Some workers already let AI do the thinking for them, Microsoft researchers find
(2025/02/11)
- Reference: 1739233869
- News link: https://www.theregister.co.uk/2025/02/11/microsoft_study_ai_critical_thinking/
- Source link:
Some knowledge workers risk becoming over-reliant on generative AI and their problem-solving skills may decline as a result, according to a study penned by researchers from Microsoft Research and Carnegie Mellon University.
In a [1]paper titled “The Impact of Generative AI on Critical Thinking”, the seven researchers report and analyze a survey in which they asked 319 knowledge workers who use generative AI at least weekly how and if they apply critical thinking when using tools such as Copilot and ChatGPT.
The research found that workers who are confident tackling a task are more likely to apply critical thinking to the output of a generative AI service, while those less comfortable with a task often assume generative AI produced adequate answers and don’t bother to think about what the brainbox delivered.
Confidence in AI is associated with reduced critical thinking effort, while self-confidence is associated with increased critical thinking
The researchers suggest their findings point to a need for a re-think about the design of enterprise AI tools.
"Confidence in AI is associated with reduced critical thinking effort, while self-confidence is associated with increased critical thinking," the paper states, adding "This duality indicates that design strategies should focus on balancing these aspects."
[2]
AI tools, the team suggests, should incorporate mechanisms to support long-term skill development and encourage users to engage in reflective thinking when interacting with AI-generated outputs.
[3]
[4]
"This aligns with the goals of explainable AI," the researchers said, referring to the practice of having AI outline how it delivered its output. The call for AI to show its workings is good news for the latest [5]chain-of-thought AI models from DeepSeek and OpenAI - but merely explaining AI's reasoning isn't enough.
Good AI tools should foster critical thinking through proactive design strategies that encourage user reflection and provide assistance when necessary, the researchers wrote.
[6]
That might seem like criticism of current AI tools, but the paper doesn’t go there. The authors also stop short of recommending that knowledge workers reduce AI use to avoid "cognitive offload" and the potential "deterioration of cognitive faculties that ought to be preserved."
The authors didn’t respond to questions from The Register .
[7]Using AI in science can add to reproducibility woes, say boffins
[8]If you use AI to teach you how to code, remember you still need to think for yourself
[9]University students recruit AI to write essays for them. Now what?
[10]Meta warns bit flips, other hardware faults cause AI errors
But please don't stop using our enterprise AI products
The paper concludes that we should adapt to an AI-infused world by applying critical thinking to verify AI outputs and how they can be used in daily work. Which may be what one would expected given six of seven authors work at the company that sells [11]Copilot .
Yes, the researchers admit, knowledge workers should be taught to "maintain foundational skills in information gathering and problem-solving [to] avoid becoming over-reliant on AI," just not too much. Those working with systems like ChatGPT, Copilot, and other generative AI tools should be trained "on developing skills in information verification, response integration and task stewardship."
This isn't the only study to conclude that more reliance on AI is having a negative impact on critical thinking skills, but [12]previous work has concluded that we need to preserve our current critical thinking skills - not offload them to AI and change the way we engage those crucial faculties to simply validate and integrate AI output.
[13]
"When using genAI tools, the effort invested in critical thinking shifts from information gathering to information verification; from problem-solving to AI response integration; and from task execution to task stewardship," the authors conclude.
The paper will be presented at the [14]2025 Conference on Human Factors in Computing Systems which starts in late April. ®
Get our [15]Tech Resources
[1] https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://www.theregister.com/2025/01/26/deepseek_r1_ai_cot/
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2024/05/29/using_ai_in_science_can/
[8] https://www.theregister.com/2024/01/27/ai_coding_automatic/
[9] https://www.theregister.com/2022/12/27/university_ai_essays_students/
[10] https://www.theregister.com/2024/06/21/hardware_faults_create_inferencing_errors/
[11] https://www.theregister.com/2025/02/07/microsoft_365_price_rises/
[12] https://www.mdpi.com/2075-4698/15/1/6
[13] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[14] https://chi2025.acm.org/
[15] https://whitepapers.theregister.com/
In a [1]paper titled “The Impact of Generative AI on Critical Thinking”, the seven researchers report and analyze a survey in which they asked 319 knowledge workers who use generative AI at least weekly how and if they apply critical thinking when using tools such as Copilot and ChatGPT.
The research found that workers who are confident tackling a task are more likely to apply critical thinking to the output of a generative AI service, while those less comfortable with a task often assume generative AI produced adequate answers and don’t bother to think about what the brainbox delivered.
Confidence in AI is associated with reduced critical thinking effort, while self-confidence is associated with increased critical thinking
The researchers suggest their findings point to a need for a re-think about the design of enterprise AI tools.
"Confidence in AI is associated with reduced critical thinking effort, while self-confidence is associated with increased critical thinking," the paper states, adding "This duality indicates that design strategies should focus on balancing these aspects."
[2]
AI tools, the team suggests, should incorporate mechanisms to support long-term skill development and encourage users to engage in reflective thinking when interacting with AI-generated outputs.
[3]
[4]
"This aligns with the goals of explainable AI," the researchers said, referring to the practice of having AI outline how it delivered its output. The call for AI to show its workings is good news for the latest [5]chain-of-thought AI models from DeepSeek and OpenAI - but merely explaining AI's reasoning isn't enough.
Good AI tools should foster critical thinking through proactive design strategies that encourage user reflection and provide assistance when necessary, the researchers wrote.
[6]
That might seem like criticism of current AI tools, but the paper doesn’t go there. The authors also stop short of recommending that knowledge workers reduce AI use to avoid "cognitive offload" and the potential "deterioration of cognitive faculties that ought to be preserved."
The authors didn’t respond to questions from The Register .
[7]Using AI in science can add to reproducibility woes, say boffins
[8]If you use AI to teach you how to code, remember you still need to think for yourself
[9]University students recruit AI to write essays for them. Now what?
[10]Meta warns bit flips, other hardware faults cause AI errors
But please don't stop using our enterprise AI products
The paper concludes that we should adapt to an AI-infused world by applying critical thinking to verify AI outputs and how they can be used in daily work. Which may be what one would expected given six of seven authors work at the company that sells [11]Copilot .
Yes, the researchers admit, knowledge workers should be taught to "maintain foundational skills in information gathering and problem-solving [to] avoid becoming over-reliant on AI," just not too much. Those working with systems like ChatGPT, Copilot, and other generative AI tools should be trained "on developing skills in information verification, response integration and task stewardship."
This isn't the only study to conclude that more reliance on AI is having a negative impact on critical thinking skills, but [12]previous work has concluded that we need to preserve our current critical thinking skills - not offload them to AI and change the way we engage those crucial faculties to simply validate and integrate AI output.
[13]
"When using genAI tools, the effort invested in critical thinking shifts from information gathering to information verification; from problem-solving to AI response integration; and from task execution to task stewardship," the authors conclude.
The paper will be presented at the [14]2025 Conference on Human Factors in Computing Systems which starts in late April. ®
Get our [15]Tech Resources
[1] https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://www.theregister.com/2025/01/26/deepseek_r1_ai_cot/
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2024/05/29/using_ai_in_science_can/
[8] https://www.theregister.com/2024/01/27/ai_coding_automatic/
[9] https://www.theregister.com/2022/12/27/university_ai_essays_students/
[10] https://www.theregister.com/2024/06/21/hardware_faults_create_inferencing_errors/
[11] https://www.theregister.com/2025/02/07/microsoft_365_price_rises/
[12] https://www.mdpi.com/2075-4698/15/1/6
[13] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z6rZcSqfLBQIO550D__mgAAAAQg&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[14] https://chi2025.acm.org/
[15] https://whitepapers.theregister.com/
Re: ChatGPT, comment on this article in a pithy and thought provoking way.
cyberdemon
I suck.
There, I did the bot's job for it. Hail to Humanity
Here is how far I trust MS information.
IGotOut
I came into work today and that stupid advertising, I mean lock screen had the following useful information.
Top right: "The end of the year is upon us once again and the holidays are coming up"
Centre: "Prioritise your well being this holiday season with Microsoft Search ...."
Checks date. Yup 10th of February.
Still makes a change from all the Copilot crap...
So AI just amplifies Dunning-Kruger
The Man Who Fell To Earth
Gee, what a shock.
The less competent rely more on AI & while the more competent rely less on AI.
Yet those who are less competent tend to over rank themselves, while those of high competence tend to under rank themselves.
Looks solid to me
HuBo
They cite Bainbridge's 1983 " Ironies of automation " right in the intro, after Socrates, Trithemius, and teachers of arithmetic, which is quite sensible imho. They then go on to state:
" by mechanising routine tasks [...] you deprive the user of [opportunities to] strengthen their cognitive musculature, leaving them atrophied and unprepared when [thinking is needed]. "
Homerun! Welcome to the age of cerebral obesity, clogged neurons, and synaptic arrhythmias, for which new types of diets, pacemakers, and bypass surgeries will likely be required. Couple that with the hormonal tendency of grandiose models of language to turn any red-blooded meatsack into a [1]stark raving lunatic (by speech alone), and you end up with a public health nightmare scenario of stupendously colossal mammoth proportions!
GenAI should be available by MD prescription only, in experimentally-verified metabolically appropriate dosages IMO!
[1] https://en.wikipedia.org/wiki/How_Real_Is_Real%3F
ChatGPT, comment on this article in a pithy and thought provoking way.
I'm going for a cuppa.