🎥 Watch the full webinar recording (April 16, 2026)

Key Highlights: click link for additional research presented 

• The Hard Data: More tech = lower scores (📝 Summary, 📺 Video clip)

• Classroom tech fails the evidence bar (📝 Summary, 📺 Video clip)

• Multitasking is a myth, and it's destroying student focus (📝 Summary, 📺 Video clip)

• Teacher-student relationship is the single biggest driver of learning (📝 Summary, 📺 Video clip)

• Screens make learning shallower and harder to transfer (📝 Summary, 📺 Video clip)

• No quality evidence that AI supports K-12 learning (📝 Summary, 📺 Video clip)

• AI outsources thinking — and novices pay the highest price (📝 Summary, 📺 Video clip)

• What can parents do (📝 Summary, 📺 Video clip)

• Neurodivergent learning and screens (📝 Summary, 📺 Video clip)

• Tech can erode the executive function skills students need most  (📝 Summary, 📺 Video clip)

PRESENTATION OVERVIEW

1. Jared’s Background: teacher turned neuroscientist

Horvath started as a classroom teacher - that was the passion. He was teaching during the "decade of the brain," when brain-based learning was everywhere.  He went back to school to learn about neuroscience, expecting to return to the classroom in a year or two.  But instead of returning to the classroom, he decided to help schools, teachers, students, and parents find answers to the question: how do human beings actually learn, and what does that mean for how we teach?

His angle on ed tech flows directly from that. He is not anti-tech and says he does not really care about tech one way or the other. He is pro-learning. Learning is a biological process that works in specific ways. So the question he asks of any tool, including screens, is whether it aligns with that process. His reading of the evidence is that most classroom tech does not. 

2. There is a history of getting tech improvement predictions wrong 

Every generation gets told a new tool will transform education.

• 1922, Thomas Edison: "Textbooks will soon be obsolete in schools. The motion picture is destined to revolutionize education." Textbook sales will clear $300B this year.

• 1945, William Levinson: "A portable radio receiver will be as common in classrooms as a blackboard." Horvath polled 100 schools recently - all had a board, three had a radio.

• 1965, Popular Science: Over half of students will be learning from an "automated schoolmarm."  No schools uses an automated schoolmarm (or knows what it is).

• 1985, B.F. Skinner: Computers will let students learn twice as much in the same time.

3. Is digital the future of education?

Data has been coming out that questions the effectiveness of ed tech solutions:

• OECD (runs the PISA test): "People who use computers very frequently at school do a lot worse in most learning outcomes."

• J-PAL at MIT: "Computers do not improve grades and test scores. Online courses lower academic achievement compared to in-person courses."

• A 2024 study of 300,000 U.S. elementary students on reading comprehension: even small amounts of digital device use in class (30 minutes) are negatively related to reading scores.

• A review of 105 learning moderators at the university level: "Expanding the use of digital technology at the expense of any other form of instruction is likely to have detrimental effects on achievement."

Larry Cuban, the scholar who coined the term "ed tech", says: "Computers were supposed to improve learning and alter teaching. Neither of these things has occurred."

4. The data: PISA and NAEP

PISA (international). Every three years, 90+ countries test 15-year-olds in math, reading, and science. When you plot total score against in-school computer use for learning (not for fun, not at home):

• 2012, 2015, and 2018 all show the same pattern: as in-school tech use goes up, scores go down.

• Students using tech six or more hours a day at school score about two-thirds of a standard deviation below peers who never use tech at school.

NAEP (U.S.). The "nation's report card," given to fourth- and eighth-graders every two years since 1992. Each state adopted heavy classroom tech at a different time.  When NAEP scores are aligned to each state's own year of heavy tech adoption - not to a single calendar year - the curve is the same everywhere:

• Fourth-grade math, fourth-grade reading, eighth-grade math, eighth-grade reading: scores go up for two decades. Then, within a few years of digital adoption, all four start to fall.

• The down-curve is tied to tech adoption, not to any particular calendar year.

5. Effect size of Ed Tech

Researchers measure whether an intervention helps or hurts using a single number called an effect size. Positive = helped, zero = neutral, negative = hurt. In most sciences, zero is a fair baseline.

Education is different. John Hattie analyzed 350,000 effect sizes across education; over 95% are positive. Why? Because learning is biological. It happens under almost any condition: change the lighting, rearrange desks, lock a door, and you will see some measurable improvement. So education needs a higher baseline.

The widely used baseline is 0.42, the average effect size across 21,000 studies. Anything below that is doing worse than ordinary school.  The below table is where ed tech sits against the 0.42 bar. None clear the bar. "It helps a little" turns out to mean "it helps less than almost anything else we do in schools."

6. The two narrow exceptions 

Two ed tech use cases do clear the 0.42 bar:

1. Intelligent tutoring: adaptive drill-and-practice that gets harder when the student is right and easier when they're wrong.

2. Remediation for specific learning disorders: using the machine to drill a specific missing skill.

This may be more about learning than about technology - repetition and drilling may be harder to skip than we hoped. These are exactly the "drill and kill" activities that early ed tech promised would be eliminated. The tool only clears the bar when we use it to do the one thing that ed tech marketers said we'd never have to do again. 

Additional note: while these results clear the bar, transferring these skills beyond the technology tutors themselves proves difficult. The data reflects learning within the program, but as Jared discussed later in this webinar, skills developed through ed tech often fail to transfer.  (Video clip later in this webinar where explains this).

7. Three structural problems with ed tech:  Problem #1: Primary Function

It is one thing to know the tool doesn't work. It's another to know why. Horvath identifies three structural reasons.

Mechanism 1: Primary function

Every tool carries a story about what it is for. Hand someone a hammer and 90% of them activate the same story: hit something.  So what is the primary function of a computer for a child? Kids ages 8–18 spend their device time roughly like this:

Over a year: roughly 450 hours of learning use and 2,500 hours of entertainment and multitasking use on the same device. When a child sits down at a school computer, the story already in their head is the one they've rehearsed for 2,500 hours.

Horvath's says "Tech, in the battle of entertainment versus education, entertainment won."

And multitasking is the single worst thing a human brain can do for learning. It's not a skill problem; it's a hardware problem. Human brains have one attentional filter. Each swap costs about two-tenths of a second and usually erases the last half-second of information. Speed drops, accuracy drops, memory tanks.

The behavioral data bears this out:

• At home, students start multitasking within 6 minutes of sitting down with a computer for schoolwork.

• In a Zoom-based remote class, students start multitasking within 15 minutes early in the term, and within 2 minutes by the end.

• In a live classroom with a teacher present, students are off-task 24 to 38 minutes of every hour of computer use. That doesn't count the time it takes the teacher to get them back.

Problem #2: Relationships and empathy 

The strongest predictor of classroom learning is the student-teacher relationship, which has an effect size 0.57. Inside that, the specific ingredient doing the work is empathy, at 0.68 - well over the baseline.

Empathy, in a neuroscience sense, is not a feeling. It's the synchronization between two people.   When a teacher and student are in sync, two things happen:

1. The student isn't just learning from the teacher.  In that moment, they're thinking like them. Understanding rises sharply.

2. When the student hits a struggle point, the teacher knows exactly why and how to pull them through, because they're biologically in sync.

Digital tools, by definition, have no biology to sync with. This is not a design flaw that better software can fix. It's a structural ceiling.

Problem #3: Transfer - additive vs. subtractive

Transfer is the gold standard of education: can students use the skill outside the place they learned it?

Transfer depends heavily on context: where you learn is where you can best perform. Tech is an extremely narrow context: screen, keyboard, track pad, no friction. Learning that happens on a screen tends to stay on a screen.

But the bigger factor is difficulty direction:

• Subtractive transfer (harder → easier): works well. Learn to drive a manual transmission and you can drive an automatic. Athletes train harder than they play.

• Additive transfer (easier → harder): barely works at all. Learn on an automatic and the manual is foreign. You basically start over.

Computers are the easiest environment humans have ever built. Frictionless by design. So:

• Learning on a screen → applying it in the real world is additive transfer. Students essentially start over.

• Learning in the messy real world → using it on a screen is subtractive transfer. It works.

This is precisely why Gen X and Millennials, who never used tech in school, are more digitally literate than Gen Z. And why Horvath's 75-year-old father texts fluently. Easy contexts are easy to enter from hard ones. The reverse is not true.

8. AI in classrooms: what the research actually says

AI has landed in schools across the US faster than the research on it. So it is worth asking what the research says now, before it is everywhere.

Stanford's SCALE Initiative reviewed every peer-reviewed paper on AI and education - 1,100 papers. Stanford found only 20 papers produce strong causal evidence (in either direction). None of those 20 studied K-12 education in the U.S.  Stanford highlights that “much of the research focuses on individuals above 18, is conducted in international settings, is studied in constrained conditions (such as a one-time 20-minute experiment) and focuses on short-term outcomes.”

Horvath examined three lower-quality meta-analyses of AI in K-12 have been published. Their combined effect size looks promising on paper: +0.54. But once you strip out:

• Studies with no methods section (non-replicable, so not actual research)

• Studies that compared AI to nothing rather than to an equivalent method

• Studies that measured feelings of learning rather than learning

• One study that found AI harmed learning but was flipped to positive in the meta-analysis

…the real combined effect drops to +0.17, well below the 0.42 baseline.

The operating principle: AI is not a learning tool. There is no current evidence it will become one.

9. The vetting problem: AI is a tool for experts, not a tool for becoming one

AI is a production tool for people who already have expertise. Horvath explains he can run a statistical analysis by hand in two hours, or through AI in two seconds, but only because his expertise lets him immediately spot when the AI produces something wrong.

A novice does not have that ability to check.  When a student uses the same tool, they cannot tell the difference between a right answer and a confident wrong one. The best they can do is copy it and hand it in.

• Cognitive offloading (what adults do): you already know how to do the task; you let a tool handle it so you can focus elsewhere.

• Cognitive outsourcing (what novices do): you've never done the task; the tool does the whole thing, and you hand it in. You never learn it.

10. What parents can do

Horvath shared some practical recommendations for parents:

Start at home (not at the school).  Build the context for tech at home first, in a way that mirrors what you'd want to see at school.

• Buy a printer. If it's homework or reading, print it out. Get it off the screen.

• Run your own tech audit. Look honestly at the adults first, then the kids. How much time on tech? What for? Is it hurting relationships? Is any of it actually useful?

• Try a tech-free day. A Saturday or Sunday where the whole family goes off screens and does something together.

Once you’ve done that, then build a coalition at school. One parent speaking up is a nuisance. Five parents speaking together have to be taken seriously.

Bring data and narrative. The data is on the parents' side, but to include the narrative. When parents can say not just "scores go down when tech goes up" but "and here's why — multitasking, loss of empathy, failed transfer," the argument lands more effectively.

11. Neurodivergent screens and learning

There are real cases where a child genuinely needs tech - a learning disorder that makes handwriting impossible, a child who cannot be in a classroom with other people, a global pandemic that closes schools. Horvath is explicit that he is not arguing those kids should go without the tools they need. Learning from tech is not zero; it’s just weaker, shallower, and less transferable and durable than other forms of learning.

The caution is about preference versus benefit. He states the research is fairly clear that students with ASD and ADHD, as a group, do not benefit from tech-based learning - they are disproportionately harmed by it, even when they prefer it. Many kids will say they learn better online because it feels easier, quieter, less socially demanding.  But in learning, what feels good may not be what's actually working.  Sitting in a classroom with other people is often more effortful than being alone with a screen, and that effort is doing work.

The practical test for parents: if a child genuinely cannot do the learning without tech, use it. If they are clearly succeeding in one mode and struggling in the other, go with what works. But if the decision is being made on preference alone, preference is not a good guide.

12. LMS (e.g. Google Classroom) impacts executive function

Executive function (planning, self-organization, remembering what's due) used to operate differently at school. A teacher said "this is due Tuesday," and the student had to remember, organize time, bring it home, bring it back, hand it in. The system broke down often, and students learned from the breakdowns.

Learning management systems have smoothed all of that away. The LMS reminds the student, holds the work, submits it, organizes the calendar. The student never builds the muscle. College professors now report students arriving unable to organize their time, remember deadlines, or drive their own work.

Graham Nuthall's line applies: about 60% of what kids learn at school is not content - it's what the structure teaches them. Change the structure, and you change that 60%. Heavy tech has changed the structure dramatically, and the thing being subtracted is self-direction.