What trends might emerge in the fields of neurotechnology and neural data in the years to come, and in light of those considerations, what projects might be most useful for an organization like NeuroTechX to pursue?
This post will suggest the following:
- the NeurotechX community is better suited to creating tools and analytics for neural data, rather than trying to create hardware or collect data ourselves.
- because we are a distributed network (rather than a concentrated group of full-time researchers), we will succeed most at projects that are modularizable and focused on engineering (as opposed to generating new algorithmic insights).
Framing the Question
When starting any technology development project, don’t impulsively seek to extend the existing technologies. Start with the problem that you want to solve, and then go find or create the tools you need to solve it.
Since we have limited time in life, it’s a good policy to choose problems that are important, neglected, and tractable. In other words: What is the most impactful area that isn’t otherwise going to be solved, where you can make progress given your current resources and constraints?
Notice there is a mapping between structures or types of problems and the types of people or organizations that are best fit to solve them. There are an analogous set of questions in computer science, where different algorithms perform better depending on the structure of the data.
Current Resources and Constraints
The NeuroTechX Community has varying levels of experience in machine learning, hardware, software development, neuroscience research. In our member survey, out of 69 respondents, 45% reported they had completed a graduate program, and another 16% reported being in a graduate program. This is one visible marker of advanced scientific/technical training, but certainly isn’t the only one! 22% also reported that “[they] learned most of what [they] know outside of school.”
NeuroTechX currently has 2,300+ members on Meetup, 500+ members on Slack, and active groups in 10 cities in five countries around the world. Through this network, we are able to draw unique talents and perspectives from many universities and big cities around the world.
In any group, projects are rate-limited by the number of people with the capacity and willingness to be project initiators. Based on the members survey, we were able to identify some major constraints on the NeuroTechX members:
- Time availability: Most people in the community have day jobs, or are full time students.
- Motivation and interests: Members won’t just work on any project because it’s useful on paper. They show a preference for projects that are tangible, fast feedback loops, and result in a cool demo within a few months
- Distributed locations: The international nature of NeurotechX means that we cannot concentrate our 50 top people in the same workspace to work on a project.
Trends in the Neurotechnology Landscape
Given the resources and constraints, what are our comparative advantages? What projects could a group like ours do extremely well, both by developing strong internal abilities, and by anticipating projects that other organizations wouldn’t do? What has already been tried in the space? Where have people found problems and solved them, or failed at solving them? What problems or opportunities still remain?
What’s already being done well?
Currently, academic labs are much better suited for data collection than the NeuroTechX community. The infrastructure to store test animals, perform small neurosurgeries, house large equipment, etc. requires centralization of data collection. The hardware for neural recording is expensive and too complicated to build in-house. Human or animal experiments require coordination with institutional review boards.
Thus, collection of the interesting neural data (through electrophysiology, fMRI, calcium imaging, and other invasive methods) is mostly limited to academic groups. (EEG data is more accessible, but is unlikely to lead to future breakthroughs in neuroscience.)
That might be OK, because researchers often report being bottlenecked at processing the data.
Next question: What is our comparative advantage?
What types of challenges will we be able to solve better than any other organization?
It depends on the class of problem. For problems that require conceptual breakthroughs, academic research labs are probably better suited. We have more of an advantage at problems that are sequential and modularizable.
Given the talented and distributed population of members, two general candidates are online software (as opposed to hardware development) and running experiments (given the potential to run human subjects in parallel).
Which types of projects would be best addressed by the NeurotechX community?
Design and Synthesis problems
Some projects in neural data analysis are bottlenecked by algorithm development. These are the types of problems that are typically solved by collaboration between a few experts who can hold many complex pieces of information in their head and derive new algorithms for processing the data.
Perhaps algorithm-bottlenecked projects could be tackled via crowdsourcing if the problem was framed in the right way and the right incentives were in place to solve it. (For example, see the crowdsourced solution to Erdos math problem.)
However, solving these analysis problems can be a matter of “iteratively searching for the right question to ask”, which requires insight and consistent testing, and thus perhaps best done by people who can devote significant hours to it (i.e. full-time researchers).
Call these “design and synthesis” problems: the kinds that have historically been solved by a relatively small group of experts rearranging complex pieces of information in their minds.
Design and synthesis problems: Specialized working groups?
One strategy to be able to reach into the space of “design and synthesis” problems would be to have small cells or working groups clustered around specific expertise. These working groups could be centered around things like:
- Algorithm design: generating new, better, algorithms for analysing data
- Experimental design: designing experiments and testing hypotheses that would otherwise not be explored
Of course, other potential working groups exist.The NeuroTechX community can discuss what these working groups could be.
Sequential, Modular Problems
These are problems that can be solved by more people making more linear, iterative contributions. Historically, what have large groups of open-source hackers been good at?
- Toolkits, open source frameworks
- Problems that are mostly technical, or can be well-described early on with engineering specs (as opposed to generating new designs or algorithmic insights)
- Modular problems, where there are not too many interdependencies between development paths, so stop and go progress is OK
What are the conditions necessary for sequential, modular problems to be solved by specialized sub-groups in the NeurotechX community?
Gary Pisano addresses this class of problems in a survey of the biotechnology sector:
“[These are] the conditions that enable a market for know-how to provide the requisite degree of integration among specialized firms….
First, it helps to have modular designs with clearly defined, codified interface standards. modularity enables a big problem to be broken into a set of quasi-independent subproblems. with clearly defined interface standards between the subproblems, modularity reduces communication and coordination costs between organizations working on different pieces of the puzzle. Software and semiconductors tend to be highly modular. Thousands of individual Linxus programmers scattered around the globe can contribute to the Linux operating system because of its modular structure. They do not need to communicate with one another; as long as they follow the well-documented Linux protocols and standards,their code will be compatible. In software, well defined, broadly accepted platforms (like the Microsoft operating system) provide a mechanism to integrate the efforts of a diverse group of specialist.
A second requirement is technology that can be communicated and transferred in codified form (e.g. blueprint, lines of code). Software is completely codified (although the expertise to develop software is not). Semiconductor designat have now created almost completely via computer are also highly codified. once technology can be codified in digital form, it can be transferred around the world at trivial cost and time. This allows developers from dispersed parts of the world to collaborate and to transfact and transfer their intellectual property at relatively low cost.”
– Science Business, Gary Pisano, p.150
Given the shape of the distributed NeuroTechX community, what outstanding challenges could we tackle? What are the most important types of sequential, modular problems? Is there a list of outstanding problems in neuroscience and neurotechnology?
One way to start develop insight into project organizational models is to look at other open-source neural data projects:
Thunder: large-scale analysis of electrophysiological and calcium imaging data
OpenWorm: computational model of the brain and body of a C.elegans nematode
Open Source Brain: A list of other open-source neural data initiatives
Over the next few weeks #moonshots working group will systematically explore these questions, and generate a list of possible projects to pursue.
Want to join the conversation?
Check out the #moonshots channel on the NeuroTechX Slack or email stephen[at]neurotechx.com