How do I Pursue a Career in Brain-Based AI?

Do you have any insights into a career in brain-based AI?  What do I need to study if I’m interested in developing Numenta’s technology? Where can I learn the necessary skills?

These are some of the questions we receive quite frequently from students interested in pursuing a career in brain-based AI. Not too long ago, the AI/ML conversation was largely centered around making machines that can outperform humans at specific tasks by identify patterns in data. These machines cannot think and plan like humans do, nor are they flexible in any way. Now, many researchers are looking to the brain for insights to address these limitations and to capture aspects of human intelligence that existing networks have failed to do.

While there may not be a single answer or one clear path, I’ve gathered a list of tips and advice from our research team, designed to help anyone kick off their professional journey with brain-based AI.

1. Master the fundamentals of brain-based AI

Many people wonder which is more important: pursuing a degree in neuroscience or a degree in computer science. Although brain-based AI research focuses more on neuroscience compared to existing AI/ML research, it still requires computer science to formalize and test ideas. Studying general AI/ML topics and honing your computer programming skills are great ways to start. As you continue, you can focus on the subfields that are more likely to be useful for brain-based AI.

AI/ML research involves quite a bit of math and programming, so proficiency in the following areas is a plus:

• Algorithms and data structures: AI/ML research requires excellent algorithmic problem-solving skills and the ability to handle and understand data.
• Programming languages and coding: Python is one of the most popular programming languages in the brain-based AI space due to its flexible and exhaustive range of frameworks and libraries. C/C++ and Java are also commonly used.
• Discrete mathematics: When things become more theoretical, having a strong math background can make a huge difference.
• AI/ML fundamentals such as basic concepts like deep learning, knowledge representation, reinforcement learning, computer vision.

And ideally,

• Basic neuroscience knowledge such as the function of cortical columns, pyramidal neurons, and grid cells.

Conversely, if you have mastered the digital domain and are struggling with the biology, don’t be discouraged. Many AI/ML researchers started this way. You can fill in the gaps in your neuroscience knowledge in many ways by reading textbooks and papers, attending courses, etc. It’s generally easier to learn neuroscience as you go than to know neuroscience and learn to build programming and AI/ML foundations.

Here are a few helpful resources to get you started:

• Introduction to Computer Science and Programming in Python
• Introduction to Algorithms
• Mathematics for Computer Science
• Introduction to Neuroscience
• Companion paper to A Framework for Intelligence and Cortical Function Based on Grid Cells in the Neocortex

2. Learn about the different subfields

Once you master some of the fundamentals, find subfields that interest you. It helps to have an overall comprehension and it’s worth exploring more subfields at any stage of your career. You never know if you’ll come across a creative solution to your most challenging problem!

What you can do:

• Read review papers and blogs: The AI/ML world moves quickly and it’s easy to miss new ideas and progress. You can find many blogs that do a good job summarizing key ideas and hot topics on platforms like Medium or Lillian Weng’s blog. If you are looking for peer-reviewed work, you can find review papers that give an overview of interesting work in the area.
• Attend conferences, workshops, and talks: Major conferences such as NeurIPS, ICLR, and Neuromatch, offer talks and workshops that dive into particular themes. Meetups such as Brains@Bay are also a great way to explore new topics and find review papers.

Quick tip! Look for workshops at main conferences to get a good grasp of a particular theme you are interested in. They often have an interesting discussion panel at the end on the current state-of-the-art research and open questions in that subfield.

• Audit online courses: In this day and age, accessing information is as easy as lifting a finger. Many courses taught in top universities such as MIT, Stanford, and Harvard make their videos and slides available online. It is helpful to watch the lessons and try the exercises provided to gain an understanding of the topic you’re interested in.

 Here are a few brain-based AI subfields to get you started:

• Sparse representations
• Sensorimotor learning and robotics
• Predictive processing
• Grid cells and cortical columns

3. Get hands-on experience with personal projects

It’s equally important to put these skills to the test – and to acquire new ones – with personal projects. If there’s a topic that interests you, try pursuing small projects or experiments. These projects help you learn skills you didn’t know you needed and allow you to learn by doing, rather than learning about it through a textbook.

Quick tip! When doing a personal project, make sure you understand all your results. If you get a result that you didn’t expect, dig deeper and figure out why you are getting this result. When the result is as expected, make sure you can effectively explain it to someone else. You can also perform additional experiments and create many plots to gain a deeper understanding of what your algorithm is doing.

Sometimes it’s hard to settle on an idea for a project. A great starting point is to look at open-source projects and try to extend them. For instance, Numenta has an open-source project called NuPIC (the Numenta Platform for Intelligent Computing) and many people have developed solutions based on our software.

4. Stay up to date with state-of-the-art applications

The field of AI/ML evolves very fast and the brain-based approach to computing is becoming increasingly prevalent. With hundreds of papers published every month, methods that were state-of-the-art a year ago are already obsolete so get into the habit of reading new papers regularly. Staying up to date with current applications can help you parse new ideas and make decisions faster. You can follow specific brain-based AI journals such as Frontiers in Neural Circuits or Frontiers in Neurorobotics.

Reading and understanding papers can be a time-consuming endeavor and it is nearly impossible to read all of them. Here are a few valuable resources that can help you:

• Newsletters: AI/ML newsletters are a great way of keeping up with the latest advancements in the field. What’s better than having a curated collection of news and resources delivered straight to your inbox? Subscribe to Numenta News Digest for the latest developments in brain-based AI. For the latest AI/ML news stories, we subscribe to The Sequence, The Gradient and The Batch.
• Twitter: Twitter is a great place to follow real-time updates from researchers and companies that are working on related areas to you. We frequently discuss our latest research activities on Twitter (@Numenta). Our VP of Research and Engineer Subutai Ahmad is also active on Twitter (@SubutaiAhmad) and often shares interesting papers and news.
• Podcasts and webcasts: Tuning into podcasts and webcasts is a great way to parse through research when you can’t read more papers. They also allow you to dive into cool ideas while running errands or taking your morning stroll at the park. We recommend:
  • ML Street Talk and This Week in ML for interviews with influential figures in the AI/ML space;
  • Yannic Kilcher for in-depth and easy-to-understand paper reviews;
  • Gradient Dissent for the latest industry news.

• Conferences: Talks and workshops at conferences often discuss the current state-of-the-art research and future directions.

5. Put yourself out there!

Showcase your capabilities and understanding by sharing your projects and ideas. The insights and critiques you get from others can often help improve your work. They can expose the areas you need to address with additional experiments and push the boundaries of what you think you know.

Quick tip! Even explaining your work and ideas to members of your family or friends who don’t know anything about your field can be incredibly valuable.

You can publish your code on GitHub, publish research articles and preprints and write blog posts. You can go to conferences, give talks, and present posters to different audiences. The possibilities are endless.

6. Don’t be afraid to seek advice from others

Joining a community that can help, support and inspire you may make your learning journey easier. I encourage you to join Numenta’s open-source community forum (HTM Forum) to ask questions, engage in discussion, and share your work.

Twitter can also be a useful tool to connect with other academics in your field, ask questions and publicize your work.

7. Look into related fields for inspiration

Biological neural networks are heavily inspired by the workings of the brain and offer a path to machine intelligence. For instance, Numenta researchers found that the activity of dendrites in the brain provides great insight into creating machines that can learn without forgetting previously learned information.

You can read books, review articles or textbooks from other related fields such as neuroscience or psychology to get inspiration. Those fields provide insights into how learning works in biological organisms and what biological intelligence can do.

Quick tip! Textbooks may sound daunting, but they can help in learning the foundations and having a birds-eye view of a particular topic. You can go through them slowly, anywhere from five to twenty pages a day, so it doesn’t require a big commitment and it helps with digesting the ideas better over time.

8. Read books to get a fresh perspective

Books are always a good resource to get a sense of the breadth of ideas and depth of the AI/ML field. Most authors are immersed in the field and have a good understanding of the area they’re writing about.

I recommend starting with A Thousand Brains by Jeff Hawkins. In the book, Jeff explains what the brain tells us about intelligence, what components are needed to achieve machine intelligence, and how our understanding of the brain can impact the future of AI and humanity.

Most importantly — Don’t get overwhelmed!

It’s easy to get overwhelmed by all the resources at your disposal. Even after a few months of learning, you might feel like you’ve barely scratched the surface. This is normal, you will definitely miss things and most of your ideas will not be new. But don’t be discouraged. Establishing a foundation for a career in brain-based AI takes a lot of time, effort, practice, and a commitment to lifelong learning (which is part of the fun!).

Whether you’re an experienced computer scientist, a fresh college graduate, or a complete newbie, brain-based AI provides a host of exciting opportunities with important problems to solve. Naturally, the more tools you have under your belt, the more ways you’ll be able to tackle those problems.

At Numenta, we’ve laid a solid neuroscience foundation upon which intelligent machines will be built, and we’ve made significant strides in modeling them in software. We believe creating machines that closely match the design principles of our brains will benefit humanity in many ways — machines can take on jobs that are too risky for humans such as cleaning up toxic spills or mining coal, lead to many medical breakthroughs such as drug discovery, and unlock vast amounts of new knowledge of the universe by traveling to places we cannot reach. Building these intelligent machines will require people like you to study it, champion it and experiment with it.

To learn more about the difference between current AI approaches and Numenta’s brain-based approach, read our blog “The Path to Machine Intelligence: Classic AI vs. Deep Learning vs. Biological Approach.”