and accelerate the progress of literature review and other processes. But when the "illusion" generated by AI appears in paper review or research, it is very concerning, and AI may also carry biases contained in its training data, thereby exacerbating inequality issues such as race and gender.

We have outsourced our judgment while using AI, no longer verifying the answers provided by AI, and no longer reflecting on the process of using AI. Unconditionally accepting any information given by AI will lead to a decrease in subjective initiative, and the ability to read, think, and express deeply will also deteriorate. ”Chen Qifan, a science fiction writer and deputy director of the Science Fiction Literature Committee of the Chinese Writers Association, said in an interview with Xinhua News Agency.

Establishing an AI buffer zone for thinking

In the future, AI will definitely be more deeply embedded in our work and life. In this new era of human-machine symbiosis, coexisting with AI requires greater intelligence. Greg Schiff, CEO of Section 4, an American AI education company, predicts that in the next 10 years, knowledge-based workers using AI will be divided into two groups: "AI passengers" and "AI drivers".

AI passengers are delighted to entrust their cognitive work to AI management, and may gain recognition in the short term due to improved work efficiency. But with the iteration of AI capabilities, these people will eventually be replaced by AI. The 'AI driver' insists on leading the operation of AI, treating the content generated by AI as a preliminary draft, rigorously verifying the results, and sometimes even turning off AI independent thinking.

We need to learn in different learning and work scenarios to distinguish which scenarios are suitable for using AI, which scenarios require human assistance, and judge and distinguish the results given by AI. ”Chen Qifan holds the same view. In his view, people should take the initiative in their own minds and cognition, consciously establish an "AI buffer zone", that is, when facing new problems, they should first provide answers from their own perspective, and then seek the deepening and expansion of AI.

We can even engage in 'adversarial generation', which means that when AI provides answers that are very consistent with their own views, critical thinking should be used to question whether I am influenced by AI or whether AI is becoming increasingly convergent under my continuous training? Can we break free from our habitual thinking and make different choices, providing answers that AI did not anticipate. ”He said.