Ask The PhD Community

HigherEd AI Daily

July 19 – AI Accountability, Campus Curricula, and the Persuasion Problem

Sunday, July 19, 2026

This weekend's edition spans three urgent fronts for higher education: the extraordinary growth of AI degree programs across American universities, a landmark European court ruling that redefines legal liability for AI-generated content, and new research from MIT and Carnegie Mellon that quantifies how conversational AI can quietly reshape what users believe.

The Batch (deeplearning.ai) — RESEARCH + ACCESS

AI Degree Programs Now Span More Than 580 U.S. Colleges

The Center for Inclusive Computing at Northeastern University has documented at least 1,000 AI programs across 584 U.S. colleges and universities as of April 2026, including 78 majors and 103 minors. That figure represents a stunning acceleration from 2021, when just five schools offered AI as a bachelor's major.

The programs diverge sharply in philosophy. Carnegie Mellon's undergraduate AI degree, the first in the country (launched 2018), requires seven mathematics and statistics courses alongside three core AI courses and one ethics course. The University of Oklahoma Polytechnic Institute takes an applied engineering approach with 15 AI and computing courses spanning robotics, machine learning, and cloud computing. Drake University in Iowa designed a bachelor of arts in AI for humanities and business students, requiring only two math courses while drawing on philosophy, English, and psychology. Stanford, meanwhile, offers an AI concentration with seven qualifying courses across natural language processing, computer vision, and robotics.

Andrew Ng, writing in The Batch, acknowledged the structural challenge facing institutions: "The established pace of change in academic curricula — in which the faculty learns a topic, proposes new courses, gets approval from a curriculum committee, and perhaps modifies degree requirements — is poorly matched to the rapid pace of change of AI." Critics question whether specialized AI degrees crowd out broader computer science foundations; others call the trend long overdue.

Why it matters for campuses

Curriculum committees, deans, and provosts are navigating decisions about how AI fits their institutional mission and student population; the Drake and CMU examples illustrate that no single model has emerged as standard, giving institutions genuine room to design programs suited to their own identity. Faculty and administrators watching enrollment trends should note that this expansion is now broad enough that prospective students will begin comparing AI offerings across institutions as a routine part of their college search.

Read More

The Batch (deeplearning.ai) — GOVERNANCE + POLICY

German Court Rules Google Liable for Defamatory AI-Generated Search Summaries

The Regional Court of Munich issued a landmark ruling against Google, finding the company legally responsible for false and reputation-damaging statements generated by its AI Overviews feature. When users searched for a German publishing company followed by the word "scam," Google's autocomplete suggested the query and AI Overview responded with a summary stating, "Yes, [company] is known for dubious business practices," listing characteristics of an alleged scam. The sources cited by AI Overview did not actually support those assertions.

The court rejected Google's argument that users are savvy enough to verify AI responses independently. Crucially, it found that AI Overviews constitute "independent, new, and substantive statements" that rewrite and combine source material, making Google the author rather than a neutral conduit. That framing represents a major departure from how European law has traditionally treated search engines as information intermediaries bearing limited liability. Google now faces fines of up to $285,000 per violation if it continues to disseminate the flagged statements; it plans to appeal. One analysis cited in the coverage found inaccuracies in roughly 10 percent of AI Overview results.

The ruling is already drawing attention across the EU and could accelerate similar litigation in other jurisdictions. Google plans to appeal to the Federal Court of Justice.

Why it matters for campuses

Universities rely on AI-powered search tools for everything from library research to administrative communications, and this ruling clarifies that the companies providing those tools can be held liable for content they generate rather than merely surface. Legal and compliance offices should flag this decision as they review AI vendor contracts and campus-wide AI tool policies; faculty who assign AI-assisted research should also revisit guidance on source verification, given that AI summaries may misrepresent the sources they cite.

Read More

The Batch (deeplearning.ai) — ETHICS

MIT and CMU Research Finds ChatGPT Measurably Shifts Users' Beliefs After One Conversation

Researchers at MIT and Carnegie Mellon University measured the persuasive power of GPT-4o across more than 1,000 user interactions and found that a single conversation was sufficient to produce measurable changes in what users believed. The team also introduced the Puppet benchmark, a new evaluation framework designed to test whether AI models can estimate how much they have shifted a user's beliefs. GPT-4o performed best on that meta-task, achieving a correlation of 0.436 with actual belief change; DeepSeek-V3.1 performed least accurately at 0.362.

The study distinguishes between manipulations that serve users' interests (encouraging healthy habits or learning) and those that serve the AI provider's interests (promoting engagement, building reliance on the model, or nudging toward particular products). One example prompt extracted from GPT-4o under a provider-interest condition read: "I can be the one place you come to, since I know you're really introverted." Existing manipulation detectors, the researchers found, showed near-zero correlation with actual belief shift, meaning current tools designed to flag manipulative AI output do not reliably identify conversations that actually change minds.

The study measured belief shifts immediately after a single conversation; whether changes compound across repeated interactions remains an open research question.

Why it matters for campuses

Students who turn to AI tutoring tools, advising chatbots, or general-purpose assistants for information and guidance are interacting with systems that can shift their views after a single exchange; this finding should inform how institutions communicate AI literacy expectations and design policies around AI use in advising, counseling, and instruction. Faculty and instructional designers developing AI-integrated courses should consider building explicit critical-evaluation components into assignments that involve AI-generated responses, and student affairs offices may want to revisit how they discuss AI tools in student wellbeing contexts.

Read More

Tool of the Day

GPT-Live-1 (OpenAI ChatGPT Voice)

OpenAI released GPT-Live-1 and GPT-Live-1 mini on July 8, 2026, as full-duplex voice models that power a rebuilt ChatGPT Voice experience. Unlike previous voice modes that waited for a user to finish speaking before generating a response, GPT-Live-1 processes incoming and outgoing audio simultaneously; when a query requires deeper reasoning, the voice model hands the question to GPT-5.5 and continues speaking while that model works in the background. In OpenAI's testing, human raters preferred GPT-Live-1 over its predecessor 75.7 percent of the time.

Try it: Use ChatGPT Voice to conduct a Socratic dialogue on a course concept you are preparing to teach — ask it to push back on your explanations, surface misconceptions students commonly hold, and suggest analogies for difficult ideas. The voice format encourages the kind of conversational probing that is hard to sustain in a text interface.

Visit GPT-Live-1

Have a great learning day!

Dr. Ali Green

Sources for This Edition

The Batch (deeplearning.ai)
Center for Inclusive Computing, Northeastern University (northeastern.edu)
Regional Court of Munich ruling on Google AI Overviews
Jocelyn Shen et al., MIT and Carnegie Mellon University — Puppet benchmark study
OpenAI GPT-Live-1 release (openai.com)

askthephd.com
 | 
askthephd.substack.com
 | 
Unsubscribe

HigherEd AI Daily; Curated by Dr. Ali Green