It's awesome to see you're interested in Artificial Intelligence. I do applied AI research in a space called natural language processing (NLP) for a tech startup. Our company develops conversational agents that use AI and machine learning to answer customer questions (both on the web and over the phone), perform bank transactions on their behalf, and automate other processes (starting loan applications).
I lead most of our research in deep learning which is a particular subset of AI involving neural networks. My research focuses primarily on natural language understanding, which involves training deep learning models to better understand human language (usually in the context of conversation and question answering). While there is no standard day, in general my works can split into the following categories, 60% applied research and experimentation, 20% meetings and organization related things, 10% writing production code, and 10% maintenance (that is fixing bugs, writing documentation, etc) and customer analysis.
Applied research involves reading lots of scientific research papers to stay on top of new academic research and findings that may be useful in helping us develop more intelligent models. Research also involves developing neural network architectures and writing model code (that is taking the mathematical and theoretical representations of various a neural network layers and converting into code that can be run on a computer). Additionally, much of my time is spent cleaning and analyzing our data (historical chat and voice logs of customers interacting with humans and our bots) to create well formatted training data that our models can ingest to more effectively learn. Computers don't understand words that we speak, and much of my work involves developing experiments around different ways to translate human language into numerical representation (long lists of numbers) that can be understood by the computer and are ideally effective for the various tasks we want the computer to do automatically. The final part of applied research involves running many experiments. These experiments explore various language representations, neural architectures (that is building different neural network models) and machine learning models (another subspace of AI involving complex algorithms that based on probability) and seeing how effective these models are on our historical data and in actually solving our NLP problems (like question answering or sentiment analysis).
If an area of research works or is promising, the next I do to is write code to take my research code and convert into a service or feature that we can be used directly in our product. Usually our models are served which means that wrapped up and put on a remote server to make predictions on novel language that comes in through chat bot or voice bot. Programmers on the product side who build out our digital agents will query our AI models when they need to digital agent needs to interpret a user question, provide an answer for a question, or drive an automation process. Additionally, if our production models are acting weird or not working, I will having debug them and fix them immediately so that it doesn't impact our customers.
Outside of research, I also do more traditional data analysis work looking customer data. This involves examining customer conversations and interactions with our bots and voice agents and seeing how well they do in providing the customer value. This data analysis involves a mixture of using summary statistic and visualization techniques to look for patterns, identify interesting findings, and develop visual charts and graphs to communicate our findings. My analysis will involve things like what types of questions is our bot struggling to answer, how much money are we saving our customers when digital agents are successful and many other business related questions.
Finally, like most workers, I have to conduct meetings or participate in meetings. I don't work alone and collaborate with not only other research scientists on my team but also other programmers and non-technical team members and our clients directly. The goal of meetings to communicate projects we're working, brainstorm idea for new research and features, and learn about other work the other team members are doing, as well find ways to provide new value for our customers.
Hopefully that gives you an idea of the many things an AI scientist does. It's worth noting that since I work at a startup, I often do many things both related to my job and as well things related to helping build and grow our business. As for your second question on options in the AI space, there are many opportunities.
In larger organization much of what I do would be separate jobs like the following:
- data engineers: responsible for building pipelines and infrastructure to clean, process, and annotate data that is used by researchers in developing models
- AI research scientist: usually folks coming from academic background in AI and machine learning who run experiments and build models to solve various problems (including computer vision, language, self driving cars, and many other interesting areas) using machine learning and AI approaches
- ML / AI engineer: these folks are software engineers whose responsibility is to take a research model developed by the scientist and put it into production. They often focus on issues of scale (ensuring the models can handle up to millions of requests and ensure predictions are return in a reasonable amount of time), availability (support simultaneous requests and have backup is things break), and integration (finding ways to incorporate the code into a larger code base)
- data analysts: these folks are more on the business side analyzing customer data and business data for valuable business insights
- data scientist: also folks coming from an academic background though usually in stats, math, and data science and focus building models related to business decision making challenges (e.g. making product suggestions or customer targeting for advertising)
All those roles are not strictly independent, depending the organization and team you may do some or all those things while having different titles like: data scientist, AI scientist, ML engineer etc. And as you can tell all these roles require lots of different skills. So there are many pathways into an AI career. Some folks have phds and masters degrees in things like computer science, physics, linguistics, etc, others have software engineer backgrounds, and some even come from non-science backgrounds like finance and business analytics.
I took a non-traditional path. My undergrad degree was in creative writing and social sciences. I worked several different jobs including project management and business analysis before transitioning to data science and eventually AI and NLP research. Along the way I got my masters in software engineering with a focus on machine learning, and soon will be heading back to school to do a phd in AI. I've spent about 5 years years doing data science and AI research professionally and decided I wanted go back for my phd to dive deeper into foundational research. So there's many different paths into AI and you'll find your own way too.
If you're interested in learning more, I recommend looking into these various roles and career paths to find what resonates with you. Take a class on data science or machine learning online (using sites like coursera or Kaggle) and see if you enjoy it and dive deeper! If you're in college or about to go to college, there are more degree paths for AI careers that your advisor can help you with. Feel free to ask more specific questions and good luck on your journey!
There are many use-cases in daily life to apply AI, and I'll share this podcast in case it is helpful to you (over 100 episodes explaining various use-cases with folks in the industry)
Or search on your podcast app on your phone/computer for other podcasts covering Artificial Intelligence / Machine Learning / Data Science. These topics overlap a lot so you can explore what is interesting to you.
There are variety of exciting roles in the field of Artificial Intelligence (AI).
With AI we have the opportunity to capture, harness and apply intelligence to the constant stream of data flowing through our digital and physical worlds to make better decisions & transformative processes.
You can be business analyst who works to understand the business and consumer problems that AI can be used to transform & use the AI insights to do things differently, create new experiences, new ways of working & new outcomes. You can be a data scientist that gathers data, analyzes data, shapes it and visualizes using Business Intelligence software, a software engineering who develops the AI models and algorithms that helps derive insight from data and even a product manager for the hardware that helps gather data in the physical world.
Wishing you the very best in learning more and potentially pursuing a career in data science.
Riste recommends the following next steps:
Dr. Sudeep Mohandas
Dr. Sudeep’s Answer
As AI is something that will continue to be disrupted and challenged especially during this Pandemic of Covid 19, one will find AI to explore possibilities and opportunities never thought of before. In other words, it will be a world so much to do that will benefit people and planet.