[[{“value”:”
There is a crisis in technical education. The golden road to a career has always been through a college education. However, this “golden road” has developed deep cracks and is badly in need of maintenance. Postsecondary education is rapidly becoming unaffordable, even at public colleges and universities. Tuition has risen at a rate 50% greater than inflation. But there’s a deeper issue. Beyond the out-of-control cost, there is evidence that degrees do not map to the skills needed in today’s job market, and there’s an increasing disconnect—particularly in computer science—between the skills employers want and the skills colleges teach.
Employers are struggling with a related problem: keeping the people who are already on their staff up-to-date with the skills they need. It’s common for experts who spend their waking hours working at the cutting edge of the technology industry to feel like they’re falling behind. The trend has only increased in the era of generative AI. A graduate degree is an option for employees who can afford it, but it doesn’t help employers. After spending a year getting a master’s degree, an employee is unlikely to return to the same employer, let alone the same job.
Why, and more important how, are colleges and universities failing? And what can companies that need to hire junior staff and upskill their current staff do about it?
The Problem with College: Agility and Fragility in Disruptive Times
Colleges and universities are rarely agile. They don’t respond to changes quickly, and that leaves them particularly vulnerable when providing training for industries where change is rapid. The traditional CS major may be where colleges and universities are at their weakest. The pace of change is very rapid, particularly when compared to the career of tenured faculty, and the resistance to change can be especially strong when change is rapid. CS departments have adapted well to AI, partly because AI originated in academia. But many jobs require skills that frequently aren’t taught in traditional CS departments, such as cloud development, Kubernetes, and microservices.
Why aren’t these institutions able to adapt to changes in technology? Professors spend much of their time doing research—well, in reality, they spend most of their time serving on committees. There’s little time left over to find out what industry is doing, let alone develop courses to teach it. Staying current in the tech industry is a bit like being a professional athlete: You have to train daily to maintain your physical conditioning. Entirely new paradigms rise quickly: cloud computing, data engineering, machine learning engineering, mobile development, and large language models. To further complicate things, topics like cloud computing, software operations, and even AI don’t fit nicely within a university IT department. They require going outside to commercial cloud vendors, which requires expense accounts and budget commitments that aren’t covered by research grants. No university has the computing resources comparable to Google, or even to a well-funded startup. Nor do they have experience building and operating highly distributed systems.
Topics like microservices and cloud native computing present an additional problem: salary commitments. Can an administrator justify the salary of a senior faculty member who specializes in a topic that might be forgotten five or ten years in the future, even if that’s what industry wants right now? Can the administration commit to paying a tenured faculty salary for 30 or so years if that professor’s specialty may be irrelevant long before they retire? It’s less risky to hire adjunct professors with industry experience to fill teaching roles that have a vocational focus: mobile development, data engineering, and cloud computing.
Using adjuncts to teach the skills that industry wants creates its own problem: an underclass within the university teaching staff. It’s no wonder that I have heard professors say “Cloud computing is a fad and not worth teaching.” It’s no wonder that many faculty members see these topics as “vocational education” while they’re trying to teach long-term verities: these “vocational” topics fulfill the needs of industry rather than the research community and are taught by faculty with short-term contracts who come and go each year. It’s understandable that professors are reluctant to teach a subject that is based almost entirely on proprietary technology that can change with minimal notice. However, if that’s the kind of expertise companies want, students who need that training will find it elsewhere—and if universities can’t provide the training students need, they will lurch toward irrelevance.
In a recent Wall Street Journal article, “Why Americans Have Lost Faith in the Value of College,” Douglas Belkin highlights another issue: critical thinking. As Belkin explains,
The misalignment between universities and the labor market is compounded by the failure of many schools to teach students to think critically. Professors compete for tenure on the basis of the quality of their research and publishing track record. Teaching is mostly an afterthought. Professors who earn tenure negotiate lighter teaching loads. To fill the gap, schools hire less expensive adjuncts with little job security. These precariously employed adjuncts depend on strong student performance reviews for job security, a system that incentivizes them to make few demands in exchange for high ratings.
Any metric can be gamed (often called Goodhart’s law)—and grades are no exception, gamed both by faculty who need good ratings from students and by students who want good grades from faculty. Grades are part of the critical thinking problem, as is a dismissal of writing, reading, and non-STEM skills. It’s not as if critical thinking isn’t taught in universities. The humanities are nothing without critical thinking, whether or not they’re taught by overwhelmed and underpaid adjuncts—but humanities departments are the ones most threatened by budget cuts and, at some schools, outright elimination.
Vocational skills are a necessity, whether or not CS departments want to teach them. Assessment is a necessity, and it’s something corporations take very seriously, at least for in-house training programs. But making vocational skills and assessment priorities risks letting grades become a motivating factor, and that is counterproductive. Is anything more conformist than aspiring to do what your teacher says to get an A? Or to build your academic career around getting a job at a prestigious, high-profile company? Students need to learn how to make mistakes. They need to learn how to push their ideas as far as they can and then a little farther. In my classes I encourage students to fail early and often. A failure is a desired outcome: It means they tried something hard and unique or that they learned a big lesson.
Teaching students to consider problems from many perspectives, including those that are uncomfortable, is a necessity. Too many students graduate thinking that science is a set of facts rather than understanding that it’s a process of skeptical inquiry driven by experiment. Too many students think that engineering is about getting the answer in the back of the book, not about making the trade-offs that are necessary in the real world. And too many companies fail because they can’t question their own assumptions. This is all critical thinking—and anything less shortchanges both students and the companies that eventually hire them.
Companies need well-trained talent
So—colleges and universities are failing industry. They aren’t providing graduates who are trained in the skills companies need; they aren’t nurturing critical thinkers; and they are pricing themselves out of the range of all but the ultrawealthy. What can companies do to acquire and retain the talent they need?
Understanding industry needs isn’t a static project. In 2020, the World Economic Forum estimated that automation will displace 85 million jobs by 2025 but will also create 97 million new jobs. Another publication estimated that there were 13 million unfilled technology jobs. In the 21st century, churn is ever present. Whatever your job is now, it will be different in five to ten years: Your skills will be obsolete, and you’ll need to learn new ones. That’s a problem for both new graduates and experienced staff, to say nothing of the companies that employ them. And as we’ve seen, it’s an even bigger problem for colleges and universities.
To start, we’ll look at what companies actually need, using data from O’Reilly’s learning platform. There are two factors: course enrollments, which show what students are studying, and course completion, which may reveal skills in high demand.
Course completion
The median course completion percentage on the O’Reilly platform is similar for B2B users and B2C users, and roughly matches industry standards. A higher completion rate could indicate that the course teaches an emerging skill that is required in industry. Examples of these skills are artificial intelligence (prompt engineering, GPT, and PyTorch), cloud (Amazon EC2, AWS Lambda, and Microsoft’s Azure AZ-900 certification), Rust, and MLOps. It’s important to note that CISSP (the Certified Information Systems Security Professional) certification is on the list; although security skills are hardly a new requirement, corporate attitudes toward security have changed drastically in the past few years. Executives don’t like seeing their companies in the news for a security breach. Some other topics with high completion rates are ggplot (for data-driven graphics in R), GitHub, and Selenium (a software testing framework). SolidWorks is an outlier; SolidWorks courses have relatively few users, but almost all the users complete them.
What are people studying? The previous graph includes all users of the O’Reilly learning platform. What can we see if we look at B2B and B2C users separately? There’s a high correlation between completion rates for both types of users on the platform. The Pearson correlation is 0.8, meaning B2B and B2C users move together 80% of the time, but there are some important differences:
- B2C users complete technical courses on topics like Java, web development, and security at a higher rate than B2B users.
- B2B users complete courses in management and “soft skills” at a much higher rate than B2C users. Those courses include topics like design thinking, communication, entrepreneurship, and project management, in addition to courses on Microsoft Word and Excel.
Individual users (B2C) are learning about technologies—perhaps to help in getting a new job or to acquire skills they need in their current job, perhaps to help with their personal projects. Users who are coming from corporate accounts (B2B) behave differently. They’re learning skills that are important in a business environment: communications, teamwork, project management. Also, keep in mind there may be outliers here like SolidWorks, which could be a course an employer requires an employee to complete.
Mapping skills to jobs
Linkedin’s research on mapping skills to jobs led the company to develop a skills genome. Here’s how it’s described:
For any entity (occupation or job, country, sector, etc.), the skills genome is an ordered list (a vector) of the 50 “most characteristic skills” of that entity. These most characteristic skills are identified using the TF-IDF algorithm to identify the most representative skills of the target entity while down-ranking ubiquitous skills that add little information about that specific entity (e.g., Microsoft Word).
Essentially, this approach shows that you can rank skills by how often they show up in job postings. Skipping the math, here’s an intuitive description of TF-IDF:
Term Frequency (TF): Measures how frequently a word (or skill, in this case) appears within a document or job posting. A higher frequency might indicate relevance.
Inverse Document Frequency (IDF): Measures how common or rare a word is across a larger collection of documents (or job postings). Common words like “the” or “and” receive a lower IDF score, de-emphasizing their importance.
We can accomplish something similar by doing topic modeling on O’Reilly’s data. First, we find the top words associated with each topic. Then we use zero-shot classification to map the topics to jobs. That process yields results like these:
Cybersecurity professional:
– Matched Topic 1: [‘kubernetes’, ‘ckad’, ‘developer’, ‘application’, ‘certified’] | Score: 0.976
– Matched Topic 2: [‘security’, ‘professional’, ‘certified’, ‘systems’, ‘information’] | Score: 0.918
Technology consultant:
– Matched Topic 1: [‘kubernetes’, ‘ckad’, ‘developer’, ‘application’, ‘certified’] | Score: 0.579
– Matched Topic 2: [‘azure’, ‘microsoft’, ‘az’, ‘fundamentals’, ‘900’] | Score: 0.868
– Matched Topic 3: [‘linux’, ‘gpt’, ‘artificial’, ‘intelligence’, ‘go’] | Score: 0.623
– Matched Topic 4: [‘learning’, ‘machine’, ‘deep’, ‘design’, ‘driven’] | Score: 0.527
The job “cybersecurity professional” requires skills in Kubernetes (including CKAD certification), along with security skills. A job as a technology consultant requires a broader group of skills: cloud development, linux, AI, and more. “Technology consultant” doesn’t match to topics as sharply as does “cybersecurity professional,” but it still gives us a good starting point.
After a bit of data cleansing, we can invert this mapping to find out what jobs are associated with any given topic. For example, take the titles of courses, then map them to topics, then take the topics and map them to job titles. For example in Topic 1, the skills “AWS” and “cloud” map to the job titles cloud engineer, AWS solutions architect, and technology consultant. This result is exactly what we should expect, showing that this approach to discovering the labels of skills mapping to jobs has merit.Topic 1 (AWS, cloud):
Topic 1 (AWS, cloud):
- Cloud Engineer
- AWS Solutions Architect
- Technology Consultant
Topic 2 (Python, AI design):
- Machine Learning Engineer
- AI Software Engineer
Topic 3 (Software architecture):
- Software Engineer
- Software Architect
Topic 4 (Kubernetes, developers):
- Platform Engineer
- DevOps Engineer
Topic 5 (Java development):
- Back-end Developer
- Full-stack Developer
Topic 6 (Microservices):
- Back-end Developer
- Platform Engineer
Topic 7 (Security systems):
- Cybersecurity Analyst
- Information Security Engineer
Topic 8 (Microsoft Azure):
- Cloud Solutions Architect
- Azure Developer
Topic 9 (Linux, AI):
- Machine Learning Engineer
- AI Engineer
Topic 10 (Deep learning):
- Machine Learning Engineer
- Data Scientist
Topic modeling can play an important role in identifying job skills based on the topics learners consume. This could certainly be used by educational institutions to give them a competitive advantage. It certainly is used by companies like O’Reilly, which provide training services to individual and corporate customers. But more importantly, it provides valuable information to HR departments about the skills they need to hire for.
For institutions that can make use of this data, it serves as a competitive advantage. It tells them what roles the topics they teach are preparing the students for, and can help them plan curricula that are more relevant to the needs of industry. A university could use this analysis to look at external trends along with internal course popularity. Students may have unique intuitions about what skills they need based on job interviews and internships. Analyzing alumni data could show what job titles their alumni have had, which could be compared with the courses those alumni took while enrolled.
The role of industry
What does industry need? The course completion data shows that students from our corporate clients are looking for soft skills like management, communications, and product management in addition to technical skills. While this might reflect students’ desires to “get ahead” rather than corporate needs, companies are aware that good communications and management skills are essential and not taught in degree programs. And, let’s face it, everyone wants product managers.
Topic modeling shows that corporations are looking for cloud skills, software architecture (a more senior skill to aspire to), AI skills, Kubernetes, Java, Python, microservices, security, and Linux. Except for AI, Java, and Python, it’s difficult or impossible to find courses on these topics in college or university CS departments. We won’t name names, but we challenge you to do your own research. Most of the schools we looked at offered one or two courses on cloud computing (though nothing on specific cloud vendors); we were unable to find any university that offered courses on microservices or Kubernetes, though no doubt some exist. If you take up our challenge, we suggest that you look at the course offerings in your state’s flagship university, one of its second-tier universities, a community college, and two private institutions (one prestigious, one not). The less prestigious schools are more likely to provide training in specific job-related skills.
If colleges and universities don’t provide training on skills that are important to industry, who will? Responsibility would seem to fall squarely on the shoulders of industry. If you can’t hire people with the skills you need, hire good people and train them. But is training available on the job? Too often, the answer is no. Why is that?
An increasing number of companies are waking up to the need for corporate training programs, but in doing so, they’re going against the last few decades of corporate thinking. For years, the incentives have been wrong. Stockholders want to see the price of the stock increase and pressure executives to use buybacks and layoffs to maximize their stock’s near-term value, often at the expense of long-term thinking. In The Man Who Broke Capitalism, David Gelles notes,
Before [Jack] Welch, corporate profits were largely reinvested in the company or paid out to workers rather than sent back to stock owners. In 1980, American companies spent less than $50 billion on buybacks and dividends. By the time of Welch’s retirement, a much greater share of corporate profits was going to investors and management, with American companies spending $350 billion on buybacks and dividends in 2000.
Training is an investment in the company—and it’s a kind of investment that has gone out of style.
However, forward-thinking companies realize that an investment in upskilling their employees is a critical part of long-term strategic thinking. Running a company as lean as possible to maximize short-term profit has dire effects on training: If expenses are cut to the bone, companies can’t help their staff keep up with changes in technology, nor can they prepare recent college graduates to make the transition to the “real world.” In turn, a workforce that lags behind current technologies leads to poor long-term outcomes. A staff that falls behind the curve or never makes it to the curve to begin with will have trouble developing successful products for the future. Shortchanging training only leads to a company that underperforms in the long term.
Historical evidence supports the value of skilled apprenticeship. Internships can be small “tiger teams” that allow students to focus on specific problems with a mentor. Although we don’t hear much about apprenticeships in the 21st century, internships (and even PhD programs) share many aspects of apprenticeship. Apprenticeships are an ideal way to bring recent college graduates up to speed on skills they need. They are less applicable for more senior employees who need to sharpen their skills or learn new ones as the industry evolves. It’s important to remember what senior employees gain from mentoring junior employees. When done well, mentoring exposes the seniors to new ideas from their students. It requires them to think through everything they already know; talking and explaining solidifies their own knowledge.
Many companies provide in-house training programs through products such as the O’Reilly learning platform. Products like these can be integrated with the company’s own learning management system (LMS) to create custom curricula depending on their staff’s needs and track progress through the learning program. This kind of solution works well for both senior and junior employees: A senior developer may only need to get up to speed on a few topics of interest, like AI, while a new hire might need to fill in basic knowledge they didn’t get in school.
Critical thinking presents different issues. Companies in which everyone is indoctrinated with the marketing literature and the annual report eventually fail; they’re blindsided by new developments because they can’t think outside of their boxes. Critical thinking isn’t tied to any specific topic or skill, like microservices, but it can be learned in any context. Recently, our learning platform has begun to introduce options for interactivity, including interactive quizzes, coding sandboxes and labs where you can try out ideas, and challenge exercises that test new skills. All of these learning tools help teach critical thinking. Critical thinking skills can also be developed by reading books, writing about what you learned, and participating in study groups. Another key to critical thinking will be valuing teaching as such—the kind of patient teaching or mentoring that doesn’t revolve around grades or student evaluations but that understands that all teaching is a process of exploration. To build critical thinking skills, companies need to go beyond providing courseware. They need to build a culture where all ideas are respected, a culture that encourages discussion, exploration, and failure.
The need to train, upskill, and reskill job seekers isn’t being fulfilled. Universities alone aren’t enough to meet the demands of a changing workforce. There are no shortcuts. Learning requires doing; it can be messy, stressful, awkward, and difficult. But without the struggle to learn, there is no future: not for individual job seekers and not for the rest of us who rely on their productivity. For most students, learning is a matter of filling the gap between academic study and pragmatic skills. Platforms like O’Reilly bridge the gap in bringing cutting-edge skills, certifications, and knowledge to students.
“}]] There is a crisis in technical education. The golden road to a career has always been through a college education. However, this “golden road” has developed deep cracks and is badly in need of maintenance. Postsecondary education is rapidly becoming unaffordable, even at public colleges and universities. Tuition has risen at a rate 50% greater Read More AI & ML, Education, Radar Column, Deep Dive