We’re reaching a point where the harm of AI is going to outweigh the benefits
Follow Keypoint Intelligence on LinkedIn, for the latest insight and updates!
The Internet has a new genre of videos to watch: College commencement speakers being booed for promoting artificial intelligence (AI) as an inevitable future. So far, we’ve seen it happen at the University of Central Florida, University of Arizona, and Middle Tennessee State University. These situations often unfold with a leader in a business or tech field announcing that AI is the future and that the graduates (many of whom will struggle as AI replaces some lower-level positions) need to embrace it to better promote themselves in their careers.
Sorry about the English degree you just received, but you can always go back and study AI to become an expert in creating generative prompts and vibe coding. Then you can have the benefits of human and artificial intelligence, right? The irony of this sentiment is that AI use can actually reduce cognitive function.
A new study out of MIT compared the brains of people tasked with an assignment who were broken out into three groups: those who used large language models (LLMs) to complete it, those who used a search engine to help complete it, and those who used only their brains. When observing the neural pathways formed from performing the task, the study found that “Brain connectivity systematically scaled down with the amount of external support: the Brain‑only group exhibited the strongest, widest‑ranging networks; Search Engine group showed intermediate engagement; and LLM assistance elicited the weakest overall coupling.” This means that those who used only their brain generally had a better understanding regarding their assignment and a greater grasp on the details compared to those who utilized search engines and (especially) those who relied on AI.
A Race to the Bottom
So why does it matter if using AI rewrites neural pathways to our detriment? AI completes tasks successfully so that we don’t have to, right?
Unfortunately, the success rate of AI completing tasks may be misleading. Another study done by the Remote Labor Index found that, despite apparent AI benchmarks, current AI models fail to match human quality on over 96% of professional tasks. Researchers took over 200 real world freelance tasks that were given to contract employees and fed them into various AI models to compare both outputs. These tasks ranged from more creative jobs like video editing and graphic design to data analysis and document preparation.
Rate of Fully Automated Projects (Source: RLI: Remote Labor Automation)
The study found that there were four main categories where AI failed:
Poor Quality: The AI generated content that was far below company standards, often appearing overly simple and unprofessional.
Incomplete Work: Generated work was often missing components outright or pieces required for completion were only partially done.
Technical Failures: Files produced were corrupted, done in unusable formats, or the deliverables were lacking any content.
Inconsistences: AI generated work shifted from one view to another; for example, 3D floorplans didn’t match the architectural outputs done by the same app for the same project.
The issue is that we’re still rushing forward without reflecting on if AI should be used for tasks, let alone if they can actually do what we’re asking. According to Reuters, investors have put about $1.6 trillion into AI during the 2013-2024 period, with another $375 billion investment as of 2025. This has led to widespread abandonment of human employment in favor of AI agents: An estimated 136,000 positions were explicitly removed from the job market in favor of an AI replacement over the past three years. And, according to investment firms like Goldman Sachs, the people generally more affected by this are those just entering the job market as many of these lost jobs are things we once considered to be entry-level. (Note that the actual, exact number of jobs lost to AI is hard to provide as many companies are broadcasting “AI restructuring,” so there could be more or less than what is generally reported.)
Keypoint Intelligence Opinion
We are reaching a point where the whole AI experiment is seeming like an example of the sunk-cost fallacy: We’ve put so much money and effort into making AI agents and LLMs work that we’re now hurting ourselves. There are complaints that “no one wants to work anymore”, but there aren’t any jobs because we’re giving them to AI agents.
In a recent blog post entitled “After Automation”, Every CEO Dan Shipper remarks that “AI commoditizes the residue of human expertise” and provides examples of how his company manages to utilize AI agents alongside human employees—clearly noting that AI creates more work for people (not less) and that it commoditizes skills people have to create the same output over and over that requires a human touch to keep content from joining the piles of slop clogging the internet.
We need to stop putting blind faith in AI’s capabilities and really take a hard look at what’s being produced. Arguments can be made for utilizing generative AI for repetitive tasks, summarizing large amounts of content into easier bullet points, and assisting with research, but we can’t just assume what’s being presented is correct without any verification.
We also need to keep in mind that, for all the good it can do, AI is only a tool. Companies need to start treating it as such and not like a replacement for human employees. After all, about 30% of the companies that laid-off staff in favor of replacing them with AI chatbots and other initiatives are rehiring all these people back—effectively destroying whatever cost-saving measures they saw by having to pay for programs and people.
The industry needs to be smart…and maybe start doing its own research instead of asking AI to do it for them.
Stay ahead in the ever-evolving print industry by browsing our Report Store for the latest insights. Log in to the InfoCenter to view research and studies through our Workplace- and Production-based Advisory Services. Log in to bliQ for product-level research, reports, and specs. Not a subscriber? Contact us for more information.