AI and Machine Learning for High School Students: What a Harvard Professor Told Parents
Key Points
Dr. Fangxu Xing, a Harvard professor with 15+ years in AI, machine learning, and computer vision, joined our recent webinar to explain what students actually need to learn to become machine learning engineers.
He prefers the term "machine learning" over "AI," because today's models run on probability and pattern matching rather than reasoning, and he walked families through why that distinction matters.
The math is narrower than it looks. Students need a specific slice of linear algebra, calculus, and probability, not full semesters of each.
About half of working AI engineers came from outside computer science, including economics, finance, life sciences, and the humanities. One of his own students moved from an arts background into medical image processing.
What he looks for in students is diligence between sessions, not raw talent, and what he looks for in applicants is a portfolio he can open and read.
His 12-week Scholar Launch program, Fundamental AI and Machine Learning, starts in late September 2026 and is built for students with no coding experience.
Short answer: In a Scholar Launch webinar, Harvard professor Dr. Fangxu Xing explained that machine learning is learnable much earlier than most families assume, because it rests on a narrow slice of college math rather than the full curriculum. He argued that machine learning engineer is becoming the standard technical role, that admissions readers and employers both want to see a portfolio of real projects, and that students should start building one now. His 12-week program, Fundamental AI and Machine Learning, begins in late September 2026.
In our recent webinar, Dr. Fangxu Xing walked families through his own path from computer vision researcher to Harvard faculty, what separates a machine learning engineer from a software engineer, and how a student with no coding background gets from zero to a working model. He has trained more than 800 students over 11 years, and his lab work applies machine learning to medical images: CT, MRI, ultrasound, and X-ray. Here is what he covered.
Why He Calls It Machine Learning, Not AI
Dr. Xing opened with a piece of vocabulary he cares about. He avoids the term artificial intelligence, and he was direct about why: the phrase raises money, but it oversells what the technology currently does.
His definition of machine learning is closer to teaching than to intelligence. You show a child a dog, then another dog, and eventually the child recognizes a dog they have never seen before. A machine has no brain, so the algorithm becomes the brain, and the math is what lets the machine find the pattern.
The limitation he named is the interesting part. Every current machine learning algorithm, in his words, is built on probability. That is also why models cannot perform real logical deduction. Asked to reason through a problem, a large language model produces the continuation with the highest probability given everything it has read, which reads like deduction without being deduction. Real artificial intelligence, in his view, would require a framework nobody has invented yet, and whoever invents it will overturn the one we use now.
For parents worried about machines developing their own minds, his answer was that the day is not close. For students, the point is more useful: what looks like a mysterious black box is a set of methods a motivated high schooler can learn.
The Math Is Narrower Than It Looks
The slide that scares people, he said, is the one listing prerequisites: linear algebra, calculus, probability, statistics, a little information theory. He has taught long enough to know exactly which parts of each get used.
His argument is that you need a shallow layer of linear algebra to do deep work in machine learning, a little calculus, and a little statistics. He held up the size of his old statistics textbook and pointed out that a student needs a few pages of it, not the volume. He admitted he disliked math himself when he was young, mostly because nobody told him which parts would ever matter.
That is the reasoning behind how he structures a course. Rather than front-loading a year of prerequisites before students touch a model, he teaches the specific mathematics each algorithm requires in the session where that algorithm appears.
Half of AI Engineers Came From Somewhere Else
Dr. Xing showed a breakdown of where working AI engineers studied. Roughly half came from computer science. The rest came from electrical engineering, mathematics, economics, finance, life sciences, and the humanities.
His example was a student he supervised who arrived with a background in the arts and graduated two years later as a machine learning engineer. That student now does medical image processing at the Cleveland Clinic, work very close to Dr. Xing's own.
He also teaches working professionals through an online engineering school, many of them with 20 or 30 years in traditional software roles, retraining because the job market shifted underneath them. His view of that group is that if adults deep into a career can rebuild these skills in a few months, a high school student has both the time and the advantage.
What the Market Is Actually Hiring For
Dr. Xing's framing was that machine learning engineer is on its way to becoming the default technical role, the way software engineer became the default 30 years ago. He described watching companies shift their hiring toward people who understand the algorithms rather than people who mainly write code, since a large share of routine coding can now be produced with AI assistance.
The public data supports the direction, with more nuance on the software side. The U.S. Bureau of Labor Statistics projects employment of data scientists to grow 34 percent from 2024 to 2034, against 3 percent for all occupations, with median pay of $112,590 as of May 2024. [1] Over the same period BLS projects computer and information research scientists to grow 19.7 percent and software developers to grow 15.8 percent, the largest absolute gain of any computing occupation, and attributes much of that growth to AI adoption rather than treating it as a threat to those jobs. [2] Software engineering is not disappearing. The skills premium is moving.
Demand for the specific skills is visible in postings, too. The 2026 Stanford AI Index, built on Lightcast job posting data, found that 2.5 percent of U.S. job postings now request AI skills, up 55 percent in a single year, with Python the most requested specialized skill in the set. [3]
He added a reason companies are staffing this in house rather than outsourcing it to a chatbot: data. Working in a hospital, he cannot upload patient data to an outside platform, so every model he builds runs locally. Any company holding sensitive data faces the same constraint, which is why they need people who can build models rather than people who can prompt someone else's.
Inside a Session: Theory, Then Code, Immediately
Dr. Xing shared his first session as a sample, and the rhythm was the clearest answer to what the program is like.
Week one covers two algorithms. K-nearest neighbors comes first because its mathematics is the most approachable, essentially measuring distance between vectors, taught through examples with cats, dogs, and birds. Naive Bayes comes second, which means introducing the Bayesian equation, which he teaches through colored balls, a COVID test scenario, and the question of how the spam folder in Gmail decides what is spam.
Then the theory stops and the code starts. In the same first week, students open a breast cancer dataset where every row describes a tumor labeled benign or malignant, and apply both algorithms line by line. K-nearest neighbors, the first algorithm they ever learn, predicts at roughly 98 percent accuracy on that dataset. Naive Bayes performs comparably.
Later weeks keep the same shape with different data. Week three uses tree-based methods on telecom customer churn, predicting which subscribers switch carriers next month. Another week predicts whether a Pokemon is normal, epic, legendary, or something rarer, which he keeps in the syllabus because students enjoy it. Others work with bank deposit subscription data and socioeconomic data. Students keep the notebooks after class.
What Makes a Student Succeed
Asked what separates the students who finish well, Dr. Xing's answer was not about talent. His words were that if you are human, you are smart enough. The variable is diligence.
He described the failure mode precisely, drawn from his adult learners: a student enjoys the live session, follows along, understands everything in the room, and does nothing with it until the next one. Each session builds directly on the last, so a concept like overfitting gets used in week six rather than re-explained. The students who succeed re-run the notebook themselves during the week and break it on purpose.
His estimate is that a diligent beginner can build their own models after about a month of serious work.
Why He Pushes Portfolios and Publication
The last question of the night came from a parent asking how early exposure to AI research affects college applications. This is where he was most emphatic.
Admissions and hiring have both moved, he said. Reviewing applicants to his own group used to mean scanning a resume for coursework. Now the first thing he asks is whether there are publications, and if there are none, which is entirely normal at the high school level, he wants a portfolio. Specifically, he wants a GitHub repository he can open. A student with ten projects in it reads differently from a student with one.
He was equally clear about what does not count. Uploading a dataset to ChatGPT and asking for a prediction produces a good answer and demonstrates nothing, which is why every session in his program moves from the theory into Python that students write themselves. Publication is a real possibility for students who get there. He mentioned a high school student who had submitted a paper to SPIE Medical Imaging two hours before the webinar began.
Inside Fundamental AI and Machine Learning
The program Dr. Xing built with Scholar Launch runs 12 weeks starting in late September 2026, with 96+ total program hours across three parts:
36 hours of class time with Dr. Xing, 3 hours per week for 12 weeks
Additional office hours with Dr. Xing for hands-on coding and debugging support
48 hours with the Scholar Launch education team supporting the final project and the written report
The 12-week arc moves from Python fundamentals to a capstone on a real dataset:
| Weeks | Focus | Covers |
|---|---|---|
| 0–2 | Foundations | Introduction to AI and machine learning, Python for AI, data structures, NumPy, Pandas, Matplotlib |
| 3–4 | Core machine learning models | KNN, Naive Bayes, linear and logistic regression, decision trees, support vector machines, bias–variance tradeoff, ensemble methods, introductory neural networks |
| 5–6 | Mathematics and optimization | Intuitive calculus for machine learning, gradient descent and optimization algorithms, correlation, Gaussian models, the Central Limit Theorem |
| 7–9 | Data skills and unsupervised learning | Exploratory data analysis, data cleaning, feature engineering, dimensionality reduction (PCA, t-SNE), clustering (K-Means, hierarchical, DBSCAN) |
| 10–12 | Evaluation and final project | Accuracy, precision, recall, F1, ROC/AUC, regularization, hyperparameter tuning, and a capstone using a real-world dataset in finance, healthcare, or social data |
Students finish with weekly coding exercises, an individual end-to-end AI project, a written technical report or paper, a formal program evaluation from Dr. Xing, and a possible letter of recommendation from him. Whether a student produces a report or a full scientific paper depends on where they land by the end, and our education team pushes students toward the higher end of what they can do.
The skills the program targets are the ones Dr. Xing described as the working requirements of a junior machine learning engineer: building and evaluating models independently, understanding how AI systems behave in real applications, reading and modifying an existing machine learning codebase, and completing a project end to end on real industry data.
From there, students go in one of two directions. The academic path uses the program as groundwork for deep learning, computer vision, natural language processing, and AI research. The career path points toward internships and entry-level roles such as Junior AI Engineer, Junior ML Engineer, ML-focused Data Analyst, and AI Research Assistant.
Basic algebra is the only prerequisite. No coding or machine learning background is required. The cohort is capped at 20 to 30 students.
Apply for Fundamental AI and Machine Learning
Admission runs in two steps. Students complete an intake form on our site, or request one from our team, and shortlisted applicants are invited to a short online interview with an admissions officer to confirm the program is the right fit for where the student is now.
This is the only Scholar Launch cohort this term taught directly by a Harvard professor with this much contact time, and seats are limited. Read the full program details or start your application. Students who want a subject outside this cohort, or a schedule built around their own commitments, can look at the Custom 1-on-1 Research Program instead. To talk through fit before applying, schedule a complimentary consultation.
Frequently Asked Questions
Does my student need coding experience?
No. Basic algebra is the only prerequisite, and Python is taught from the ground up in the first two weeks. Dr. Xing teaches the specific mathematics each algorithm needs in the session where it appears rather than requiring prerequisite coursework.
How much math is really involved?
A narrow slice of linear algebra, calculus, probability, and statistics. Dr. Xing's position is that a shallow working knowledge of each is enough to do advanced modeling, and that most of a standard statistics course is not used in practice.
How much time does it take each week?
Sessions with Dr. Xing run about 3 hours per week for 12 weeks, and students should plan additional hours for coding exercises, office hours, and project work.
Is this only for students planning to study computer science?
No. Dr. Xing's own breakdown showed about half of working AI engineers coming from outside computer science, and capstones can be built around finance, healthcare, or social data. Several students use a program like this to decide whether the field is right for them before committing to a major.
What does my student have to show at the end?
An end-to-end AI project on a real dataset, a written technical report or paper, weekly coding exercises, and a formal evaluation. Dr. Xing recommends students publish the project to a public GitHub repository, since that is the first thing he looks at when reviewing applicants.
Can a high school student really publish research in this field?
It happens. Dr. Xing mentioned a high school student who had submitted a paper to SPIE Medical Imaging shortly before the webinar. Publication is not part of the standard program outcome, but it is available to students who go further than the requirements.
What is the difference between AI and machine learning?
Machine learning is the set of methods being used today, built on probability and statistics, in which an algorithm learns patterns from data. Artificial intelligence describes the broader goal of machines that reason the way people do, which Dr. Xing considers a long way off.
References
U.S. Bureau of Labor Statistics. (2025, August 28). Data scientists. Occupational Outlook Handbook. https://www.bls.gov/ooh/math/data-scientists.htm
U.S. Bureau of Labor Statistics. (2026). Artificial intelligence, information technology, and employment, 2024–34. The Economics Daily. https://www.bls.gov/opub/ted/2026/artificial-intelligence-information-technology-and-employment-2024-34.htm
Lightcast. (2026, April 13). Four takeaways from the 2026 Stanford AI Index.https://lightcast.io/resources/blog/stanford-ai-2026
