James Isaac Rosen-Birch

About Me
Repeat startup founder, former
unicornRoivant Sciences [$7.3B exit]
and
public companyAxovant Gene Therapies
chief of staff, and scientist at the intersection of technology and human society. Worked professionally in neuroscience, machine learning and artificial intelligence, socioeconomic anthropology, organizational science, and environmental technology.

Deep interest in complex systems, intelligent machines, and organizational and
sociotechnicalintegrated human-machine
systems that augment both individual and collective human capabilities. Currently developing a visual programming language for organization design, in order to make sociotechnical systems easy to understand for both people and machines.

Occasionally also write and publish professionally, including, most recently, a professional postmortem of the October 7th crisis in Israel-Palestine from an organizational science perspective (now out in New Lines Magazine). It has also been translated into Italian (published in Internazionale).

Education
  • CÉGEPpre-university college in Québec
    :
    Liberal ArtsWestern history (ancient to modern), art history, philosophy, poetry, classical literature, sacred writings, research methods, calculus, principles of math & logic
  • University: neuroscience, social-cultural and linguistic anthropology, some electrical engineering,
    cyberneticsthe study of adaptive feedback and control systems
    , and
    semioticsthe study of signs and symbolic systems
    . Graduated Hon. BSc from the University of Toronto in 2014.

    The underlying logic was that I wanted to understand the biological basis of how we navigate and experience the world (neuroscience), how culture and society tints those base inputs (social-cultural and linguistic anthropology; semiotics), and then be able to apply that knowledge to the design of intelligent machines (electrical engineering) and feedback systems (cybernetics) to solve human problems. That perspective has guided much of my work in the years since.
Work (slowly adding stories to this)
  • Early career spent developing technological adaptations to climate change with
    Environment CanadaAdaptations and Impacts Research Group, UToronto Section
    .

    Focus was primarily on designing
    living machinesintegrated biological and mechanical systems to solve problems eg. energy generation and waste treatment
    and closed-loop systems to process waste, filter water, generate energy, conduct bioremediation, and improve public understanding of the feedback loop between human behaviour and the health of the natural environment. I got to work on
    green roofs
    ,
    living walls
    ,
    microbial fuel cells
    , mycological bioreactors,
    biofiltration systems
    , and various
    combination systems
    . The overarching goal of my team was to find ways to bring the needs of a modern, prosperous society in alignment with the development of a thriving and sustainable ecology. This was where I first kindled my love of cybernetics and systems thinking, as much of the work involved the design and management of open systems, and systems-of-systems.

    Unfortunately, this work happened at a time when climate change denial was still rampant, and was cut short by a (Conservative) government that sought to destroy public sector science and considered sociology a dirty word. Otherwise, I probably would have stayed in the space longer.



  • I trained as an anthropologist, and was lucky enough to fieldwork studying the impact of World Bank policy on smallholder
    palm oil
    farmers in rural
    West KalimantanIndonesian Borneo
    .

    Investigated various forms of organized crime (including a transnational pyramid scheme and prostitution ring) preying on local communities and the growing social inequalities and debt-based desperation linked to broader agricultural policies and transnational conditionality agreements.


  • Worked for several years developing a neuroimaging protocol for diffusion tensor MRI of brain cancers of the
    seventhfacial nerve
    and
    eighthvestibulocochlear nerve
    cranial nerves. The goal here was to identify how tumour growth
    distorts the path of major nerves
    before a patient undergoes surgery, so the neurosurgeons could excise the tumour without damaging healthy nerve tissue. The lack of visibility neurosurgeons face can result in a number of side effects, from partial facial paralysis to deafness in one ear and balance issues. Being able to identify in advance if the nerve is salvageable and, if so, how to avoid damaging it while excising the tumour is therefore of great use to many neuro- and oral/maxillofacial surgeons. The work was so successful we ended up applying the same technique to identify the auditory nerve root in deaf patients to improve the placement of auditory brainstem implants (ABIs). (I also got to observe neurosurgeries firsthand, which was very cool as a young neuroscience grad.)

    A couple notes for clarity:

    1. A typical MRI works by sending electromagnetic waves into sequential 'slices' of tissue -- in this case, brain tissue -- and then counting how long it takes for the energy to bounce back to the machine. Differences in the feedback time tell you what type of tissue exists in each spot (eg. fatty tissue is faster than watery tissue). Diffusion tensor MRI takes this one step further, using sequences of tissue types to infer the direction and flow of nerve bundles (or 'tracts'). This then allows you to visualize the physical structure of groups of neurons in the brain (a practice called 'tractography').
    2. For any readers not familiar, ABIs are 'brain chips' placed on the brainstem that take signal from an external microphone and convert it into electric pulses that go to the brain. But in order to work, they have to be placed right where the auditory nerve would typically connect the ear to the brainstem -- which can be difficult for surgeons to find, especially in patients who have completely lost or never had a substantive auditory nerve to begin with. So my job was to identify the nerve tract running from the auditory cortex down to the brainstem to guide the surgeons on where to place implants.


  • Developed artificial intelligence algorithms to predict the time, severity, location, and set of likely injuries of car accidents under the American College of Surgeons' Trauma Quality Improvement Program (via
    Sunnybrook Research InstituteSunnybrook Health Sciences Centre, Toronto, Canada
    ). This one was a really fun project that started as an idea at a hackathon, and evolved to me pitching the head of trauma, surgery, and emergency medicine at the hospital I worked at.

    The approach was part-mechanistic, part-probabilistic. The basic theory is that the baseline likelihood of a person getting distracted is ~consistent across the population, and therefore negligible for systems-level modelling purposes
    ¹it is, of course, relevant for individual-level models, which is why individual-level interventions like better driver training and vehicle improvements like collision avoidance systems help decrease accident frequency.
    . As I used to explain it, I could get distracted by a buzz on my phone while driving, but if I'm driving in the middle of an open, empty, straight road with no ice on a clear day, the likelihood I'm going to get into an accident is pretty low (not that you should do that!).

    This means the baseline likelihood of an accident is modulated by various environmental features -- the big ones being time (of day, month, and year), geography and hydrology, weather conditions, road and intersection type and quality, and traffic density. This allows you to determine, for each position on a map, the set of coefficients that make an accident at that location more likely. This is the deterministic part. You can then use historical data to model the thresholds of various combinations of features that result in relatively high probabilities of an accident occuring. The same thing can also be done for injuries: different combinations of features result in different types of accidents. Different types of accidents result in different types of injuries.

    Together, the result is a system that can monitor various environmental variables and determine where and when accidents are most likely to occur, and dispatch various interventions. My preferred intervention would have been rerouting traffic flow, but unfortunately the road network was not sophisticated enough to allow that to happen; most intersections, it turns out, run on local sensors and/or timers. So the main discussion ended up being around preparing EMS teams, since a colleague of mine's work identified that patient survival hinges on how quickly they can be brought to hospital.

    [insert story about the trauma surge we solved]


  • Did some private sector work in data science in healthtech and adtech. Wrote libraries to predict patient out-of-pocket cost in the US healthcare system, ETL code, anomaly detection algorithms for online ads, among other things.


  • Guided a biotech unicorn's expansion at a critical time in their growth trajectory.


  • Turned around a publicly-traded biotech company after a Phase III clinical trial failure.


  • Worked on a hardware company doing a mix of radar sensing and projection interfaces.


  • Currently developing a visual programming language for designing and modelling organizations.
Publications

Contact Me
I am mostly active on Twitter, where I keep my DMs open.
If you prefer to reach me by email, I can be reached at james [at] provisionalideas [dot] com.