AI in Schools: Assessment is the Issue

Most of us would likely agree that younger students, in particular, need to master some cognitive skills and that artificial intelligence could impair such learning. In other words, “AI should not do the student's thinking.” 

But recent moves by some schools to ban use of generative AI in primary schools, whether temporary or permanently, are unlikely to stand the test of time, one might suggest. 

Schools arguably have good reasons to restrict AI use to the extent that they try to measure unaided learning. Most teachers are likely to agree that the purpose of any assessment method is to determine what the student knows, not what language models know. 

So cheating and a loss of foundational skills are key issues for schools. Tests and grades can lose their value as assessment tools if teachers cannot separate language model output from student proficiency. 

On the other hand, the fact that educators don't yet know how to redesign education around AI might not be a robust reason for banning AI use. Is that the learner’s problem or the instructor’s problem?

School system / jurisdiction	Policy	What is actually restricted?	Rationale
New York City Public Schools	2026–27 moratorium	Student-facing generative AI banned in grades 2K–8; limited approved use in grades 9–12	Developmental concerns, human interaction, avoiding outsourcing thinking; high schools receive AI literacy
Fairfax County, VA	2026 restrictions	Elementary students prohibited from generative AI; secondary students require specific authorization for specialized uses	"Human-centered" education, caution while formal policy is developed
Seattle Public Schools	Restricted/teacher-directed	AI permitted when teacher authorizes it; unauthorized AI use treated as academic dishonesty	Emphasis on student thinking, transparency and responsible use
Los Angeles Unified	Initially restricted, subsequently opened controlled access	In Dec. 2022 LAUSD temporarily restricted ChatGPT and other GenAI while developing safeguards; subsequently provided access for students 13+	Shift from outright prohibition toward safeguarded use
Fairfax County student devices	Technical blocking	General-purpose GenAI remains blocked on district-issued student devices even though personal-device use is governed by assignment rules	Security and control of school technology
New South Wales, Australia	Assessment restrictions	2026 reforms sharply restrict take-home HSC assessment because of AI concerns	Preserve authenticity of assessed work rather than attempting to ban AI everywhere
England	No blanket student ban	Schools choose their own rules; government recommends supervised, safeguarded student use	AI literacy plus safeguards rather than prohibition

Among the problems is that banning technology does not seem particularly effective, longer term. 

Study	Technology	Finding	Implication for AI bans
Allcott et al., NBER 2026	U.S. school cellphone bans, >43,000 schools	Lockable pouches substantially reduced phone use, but average test-score effects were close to zero; effects on well-being evolved over several years	A ban can change behavior without necessarily producing the hoped-for educational gains
Lichand et al., NBER 2026	Cellphone ban in Rio de Janeiro schools	Phone use fell and test scores increased about 0.06 SD	Restrictions can work when the technology genuinely interferes with learning
Figlio & Özek, NBER 2025	Florida cellphone bans	Short-term suspensions increased; disciplinary effects dissipated after the first year	Enforcement can generate costs of its own
Kessel et al., Sweden	Swedish secondary-school cellphone bans	Found no impact on student performance and could reject even small positive effects	Removing technology doesn't automatically improve learning
Rahali, Kidron & Livingstone, 2024	Rapid review of school smartphone bans	Meta-analysis found a statistically significant but modest overall effect (d=.162), larger for social well-being than academics	Benefits of bans are real but relatively limited
OECD/PISA 2022	School smartphone restrictions	In schools with bans, many students nevertheless reported using phones every day or several times a day	Formal prohibition does not equal behavioral prohibition
EdWeek Research Center, 2024	School cellphone enforcement	Students reported using smartwatches, alternate devices and other methods to circumvent restrictions	Students adapt around technological restrictions
SMART Schools study	UK school phone policies	Restrictive policies reduced phone/social-media use during school, but not overall weekday/weekend use or mental well-being	Restrictions often move behavior rather than eliminate it

Perhaps other bans, such as forbidding student access to smartphones during the school day can work, in a controlled setting, for some types of technology. But it remains far from clear that long-term impact is positive. 

Perhaps bans are good at reducing exposure to a technology inside a controlled environment. They are much less reliable at producing large improvements in educational outcomes. 

Also, “cheating” might not be a technological problem, but a human problem. Students cheated before ChatGPT. Calculators, phones, Google, Wikipedia, friends, answer sites and copied homework provide examples.

AI changes the cost and scale of cheating, but banning one particular tool doesn't necessarily eliminate the underlying incentives. 

The strongest argument for bans, though, is developmental sequencing. We believe students need to learn fundamentals first. That is why art or music students are schooled in classical fundamentals before they start exploring their own interpretations. 

For similar reasons, instructors of mathematics or writing are likely to continue emphasizing mastery of foundational concepts and forms. 

But practical bans will be difficult when AI is built into virtually all major technology platforms people use. 

The issue might be assessment methods more than “learning” methods, though. Probably everyone would agree that students need cognitive mastery of forms before outsourcing work to AI. The relevant issue might be that teachers do not yet have a good way of assessing such mastery when AI tools are available. 

It is one matter to “redesign assessment”methods. But some methods are sort of “brute force,” such as shifting to in-classroom writing rather than “take home” work. 

Redesigning assessment for online learning scenarios will be quite a bit harder, one would think. 

Policy	Short-term effectiveness	Long-term viability
Block ChatGPT on school Wi-Fi/devices	High	Low
Ban AI for particular assignments	High	High
Ban AI for young children	Potentially high	Quite plausible
Ban AI for all K–12 students	Moderate	Low
AI detectors as enforcement	Low–moderate	Very low
Require disclosure of AI use	Moderate	High
Teach AI literacy	Moderate initially	Very high
Redesign assessment around demonstrated competence	High	Very high
Allow AI but require students to explain/defend its output	Moderate–high	Very high

Compulsory License and AI Copyright

Language models are creating new terrain for copyright law.

Personally, I favor as much freedom as we can possibly endure. On the other hand, practical economic realities are likely to dictate outcomes that balance payments to rights owners and encouragement of innovation by AI firms. 

History strongly suggests that pattern will emerge. 

In the past, rights holders usually wanted control (veto power over new uses). Courts and Congress usually compromise by substituting compensation (compulsory licenses, collective societies, levies) for control. 

Radio stations, cable companies, physical media distributors and streaming platforms captured enormous value from others' content, then eventually reached licensing accommodations. 

Player piano rolls in the late 1800s reproduced sheet music mechanically, and publishers argued this was infringement. 

The Supreme Court ruled in White-Smith Music v. Apollo (1908) that piano rolls were not copies because copyright covered notation readable by humans, not mechanical reproductions.

Congress then passed the Copyright Act of 1909, which created the  compulsory mechanical license, under which anyone could record a song already recorded by someone else, provided they paid a statutory royalty.

Radio created a new problem:

• Composers and publishers got performance royalties through ASCAP (founded 1914), which negotiated blanket licenses with broadcasters.

• Performers and record labels got nothing. The law treated a broadcast performance as fundamentally different from a mechanical reproduction, so the actual recording artists received no royalties when their records were played on air.

Film created other issues. Sound film after 1927 meant a single film now embodied multiple copyrights simultaneously:

• the screenplay

• the musical score

• the recorded performances

• the film itself. 

The question of who owned what in a collaborative industrial production led to the work-for-hire doctrine. Studios owned everything, employees and contractors owned nothing.

International distribution exposed the territorial nature of copyright immediately. The Berne Convention (1886, continuously expanded) became the framework for international harmonization.

Magnetic tape, then cassettes, then VHS created successive waves of the same basic problem: cheap, accessible reproduction technology. 

The 1971 Sound Recording Amendment was the first law to give sound recordings their own copyright protection (previously, only the underlying composition was protected).

Sony v. Universal City Studios (the "Betamax case," 1984) is one of the most consequential copyright decisions ever. Hollywood sued Sony for making VCRs, arguing Sony was liable for the infringement its customers committed. 

The Supreme Court ruled 5-4 that:

• Time-shifting (recording TV to watch later) was fair use

• Selling a device with substantial non-infringing uses didn't make the manufacturer liable.

The Betamax decision shaped technology law for decades. It's why VCRs, DVRs and the consumer electronics industry could exist without needing Hollywood's permission. 

The Audio Home Recording Act (1992) eventually addressed digital audio tape (DAT) by requiring copy-protection technology in DAT recorders and adding a levy on blank digital media, a compromise that established the principle that hardware makers could be taxed to compensate rights holders.

Cable TV initially retransmitted distant television broadcast signals without compensation. The Supreme Court ruled twice (1968, 1974) that cable retransmission wasn't infringement because cable companies weren't "performing" the works. 

But the Copyright Act of 1976 created a compulsory license for cable retransmission.

This compulsory-license-as-compromise model recurs throughout copyright's adaptation to new media.

Xerography created frictionless reproduction of printed text.

The 1976 Act addressed this partly, and the Copyright Clearance Center was founded in 1978 as a collective licensing organization, allowing libraries and businesses to pay blanket fees for photocopying rights.

CDs introduced a paradox: perfect digital copies. This era produced the first serious deployment of Digital Rights Management concepts. It also sharpened the debate about the first sale doctrine — you can resell a CD you bought, but can you resell a digital file? 

The DMCA (1998) was the legislative response. 

MP3 and the iPod era forced the unbundling of the album into individual songs, collapsing per-unit revenue and forcing the industry toward licensing models.

Streaming is philosophically interesting because it doesn't fit the reproduction model at all — you're not copying anything, you're accessing a performance. This is actually closer to the original concept of performance rights than anything since radio.

Spotify, Netflix, and their successors forced the creation of entirely new licensing frameworks:

mechanical licenses for on-demand streaming, negotiated rates between platforms and labels, windowing strategies for film. 

The Music Modernization Act (2018) was the most significant music copyright reform in decades, largely cleaning up the licensing mess streaming had created.

Crucially, streaming finally gave performers (not just composers) digital performance royalties. 

AI companies likely will eventually fit this compulsory license pattern almost perfectly.

The internet produced many precedents that might shape AI copyright:

• The Digital Millennium Copyright Act

• The notice-and-takedown system (rights holders can demand platforms remove infringing content, and platforms that comply get "safe harbor" protection from liability

• Anti-circumvention rules making it illegal to bypass digital rights management (DRM) technology

• ISP liability limits** — shielding internet service providers from liability for what users transmit, provided they act on takedown notices

• The No Electronic Theft Act (1997) closed a loophole where non-commercial infringement wasn't clearly criminal. Before it, you had to be *selling* pirated content to face criminal liability

• The Sonny Bono Copyright Term Extension Act (1998) extended copyright terms by 20 years

• Perfect 10 v. Amazon/Google (2007) established that search engine thumbnail images and inline linking could qualify as fair use

• Authors Guild v. Google validated Google's mass scanning of copyrighted books as fair use

• Viacom v. YouTube (2012) confirmed that platforms don't lose protection simply because infringement is general knowledge

• The Napster and Grokster cases (2001, 2005) established "contributory infringement" and "inducement" doctrines: you can be liable not just for infringing yourself, but for building a platform *designed* to encourage infringement. 

Beyond law, the internet forced copyright to adapt through raw economic pressure:

• Licensing at scale became essential

• Creative Commons (2001) created a voluntary licensing system letting creators specify in advance what reuse they permit

• Terms of service became a shadow copyright system

• Geoblocking and regional licensing became standard business practice as companies realized the internet didn't respect territorial copyright boundaries that the entire global licensing system was built around.

The historical pattern suggests the outcome will likely be some combination of new licensing frameworks, platform liability rules, and compensation mechanisms rather than either "AI training is fully free" or "AI training is categorically infringing." 

The internet precedents suggest the law will bend toward enabling the technology while extracting some structural protections for creators.

My own thinking leans towards permissiveness where it comes to the use of content. As for the argument that language models unfairly infringe copyrights because they “read” or “ingest” content, my argument would be that this non-infringing work is what humans do routinely, and it is not copyright infringement. 

When a person reads a novel, watches a film, or listens to music, they:

• Absorb patterns, styles, vocabulary, narrative structures

• Develop taste and skill influenced by what they've consumed

• Produce new works that are clearly downstream of that consumption

• Do all of this without paying royalties or seeking permission

Nobody considers this a copyright violation, even when the influence is obvious. A novelist can write "in the style of Hemingway" after reading all his books, or a musician whose sound is clearly shaped by artists they grew up listening to.

AI models do that, but with some important differences, some will argue:

• Scale and speed

• Reproduction during training, which involves making copies of copyrighted content and storing them (transiently) on servers. Courts have historically treated that mechanical act of copying as significant, regardless of what happens downstream

• Memorization and regurgitation at scale that arguably collapses the distinction between learning from something and copying it

• Market substitution when an AI trained on a photographer's portfolio can produce images that replace demand for that photographer's work.

The enduring issues suggesting preserving freedom are:

• knowledge and style aren't ownable

• culture builds on culture

• the internet was built on the assumption that content could be read and indexed.

The case for creator protection rests on: 

• copyright exists precisely to ensure creators can sustain the work of creation

• the AI industry is extracting enormous commercial value from creative labor without compensation

• "humans do it too" ignores that humans don't build billion-dollar products directly monetizing others' uncredited work.

The resolution will require development of compensation structures that make the ecosystem fair to creators while not strangling a genuinely transformative technology.

In my view, compulsory license in some form.

How Zero User Interface Might Work

OpenAI is said to be working on a smartphone optimized for language models, something that might be called a " Zero User Interface" model, where the app-centric mobile environment becomes an agentic experience.

Zero UI represents a fundamental departure from screen-centric interaction, using voice, gesture, sound and biometric signals instead of graphical user interfaces or touchscreens, for example. 

It would represent a fundamental shift in how people interact with technology, much as earlier efforts have focused on form factors including glasses, pins or watches.

Instead of forcing users to navigate complex folder structures and discrete app icons, the device becomes an assistant that understands intent and executes tasks across the digital ecosystem on the user’s behalf, perhaps often without the use of a screen-based interface.

In a smartphone optimized for local language models, the interface moves from "command-based" (where the user clicks icons to trigger features) to "intent-based" (where the user describes the desired outcome).

Traditional UI forced users to know where to click and what to configure, for example. Zero UI systems shift from telling the computer what to do to specifying what outcome is wanted.

Instead of building a spreadsheet to assess customer churn, the user says “show me users who are likely to churn in the next seven days.”

Then the followup prompt might be “recommend the best channel to reach them.”

Without screens, feedback mechanisms become critical. Haptic vibrations in wearables or auditory cues must replace visual confirmations, for example.

• Unified OS-Level Intelligence: Rather than individual apps handling their own data and logic, the LLM acts as a central system service. It can perceive the current state of the device, understand the content on screen, and perform actions—such as sending messages, adjusting settings, or pulling data from services—without the user needing to manually open specific applications.

• Dynamic, Just-in-Time UI: Instead of a static home screen, the device generates interfaces on the fly. If you say, "Show me my budget for this week," it doesn't just open a banking app; it generates a concise, readable summary view tailored to your request, allowing you to act on the information immediately.

• Contextual Awareness: The system learns your routines, habits, and preferences. It becomes predictive—anticipating that you might want your calendar organized after a meeting or that you need specific controls available while you are driving—without needing explicit prompts.

• 
  

Without a visual display, the interface relies on glanceability, ambience, and human-centric feedback. 

Input/Output Method	Function
Natural Language (NLP)	Your primary "cursor." You speak, and the model understands nuance, intent, and tone.
Haptic Feedback	Provides non-intrusive alerts. A subtle tap could mean a notification, while a sustained pulse could confirm an action was successfully completed.
Ambient Audio/Chimes	Uses spatial audio and varied tones to provide system status or confirm understanding, reducing the need for constant verbal confirmation.
Gestural Recognition	Using cameras or proximity sensors to interpret hand movements (e.g., a "stop" motion to pause audio, or a "flick" to dismiss a notification).
Ambient LEDs/Light	Subtle light patterns can convey status or urgency, offering a "glanceable" way to understand system states without a full text-based interface.

The greatest hurdle: how do you know what the device can do if there are no menus or icons to guide you?

Successful "Zero UI" devices solve this by:

• Proactive Suggestions: The device doesn't wait to be asked; it learns to surface options when they are contextually relevant (e.g., "Would you like me to book your usual ride home?").

• Conversational Guidance: The AI acts as a guide, periodically informing the user of its capabilities or asking clarifying questions to narrow down intent, effectively "training" the user through natural conversation.

• Standardized Rituals: Just as we learned to "pinch to zoom" on smartphones, screenless devices will likely develop a set of universally understood physical gestures or verbal commands that serve as the "navigation system" of the future.

The idea is to present functions as a fluid, intelligent collaborator, not a collection of app silos, with the objective of minimizing the friction between your intent and the digital outcome.