What does a PhD from five years ago actually mean today, and should we reconsider how education, expertise, and intellectual achievement are measured?
In short
- A degree certifies several things at once (knowledge, skills, research capability, persistence, credibility), and AI is changing the economics of several of them simultaneously.
- When producing evidence of expertise becomes easier than developing expertise itself, traditional assessment starts losing some of its meaning.
- A PhD does not expire, but historical achievement is not the same as current capability, so degrees could be complemented with continuously renewed evidence of capability.
- Removing the effort from a task can also remove some of the learning the effort was supposed to produce, so AI may compress knowledge acquisition faster than capability development.
- The question should move from “What degree do you have?” toward what you can understand, question, and create, and how effectively you extend that with AI.
For centuries, education has followed a relatively stable hierarchy: bachelor’s degree, master’s degree, PhD. Each level represents a progressively deeper understanding of a subject, greater specialization, and, eventually, the ability to produce original knowledge.
But what happens when artificial intelligence can perform an increasing share of the intellectual work that once distinguished these levels?
Consider someone who earned a PhD in computer science in 2020. They may have spent four to six years studying a narrow research problem, reviewing hundreds of papers, developing algorithms, running experiments, and writing a dissertation.
That accomplishment was meaningful then, and it remains meaningful today.
But in 2026, a capable researcher using AI can potentially compress significant portions of that workflow (from literature exploration to implementation and experimental analysis) into a fraction of the time.
This raises an uncomfortable question:
Does the PhD still represent the same level of intellectual capability it represented when it was awarded?
And perhaps the more important question:
Should we continue measuring education primarily by the depth of knowledge someone has accumulated, or by their demonstrated ability to create new understanding in a world where knowledge is increasingly accessible on demand?
1. What did academic degrees actually certify?
Historically, higher education served several distinct purposes that we often treated as one.
- Knowledge: A person has acquired a substantial body of information about a discipline.
- Skills: They can apply established methods to solve increasingly complex problems.
- Research capability: They can formulate questions and contribute something original.
- Persistence and discipline: They can sustain intellectual work over several years.
- Credibility: An institution has evaluated their work and certified that they meet a particular standard.
A PhD, especially, was never simply about knowing more facts than someone with a bachelor’s degree. It represented evidence that a person could operate near the frontier of a discipline and contribute to advancing it.
The problem is that AI is changing the economics of several of these capabilities simultaneously.
Remembering facts is becoming less scarce. Producing technically sophisticated work is becoming more accessible. Exploring unfamiliar academic fields is becoming faster.
None of this automatically makes expertise obsolete. But it does challenge the assumption that an academic credential serves as a sufficiently complete and durable representation of someone’s intellectual capabilities.
2. The difference between knowing something and being able to do something is shrinking
Imagine three people trying to solve a complex machine learning problem. One has a bachelor’s degree and two years of industry experience. Another has a master’s degree. The third completed a PhD five years ago.
In a traditional environment, the PhD graduate might have had substantial advantages in identifying appropriate research methods, understanding relevant literature, designing experiments, and interpreting subtle results.
Now give all three access to capable AI systems.
- The bachelor’s graduate can explore unfamiliar algorithms, request explanations of advanced mathematical concepts, generate implementations, and design experimental approaches.
- The master’s graduate can accelerate analytical work and examine alternatives beyond their original specialization.
- The PhD graduate can use the same tools to extend their existing expertise.
AI raises the potential capabilities of all three. But it may not raise them equally.
The value of previously scarce knowledge may decline, while the value of recognizing mistakes, choosing good research directions, and exercising judgment may increase.
Conceptual illustration: How AI may change the value of expertise
| Educational background | Without AI | With AI |
|---|---|---|
| Bachelor’s | Lower baseline potential | Substantially augmented potential |
| Master’s | Intermediate baseline potential | Substantially augmented potential |
| PhD | Higher baseline potential | Further augmented potential |
Illustrative hypothesis, not empirical measurements. AI could raise capability across educational levels while changing the relative importance of formal qualifications.
Importantly, access to AI does not mean these people would produce equally reliable work. The ability to evaluate AI-generated conclusions still depends on domain knowledge, methodological understanding, and experience.
In fact, one of the central risks is that AI can make people appear significantly more competent than they actually are. A polished output is not necessarily evidence of deep understanding.
And that brings us to a more fundamental problem.
3. Are we measuring educational achievement or the ability to produce educational artifacts?
A university evaluates students through examinations, essays, projects, research papers, and dissertations. These artifacts have traditionally served as evidence of the underlying learning process.
But AI introduces a separation between the quality of the artifact and the capability of the person who produced it.
- A student might submit an excellent research paper without being able to independently defend its assumptions.
- A developer might create sophisticated software without understanding its architecture.
- A doctoral candidate might produce a technically impressive dissertation while delegating substantial parts of the intellectual process to AI.
This is no longer purely hypothetical.
An October 2026 NBER working paper examined AI-generated writing in US STEM dissertations and estimated that approximately 29% of dissertations filed in early 2026 contained AI-generated writing in their abstracts. The study is not peer-reviewed, and detecting AI-written text does not establish that a dissertation’s underlying research was delegated to AI. Nevertheless, the finding illustrates how rapidly academic production practices are changing.
The important question isn’t whether using AI is cheating. It’s whether our evaluation methods still measure the capabilities we intended them to measure.
When producing evidence of expertise becomes easier than developing expertise itself, traditional assessment starts losing some of its meaning.
4. Should a PhD have an expiration date?
A PhD does not expire. Nor should we pretend that the original achievement somehow disappears because technology advances.
But perhaps we need to distinguish between three things:
- Historical achievement: This individual demonstrated the ability to conduct doctoral-level research at a particular time.
- Current expertise: This individual understands the state of their field today and can make reliable judgments within it.
- Current research capability: This individual can independently define important problems, create new knowledge, and effectively use contemporary tools to advance their discipline.
These are related, but they are not interchangeable.
A PhD completed in 2020 remains evidence of historical achievement and demonstrated research training. It does not automatically certify that its holder is at the forefront of AI-assisted research in 2026.
Equally, someone without a PhD might have developed exceptional research capability through industry practice, independent study, and sustained intellectual contribution.
This distinction becomes increasingly relevant as the rate at which fields evolve accelerates.
Perhaps rather than expiring degrees, we should complement them with continuously renewed evidence of capability.
Similar to how certain professional certifications require continuing education, research credentials could be supplemented with a current portfolio of demonstrated competence. Not to invalidate a person’s past work, but to provide a better picture of what they can actually contribute today.
5. The next educational hierarchy may not be about degrees at all
The traditional hierarchy is built around academic milestones.
Traditional model
| Degree | Traditional emphasis |
|---|---|
| PhD | Original research and specialized expertise |
| Master’s | Advanced and specialized knowledge |
| Bachelor’s | Foundational disciplinary knowledge |
- Level 4: Knowledge CreatorDiscovers new principles, methods, and understanding
- Level 3: Problem FramerIdentifies valuable questions and judges competing solutions
- Level 2: AI-Augmented PractitionerSolves complex problems using tools and established knowledge
- Level 1: Knowledge ConsumerUnderstands and applies existing information
In this model, someone with a bachelor’s degree could demonstrate Level 4 capabilities in a particular field, while someone with a PhD might currently operate at Level 2 in an unfamiliar domain.
The levels would not be permanent characteristics of individuals. They would describe demonstrated capabilities within a specific context. This is an important distinction.
Perhaps the future of education is not about climbing a single hierarchy of credentials. It is about developing and continuously demonstrating increasingly sophisticated intellectual abilities. (I made a related argument about engineering careers in The Future of Software Careers After AI Becomes Native.)
6. If AI can do the research, what should a researcher learn?
We should also reconsider the purpose of the learning process itself.
For decades, significant portions of academic training involved activities such as finding papers, reviewing literature, writing code, processing data, formatting publications, and preparing experiments.
These activities are not necessarily valuable because they are time-consuming. They are valuable because performing them can develop understanding, intuition, and intellectual discipline.
If AI can accelerate them, that could be enormously beneficial. But there is a catch.
Removing the effort from a task can also remove some of the learning that the effort was supposed to produce.
A 2025 Microsoft Research and Carnegie Mellon study involving 319 knowledge workers found that greater confidence in generative AI was associated with less reported critical-thinking effort, while greater confidence in one’s own ability was associated with more. The findings are correlational and self-reported, but they identify a significant challenge for AI-assisted learning. (The same problem shows up inside engineering teams; see What Async Can’t Teach You.)
The educational objective should not simply be to minimize the time required to produce a result. It should be to maximize the depth of understanding and intellectual capability gained from that effort.
That suggests a different approach to doctoral education: instead of evaluating how independently someone performs every step of the research workflow, evaluate how effectively they use available capabilities without losing ownership of the reasoning.
For example, a future doctoral examination might ask a candidate to investigate a genuinely unfamiliar problem using AI, defend the assumptions behind their approach, identify weaknesses in generated evidence, and explain which conclusions remain uncertain.
The purpose wouldn’t be to prove that the student can outperform an AI system at every task. It would be to demonstrate that they know when to trust it, when to challenge it, and how to move beyond its existing knowledge.
7. Could we compress years of higher education?
Here is another potentially disruptive implication.
If an individual with AI assistance can acquire relevant knowledge, explore alternative approaches, and perform practical exercises much faster, why should educational programs continue requiring the same amount of time?
A four-year bachelor’s degree, a two-year master’s degree, and a five-year PhD represent conventions shaped by institutions, learning requirements, funding structures, and professional expectations. They are not universal laws describing how long intellectual development must take.
AI tutors, personalized curricula, simulated environments, and automated feedback might eventually allow people to progress through particular subjects considerably faster.
But we must be careful not to confuse faster access to information with faster development of expertise.
Scientific judgment, intuition, intellectual independence, and the ability to recognize flawed reasoning often emerge from repeated experience, including failure.
A 2025 Microsoft Research review of empirical evidence on AI in education specifically recommended that students develop foundational knowledge before relying heavily on generative AI, and that AI supplement rather than replace well-designed learning activities.
Perhaps we should therefore separate educational time into two components:
This could eventually lead to shorter programs in some fields, more intensive practical training in others, and significantly more individualized educational pathways.
The goal should not be to get everyone through a PhD faster. It should be to remove unnecessary friction while preserving the intellectual challenges that make advanced education valuable.
8. The irony: Advanced research skills may become more valuable
There is an apparent contradiction in this entire argument. If AI can perform more research tasks, shouldn’t we need fewer highly trained researchers?
Not necessarily.
AI makes generating plausible hypotheses, analyses, and explanations easier. But it does not automatically establish whether those outputs are novel, reliable, reproducible, or important.
In fact, increasing the volume of generated research may make the ability to distinguish meaningful discoveries from superficial results even more valuable.
A January 2026 Nature analysis highlighted research by Hao, Xu, Li and Evans finding that AI-augmented scientists produced more influential work while scientific activity became more concentrated within established research areas.
This suggests a potential paradox: AI may accelerate research production without proportionally accelerating the discovery of genuinely new directions.
The skills required to challenge assumptions, recognize unexplored opportunities, and design convincing experiments remain central.
In other words, a PhD might become less valuable as proof that someone can independently execute a traditional research workflow, but more valuable when it genuinely demonstrates original thinking, methodological rigor, and scientific judgment.
The degree itself is not the problem. The problem is whether its meaning evolves alongside the capabilities available to researchers.
9. From static credentials to a continuously evolving intellectual profile
Imagine an alternative educational system. Instead of representing someone primarily through a degree earned years ago, we maintain a richer picture of their intellectual development.
Such a profile could contain evidence of foundational knowledge, current domain expertise, original research contributions, practical accomplishments, methodological skills, and effective AI collaboration.
Some dimensions might be assessed through examinations. Others through peer-reviewed work, practical challenges, or independently reproduced contributions. It would evolve as the individual learns.
This does not mean replacing university accreditation with an AI-generated expertise score. Such a score could introduce serious problems with fairness, transparency, and reliability. Rather, the idea is to complement established credentials with verifiable, current evidence.
Universities could remain important institutions for intellectual development, mentorship, scientific infrastructure, and rigorous evaluation. But the emphasis might shift from certifying a completed educational journey to helping people develop and demonstrate capabilities throughout their careers.
A degree would become one meaningful chapter of an intellectual history rather than its definitive summary.
10. What would a PhD earned in 2030 represent?
Consider a hypothetical doctoral candidate graduating in 2030.
Their AI research assistants might continuously review literature, suggest experimental designs, generate and test code, and identify inconsistencies in results. They might complete experimental work that would have required a substantially larger research team a decade earlier.
Should this make their PhD less valuable?
I would argue the opposite, provided that the standards change.
We should expect such a candidate to accomplish things that were previously impractical for an individual researcher. Their contribution should be evaluated in the context of the tools available to them, not against the manual effort required of researchers in the past.
A PhD should represent the ability to advance the frontier of knowledge using the best available methods. It should not represent the preservation of historically difficult workflows.
And as those methods evolve, the definition of a meaningful doctoral contribution should evolve with them.
Conclusion: Maybe we are measuring the wrong thing
A PhD earned five years ago is not suddenly worthless. It remains evidence of a significant intellectual accomplishment, advanced training, and the ability to complete a demanding research process.
But we should question whether it is sufficient evidence of someone’s current intellectual capability.
In a world where the cost of accessing knowledge, generating solutions, and performing complex cognitive tasks continues to fall, the distinction between educational credentials and demonstrated capabilities becomes increasingly important.
Perhaps we are approaching a moment when the fundamental questions about education need to change.
Not simply:
What degree do you have?
But:
What can you understand? What can you question? What can you create? And how effectively can you extend your abilities with AI?
The ultimate purpose of education has never been to accumulate certificates. It is to develop the human capacity to understand the world and contribute something meaningful to it.
AI doesn’t make that purpose obsolete. It forces us to reconsider whether our educational systems are still measuring it correctly.
Further reading
- PhD training needs a reboot in an AI world, Nature, 2025.
- AI and the PhD student: friend or foe?, Nature, March 2026.
- Thinking out of distribution: what a PhD must train in the age of AI, AJE Advances: Research in Epidemiology, October 2026.