📍 130 million American adults read below a 6th-grade level. A new Wharton study says AI is making that harder to see — and easier to ignore. That is not a productivity story. It is a warning about what happens when the tools get smarter and human capacity lags behind. ───── Hey Builders! There is a version of AI optimism that is easy to hold from the outside. The productivity numbers look good, the tools are getting cheaper, and workers who used to need a degree to write a professional email can now generate one in seconds. On the surface, AI appears to be the great equalizer — giving everyone access to outputs that used to require years of training. But a research paper from two researchers at the Wharton School says something different — and it is making rounds in policy and workforce circles in a way that probably deserves your attention. They call it "cognitive surrender", defined as the moment a worker stops evaluating what AI produces and simply adopts it as their own judgment. And when you layer it over data on American literacy — 130 million adults reading below a 6th-grade level — what you get is not a productivity story. It is a structural risk story that almost nobody is telling. Today's brief is about that risk. And what founders, professionals, and community builders need to understand about it before the economy makes the point more painfully. ───── 1️⃣ The Wharton Study Nobody Is Talking About In a series of three behavioral experiments involving 1,372 participants and 9,593 individual trials, researchers Steven Shaw and Gideon Nave at the Wharton School tested what actually happens to human decision-making when AI enters the room. Their framework — Tri-System Theory — extends the classic dual-process model of cognition (fast intuitive thinking vs. slow deliberative thinking) by adding a third system: artificial cognition. The question they were testing was not whether AI helps people get better answers. It was whether people are actually evaluating AI outputs — or simply adopting them. The results are striking. When the AI assistant was accurate, participants' performance improved by 25 percentage points relative to baseline. Feed it incorrect outputs — hidden from the participants — and performance dropped by 15 points. That asymmetry matters. The accuracy gain from good AI is real, but so is the loss from bad AI, and the participants had no idea which one they were getting. What made it worse: confidence rose either way. People became more sure of correct answers when the AI was right — and more sure of wrong answers when it wasn't. Confidence and accuracy fully decoupled, and no one in the study could feel the gap. The researchers formally defined this as "cognitive surrender" — distinct from simply outsourcing a task to a tool. Their definition: the decision-maker no longer constructs an answer, but adopts one generated by an external system, relinquishing cognitive control entirely. It is not using AI to go faster. It is abdicating the evaluation itself. Who surrenders most? The study found the pattern was strongest among people with higher trust in AI, lower need for independent cognition, and lower fluid intelligence. In other words, the workers most likely to surrender to AI are the workers who most need their own judgment to be intact. 💡 For Founders If users act on your AI outputs without evaluating them, that is a liability — not a feature. Build verification friction into high-stakes outputs. Design for scrutiny, not just speed. ───── 2️⃣ 130 Million Americans — and the Gap AI Is Hiding Now layer the Wharton finding over what we already know about American literacy. According to the U.S. Department of Education — cited widely by the Barbara Bush Foundation for Family Literacy — roughly 130 million U.S. adults, approximately 54% of the adult population ages 16 to 74, read below a 6th-grade level. ProLiteracy puts the floor even lower: approximately 43 million U.S. adults cannot read, write, or do basic math above a 3rd-grade level. These numbers predate AI. They have been sitting in workforce development research for years. What is new is how AI is interacting with them. Workers have always found workarounds for literacy gaps — asking a family member to fill out the form, avoiding written tasks, leaning on a coworker to handle the email. Those workarounds were always at least somewhat visible. A manager could notice. A skills gap could surface in a performance review. AI is now providing a workaround that leaves no trace. The worker who could not write the report can generate it with ease. Compliance documents that were impossible to parse? AI summarizes them in seconds. Contract clauses that required legal literacy? Flagged automatically. Each of those is a genuine capability gain — right up until the AI is wrong, and there is no one in the room who can tell. Because here is what the Wharton study tells us: "cognitive surrender" is not equally distributed. The workers most likely to surrender to AI outputs without scrutiny are the workers with lower evaluative capacity. And the workers with lower evaluative capacity in the workplace are disproportionately the same workers who are operating below a 6th-grade reading level. The skills gap is not being closed — it is being made invisible. And that invisibility is the problem. Amanda Bergson-Shilcock of the National Skills Coalition describes it as "an invisible drag on productivity" that does not show up in standard performance data but compounds through entire organizations. Professor Stephen Reder of Portland State University is even more direct: "The net effect of AI on the workplace is probably going to be increased demand and need for workers with higher levels of basic skills, not lower." AI raises the floor of what workers can produce. But it also raises the floor of what workers need to be able to evaluate. 💡 For Founders The question is not whether your people can use AI. It is whether they can evaluate it. Those are not the same skill. Audit your onboarding for the difference. A worker producing AI outputs they cannot assess is a compliance and quality risk waiting to surface. ───── 3️⃣ The Wage Premium Goes to Evaluators — Not Users We covered this in the June 3 brief, but it connects directly here. PwC's 2025 Global AI Jobs Barometer found that workers with demonstrated AI skills command a 56% wage premium over workers without them. That number has become a headline in every AI and workforce conversation — and it is being misread. PwC defines "AI-skilled workers" by demonstrated competency — prompt engineering, working alongside AI tools, applying outputs to complex tasks. But layer the Wharton findings on top of that and the picture sharpens: the workers who capture that premium are the ones who can read a model's output, catch what it got wrong, apply domain knowledge to correct it, and make a judgment call. That is not a purely technical skill. At its core, it is a literacy function — and it requires the kind of reading comprehension and logical evaluation that 130 million Americans are operating without. The calculator parallel is instructive. When calculators became universal in the 1970s and 80s, they did not eliminate the need for math — they shifted the floor. Workers who understood arithmetic used calculators to go faster. Workers who did not understand arithmetic used calculators to produce wrong answers with greater confidence and less awareness. The net effect on the population was an increased need for mathematical reasoning, not a decreased one. AI is the same dynamic operating at a scale calculators never reached. If 130 million Americans are operating below a 6th-grade reading level, and the 56% wage premium flows to evaluators rather than users, then AI does not democratize that premium. It concentrates it — in the hands of workers who already had the foundational skills to evaluate what AI produces. The gap does not close. It goes underground — and resurfaces later as errors that compound, decisions that fail, and workers who cannot explain what went wrong because they never understood the process that produced the output in the first place. 💡 For Founders Stop asking if candidates use AI. Ask when AI gave them something wrong and how they caught it. That answer tells you whether you are hiring a user or an evaluator. The difference is worth 56%. ───── 4️⃣ The Opportunity Inside the Problem The gap is a problem. It is also a market. One hundred and thirty million adults operating below a 6th-grade reading level — with AI tools accelerating their output while masking their vulnerability — represents one of the largest underserved workforce development opportunities in the country. The adult education system reaches a fraction of them. Employer-sponsored training programs reach another fraction. The digital and AI upskilling industry has largely targeted the top of the skills distribution — the workers who already have the foundational capacity to learn advanced tools. The founders who build for the bottom of that distribution — for workers whose foundational skills need scaffolding alongside their AI fluency — are operating in a near-uncontested market. That includes adaptive learning platforms, AI tools designed with lower-literacy interfaces, workplace coaching programs that blend basic skills and digital literacy, and employer-facing analytics that make the invisible drag visible before it becomes a compliance or operational failure. The policy window is also opening. The OMB Uniform Grants Regulation revision — which we covered in the May 29 brief — includes changes to how federal grants can support workforce development programs. The comment period closes July 13, 2026. For founders and nonprofits building in the adult education and workforce upskilling space, this is a moment to engage. 💡 For Founders If you are building at the intersection of workforce development, adult literacy, and AI fluency — the funding is shifting, the employer demand is growing, and the research is now on your side. You cannot skip the foundation and expect the tools to hold. ───── 🔧 Three moves to make this week 1️⃣ Read the Wharton study Shaw and Nave's "Thinking — Fast, Slow, and Artificial" is available at SSRN (abstract_id=6097646). It is worth reading in full if you are building any product where users act on AI outputs. The behavioral architecture it describes — "cognitive surrender" as a predictable response to authoritative-sounding AI — should inform your product design. Learn more: papers.ssrn.com/... 2️⃣ Audit your team's AI use for evaluation, not just adoption Ask one question in your next team meeting: what is the last thing AI told you that you pushed back on? If nobody has a recent answer, that is a signal. AI tools should be generating scrutiny alongside output. If they are generating only output, your team may be in "cognitive surrender" mode — and you will not know until something fails. 3️⃣ Map your customer base against the literacy data If you are building a consumer product or a B2B2C platform — especially in healthcare, financial services, legal, or insurance — model what happens to your user experience if 30 to 50% of your users are operating below a 6th-grade reading level. Does your AI-generated content work at that level? Does your interface assume comprehension your users may not have? This is not a hypothetical. It is the actual distribution. ───── 💬 Quote of the Day "You still need to know what you're doing." — Stephen Reder, Professor Emeritus of Applied Linguistics, Portland State University ───── 🎬 Closing Thought The productivity numbers look good. Outputs that used to require a college degree are now available to anyone with a phone and a subscription, and the tools keep getting cheaper. But the Wharton study is telling us something important about what is happening underneath those outputs. Workers are adopting AI answers without evaluating them. Confidence rises even when the AI is wrong. And the workers most likely to surrender are the ones who most need their own judgment to be intact. Layer that over 130 million Americans reading below a 6th-grade level — producing AI-generated work they may not be able to evaluate, in a labor market that pays a 56% premium specifically to the workers who can — and what you have is not a productivity story. It is a risk story with a very slow fuse. The gap is not closing. It is going underground. The founders who see that — and build for it — are moving in the direction the economy is actually heading. Not toward a workforce that looks more capable because of AI, but one that genuinely is. Getting there is not glamorous work and it won't be the story journalists scramble to cover. Adult literacy programs, adaptive interfaces, skills assessments, onboarding redesigns — none of it trends. The venture world has never found it exciting enough to fund at scale. But it is the work that determines whether AI becomes a ladder or a mask. And at scale, it is the kind of work that determines something larger — whether a nation's most powerful technological moment lifts its people or leaves them behind. That is not a workforce development question. It is a question about the future of a nation and those who occupy it. Something to think about..... ───── ⚡️ OHUBNext Daily Brief - investments, edge tech, and moves that matter. For 12+ years, OHUB has been building pathways and on-ramps to multi-generational wealth without reliance on pre-existing wealth. Through exposure, skills, entrepreneurship, capital markets, and inclusive ecosystems, we've helped people create new jobs, new companies, and new wealth.