Many of us utilize AI daily in our higher education work, yet we may not have assessed the ethical and human-centered nature of the tool we have selected and trained through our prompts.

AI tools are no longer a relatively simple search engine that is driven by marketing metrics to help us conduct our research. Rather, with AI we are using more sophisticated tools that conduct research and seek answers to our prompting while making source-selection decisions, contextual settings and semantic subtleties that impact the values expressed in the results.

As I have mentioned previously in these columns, most often I seek input from a current version of each of the three frontier models when conducting research. The three-viewpoint approach allows me to survey a variety of sources, points of view, and to balance the output to address ethical and social perspectives. In the case of this article, I have hyperlinked immediately below the foundation research responses I elicited from prompts on April 19.

ChatGPT 5.4 Thinking model suggested “In higher education, the ethically preferable AI model is not necessarily the most powerful one; it is the model that performs well enough for the use case while offering the strongest evidence of human-centered design, transparency, safety testing, and institutional controllability.”

Claude Sonnet 4.6 Adaptive model suggested “choosing an AI model is now an ethical act, not just a technical one. The field has moved from “does this work?” to “does this serve?” Your column can help deans and department chairs become informed ethical consumers — not AI engineers, but critical stewards.”

Gemini 3 Thinking model noted “Given your recent work on maximizing returns in AI administration, shifting the focus toward “R-Values” (Return on Values) is a timely and necessary evolution for the Higher Ed conversation.”

Before we look at the default values and orientations inherent in some of the leading AI models, let me remind you that in crafting your prompt you can encourage the tool to put an emphasis on generating responses that include orientations and perspectives that address ethical considerations, Your prompt can direct the model to provide results that explore, highlight or emphasize pro-social or human-centered solutions and examples. Over time, if you include such directions in your prompts the more sophisticated models that retain memory of your prior prompts will learn that you are interested in those values. If your preferred perspectives are not included, you can refine the responses by including a request in an iterative follow-up prompt.

Dr. Cornelia C. Walther is a visiting scholar at the Wharton School of the University of Pennsylvania, and a humanitarian practitioner who spent more than 20 years at the United Nations. Her research focuses on leveraging AI for social good. Walther notes in a recent edition of Knowledge at Wharton that most research on AI models is done “exclusively through the lens of efficiency gains, cost reductions, and revenue lift.” However, Walther says “existing dashboards do not capture whether an AI system is fair, whether it is eroding or building trust, whether it is making the people who use it more capable or quietly deskilling them, and whether its environmental footprint is accounted for or simply ignored.”

Late last summer, Walther published an article in Forbes “Why ProSocial AI Is ProPlanetary AI. A Promise For Planetary Harmony” in which she explained an array of elements of assessing AI that may be used to determine the sensitivity to social good. Walther notes that pro-social AI is “ not just about making AI more helpful or ethical. It’s about creating technology that is simultaneously pro-people, pro-planet, and pro-potential.” She points to the 2025 AI Safety Index – Future of Life Institute as an early example of such an assessment. In that index, among seven of the largest models, Anthropic scored a C+ with a 2.64; OpenAI a C with 2.10; and Google DeepMind a C- with 1.76. Notably, DeepSeek scored an F with 0.37.

If you are seeking to look more closely at the AI tools that you use, including custom tools that your university may use for specific purposes, Walther suggests utilizing the key pro-social elements to create a four-by-four grid. Those elements are detailed in the Knowledge at Wharton article:

THE 4 Ts

• • Tailored: Is the AI system designed for the specific context, culture, and constraints of its users — not copy-pasted from a generic template?
  • Trained: Is the system built on representative, inclusive data and objectives that encode the values the organization actually wants to promote, not proxy metrics that are merely convenient?
  • Tested: Is it rigorously evaluated for bias, robustness, and unintended consequences — before deployment and continuously afterwards?
  • Targeted: Is it applied where AI adds genuine value and withheld — deliberately — where human judgment is irreplaceable?

The 4 Ps

• • Purpose: Does the system advance a mission that all stakeholders can be proud of, beyond the next quarterly cycle?
  • People: Does it improve the experience, agency, and well-being of everyone who builds, uses, and is affected by it?
  • Profit: Does it generate durable financial value — not by externalizing costs onto society, but by creating genuine worth?
  • Planet: Is its energy consumption, materials footprint, and systemic environmental impact accounted for and actively reduced?

Walther suggests assembling a leadership team ready to act now. The entry point is deliberately low friction. Choose one AI system currently in production such as a customer-facing chatbot, a hiring screening tool, a demand-forecasting model and convene a 90-minute cross-functional workshop with representatives from technology, HR, finance, legal, and sustainability. Working through the 4 x 4 grid of 16 cells of the matrix together, score each one on a simple traffic light system: green (strong), amber (developing), or red (not compliant).

You do not need a consultant or a new software platform to do this. You need intellectual honesty, a willingness to act on what you find, and the conviction that the institutions that flourish in the algorithmic age will be those that had the wisdom to decide, first, what deserves to be managed and then build the instruments to match. That 90-minute conversation is where the shift from “treasure what you can measure” to “measure what you should treasure” begins.

If your institution-wide mission or goals include ethical, human-centered or pro-social values, you must assess and, where necessary, remediate those AI tools that fall short of the collective values. Are you ready to lead the initiative to begin addressing the pro-social orientation of the AI tools you use in your university department, college, school or division?

This column was originally published in Inside Higher Ed. 

The spring semester is coming to a close with the normal host of routines. Yet, beneath those routines, something is unfolding in the labor market that will greet your new graduates who will face an incrementally tighter job market.

I asked Claude Sonnet 4.6 Extended Thinking to research the tasks relevant to preparing our new students for the new realities they will face when they begin their planned careers. The urgency is real. Entry-level hiring at the 15 biggest tech firms fell 25 percent from 2023 to 2024, according to a SignalFire report. With AI tools performing more of the work previously reserved for recent graduates, new hires are expected to slot in at a higher level almost from day one. That is not a distant forecast. That is the market your Class of 2027 will enter.

The transition from generative AI to agentic AI is the inflection point that makes this moment different from prior technology shifts. Agentic AI systems plan, reason, act, and collaborate independently, shifting from augmentation to widespread displacement and the creation of new roles. In the workplace, full AI implementation jumped from 11% to 42% year-over-year, and CIOs report that 30% of the AI budget is now reserved for agentic AI.

What does this mean for graduates? Entry-level roles are evolving from task execution to “orchestration” of digital labor and AI agents, demanding that career progression becomes a non-linear “mosaic” based on project outcomes and continuous skills development rather than tenure. As Salesforce VP of Workforce Innovation Ruth Hickin put it, “It’s no longer just about execution. It’s also about orchestration as a core skill.”

McKinsey’s “State of AI in 2025” report reveals that 62% of organizations are experimenting with AI agents, with 23% scaling agentic systems within at least one business function. Many of the tasks that a new hire has historically performed: information gathering, basic data analysis, report writing, presentation development are what AI now does well. Those basic tasks are a form of training for the new hire. If AI removes that scaffold, programs that have not deliberately replaced it with something else are sending unprepared graduates into the workplace.

Anthropic CEO Dario Amodei is among those who forecast that 50% of entry-level jobs may ultimately be displaced by AI as the technology matures. Even short of that figure, the combination of skills-based hiring, AI-driven task displacement, and employer pessimism creates a perfect storm for students in academic programs that have not deliberately recalibrated. This further raises the question about what happens to the traditional career ladder that prepares young workers to start at a firm, stay at a firm, and rise all the way to the top?

Claude Sonnet 4.6 Extended Thinking on April 5, 2026 responded to my prompt with seven important tasks for us to complete prior to the Fall term, paraphrased below. Sonnet concluded with a warning for the workplace as a whole.

1. Conduct an Immediate Program Audit Through an AI Lens

Before anything else can happen, chairs need to know what they are working with. Map every course in your degree against the question: Which learning outcomes in this course are now competently performed by a mid-tier AI agent? This is not a rhetorical exercise — it is the diagnostic that will drive everything else. Engage your faculty in this audit collaboratively, framing it not as a threat to their courses but as an opportunity to elevate them.

2. Make AI Fluency a Program-Wide Graduation Competency

A single “AI tools” elective is insufficient. The emerging consensus in higher education calls for moving beyond academic integrity concerns to “AI fluency” as a graduation standard, where assessments focus on process rather than product. Deans should work with chairs this summer to identify how AI fluency, not just AI literacy, gets threaded across every program. This means students learn not only to use AI tools but to prompt, evaluate, supervise, and critically interrogate their outputs. AI literacy is the number one skill Chief Human Relations Officers say workers need as businesses move into the agentic economy.

3.Redesign at Least One Capstone Around Agentic Task Supervision (Orchestration)

The most valuable thing a program can do in the time remaining is ensure students have supervised, reflected upon, and demonstrated the ability to direct AI agents toward meaningful professional outcomes. Salesforce has identified ten essential enterprise skills for the agentic era, organized into human competencies (adaptability, accountability, collaboration, emotional intelligence), agent competencies (AI literacy, human-agent collaboration), and business competencies (problem solving, critical thinking, data interpretation, storytelling). Capstone projects redesigned around these skills, with real deliverables and employer-connected review panels, will differentiate graduates in ways that a revised transcript alone cannot.

4. Accelerate Employer Engagement Around AI Role Evolution

Most advisory boards are too slow-moving to provide the intelligence needed right now. Chairs should convene focused listening sessions with several regional or national employers this summer to ask specifically: What does the first 90 days look like for a new hire in your organization today, compared to two years ago? What tasks have AI agents absorbed? What do you now expect humans to handle that you previously would have trained them on? The answers to those questions should drive course-level revisions for fall. Industry experience and demonstrated proficiencies are among the top factors considered by employers in NACE’s Job Outlook 2026 survey.

5. Expand and Credential Experiential Learning ASAP

To help students prepare, the education system will likely need to change — particularly by encouraging students to become proficient using AI and to take on more hands-on, experiential learning. A new model of “AI apprenticeship” is emerging where juniors use AI to bypass the experience gap and perform at mid-level capacity. Deans should work with their institution’s continuing education and workforce development offices to fast-track partnerships placing students in supervised, AI-augmented project environments before they graduate.

6. Audit and Revise Graduate Advising Language

This can be done immediately. The language your career advisors use with students including the skills they highlight, the roles they flag as targets, the industries they emphasize, should be updated to reflect the current landscape. The Federal Reserve Bank of New York’s recent labor market analysis found philosophy majors outperforming computer science graduates in employment prospects, with the shift tied to the premium placed on human reasoning, adaptability, and cross-domain thinking. That is a story worth telling students, and it changes the advising conversation considerably.

7. Launch a Rapid Faculty Development Initiative Focused on AI Integration

Faculty cannot redesign courses around competencies they haven’t personally developed. A high-engagement AI workshop series in May and June emphasizing hands-on agentic tool use in discipline-specific contexts is among the most impactful investments a dean can make this summer. Faculty who return to their courses in August having deeply engaged with these tools will redesign their pedagogy.

Finally, Claude Sonnet 4.6 Extended Thinking issued a structural warning.

There is a cautionary note that academic leaders must hold alongside all of this urgency. If companies stop hiring juniors at scale, they risk eating their own seed corn. By 2030, industries may face a critical shortage of true senior leaders; those capable of understanding systems below the AI abstraction layer. The risk, as one analyst put it, is creating a generation of “architects who have never laid a brick.”

We in higher education now must ensure that graduates bring the contextual judgment, ethical reasoning, and human relationship capacity that agentic systems cannot replicate. UPCEA’s 2026 Predictions for Higher Education foreground workforce and employer alignment as a critical convergence point, emphasizing that labor-market volatility is forcing colleges to rethink how learning connects to opportunity.

There is much “good work” to be done before the fall semester begins. How might you contribute to this important work on your campus?

This column was originally published in Inside Higher Ed.