When I started graduate school in AI back in 2000, it was not at all certain that skills in AI would actually be useful during my lifetime. Back then, career options in Natural Language Processing (my specialty) amounted to teaching in academia or working on a search engine like Google – and that's about it.
No longer
Here are 5 jobs you could land with skills in AI
1 - AI Modeler
Behind every AI system lies a human being who trained the underlying machine learning model. Whether the AI system is a chat bot backed by a large language model like GPT, a fraud detection system, or a self-driving car with a vision model behind the wheel, the underlying AI has to be trained by a human. While the official title of an AI Modeler depends on the company, typical titles include Machine Learning Scientist and Machine Learning Engineer. When I worked in industry, I built AI Models as a Machine Learning Scientist.
An AI modeler's job is 2-fold. First, since training an AI model requires training data, an AI modeler collects, curates, and cleans that training data. Finding sources for the substantial quantity of data current AI models require can be a serious undertaking. Beyond simply the volume of data, AI models also need that data to be clean and correct. Machine learning models find patterns in their training data and then apply those patterns to make predictions about new sets of data. If the training data contains errors, then the patterns the AI models learn will contain those same errors.
An AI Modeler's second task is to run the code that trains the model itself. Most machine learning training algorithms have a variety of settings (called hyperparameters) that affect the model that is built. And there is usually no way to know before hand what the best values for those hyperparameters is -- Instead, the AI Modeler must:
Apply their knowledge of the underlying machine learning theory to hypothesize hyperparameter values that are likely to work well,
Draw on experience in building similar models in the past, and
Experiment: Build machine learning models with a variety of settings and evaluate the performance of the resulting models. Pure science!
When building AI models, the most exciting moment is when you measure the model accuracy after an experiment and finally learn if all your hard work collecting data and adjusting training parameters paid off!
2 - AI Operations Engineer
While an AI Modeler can build a machine learning model, getting that model into the hands of actual users is another matter entirely. AI Operations Engineers build and support the infrastructure that serves AI results to end users. AI Operations Engineers must both be solid software engineers and have knowledge of how machine learning models work: Storing, retrieving, and serving AI models places a different set of demands on the backend than storing, retrieving, and serving raw data does. And since most real world AI systems live in the cloud, AI Operations Engineers must be comfortable working in cloud computing platforms like AWS, Google Cloud, and Microsoft Azure as well.
AI Operations Engineers also monitor the performance of AI systems in the wild. End users are unpredictable. The types of queries that well-meaning users submit to live systems vary constantly, while more devious users may try to break an AI deliberately. AI Operations Engineers put deterministic safeguards around the inherently non-deterministic AI systems to make sure a live product can't go off the rails.
One of my most successful machine learning students came to me with a software engineering background. With his existing skills, learning just the fundamentals of AI enabled him to transition into a AI Ops role where he supported a team of AI Modelers. From time to time the team would have him do small modeling tasks himself. Recently, after several years of hard work, my student interviewed for and won a position as a full Machine Learning Scientist himself -- Congratulations Kumar!
3 - Domain Specialist
Over the past few years, AI techniques have been successfully applied to a wide array of fields. From finance, to supply chain management, to translation services, to medical diagnoses. Each of these domains need humans who are both experts in their field and who are simultaneously comfortable with AI.
Humans with domain knowledge and AI sangfroid can successfully convert the output of AI systems into working products. They are the business experts who build tools to not merely predict where the weak link in the supply chain is, but to then strengthen that weakness. They are the Doctors who use AI assisted diagnoses to craft individual medical plans with their patients. The rise of AI does not mean the end of human expertise, but rather a multiplication of its influence.
4 - AI Program Manager
Building a robust AI system requires a small army of human workers. Someone needs to coordinate this work. That someone is an AI Program Manager. Program Managers must have excellent organizational skills. To compete in a crowded landscape, products must meet their technical requirements while simultaneously meeting their milestones in a timely manner. Herding humans toward that goal is never an easy task.
At Amazon I worked with multiple skilled technical program managers to help build the Alexa personal assistant. At one point I was responsible for deploying a major update to the Alexa model. The deployment process at that time was particularly onerous. The program manager who coordinated my work kept abreast of my progress, urged me to deploy the model as soon as possible, but somehow managed to not come across as overbearing or demanding. An impressive combination of skills.
If you are a technical program manager interested in AI, consider going deep like a student of mine who, after several sessions where we covered ideas from modern machine learning landed a job as a Program Manager at Apple working on AI projects.
5 - AI Business Owner
The recent incredible advancements in AI have enabled so many new applications! Image recognition can speed up product quality assurance in manufacturing. Agents based on Large Language Models can automate repetitive business tasks. 3D asset generation tools can speed up CAD engineering.
Indeed, AI is moving so fast that many entrepreneurs find themselves using AI in their companies when they didn't even plan on it. I've met with CEO's and CTO's in businesses as diverse as entertainment, customer service applications, and real estate who discover they are unexpectedly sitting on a treasure trove of domain specific data perfect for building niche AI products.
Do you have a business idea lying on a shelf gathering dust? Maybe now is the time to revisit and think through how a dose of AI might give you that competitive edge.
Comentários