TL;DR: Lets develop a safe-to-fail privacy-preserving distributed sensor network as a testbed for the future distributed platform.
Development of the revolutionary technologies currently envisioned is difficult, expensive and slow as the proven recipe of success of ‘fail early, fail often’ is beyond reach due to privacy and economic concerns. Many groups work in isolation to produce remarkable results, but they don’t find the desired adoption because they are unable to integrate with other projects. To have a path to success we need to lower the stakes, but cost of development needs to be offset with economic value. The immediate value proposed here is ecological. Birds are sensitive to environmental changes, so if a species’ calls are heard in an unfamiliar location this is a probable indication that an event with ecological impact has occurred. This could be an invasive species displacing another, or a chemical spill depleting food supplies, or a signal that spring will come early and now is the time to sow the fields.
To gather enough data for the results to be meaningful, some early adopters should be identified. Since there is popular interest in mitigating for example seagulls in residential zones the app may find plenty of adoption because of interest in mapping out targeted need for pest control. Amateur ornithologists could enjoy identifying which bird it is they are hearing and are likely to collect high quality data. Finally, by computing a trust token for the value of each contribution there may be enough interest among cryptocoin enthusiasts to download the app. With some economic incentive they are also likely to provide the adversarial examples needed for robustness, enabling the path towards the true goal of the project which is to build the foundation for the modern online platform.
Some of the technologies which could be integrated include differential privacy, distributed + online machine learning, misinformation resilience and multi-party computation, all within the context of smart contracts and bioinformatics.
This specific application, distributed bird call identification, is meant to reduce the complexity of integration as much of possible by making failure a viable option. The worst-case privacy failure that I am willing to accept is online advertisers finding cries of infants among the data, and being able to somehow target ads because of it. Speech needs to be filtered out at the edge, and security in depth is always the right choice.
I would like to hear your thoughts about how difficult or complex you think this integration could be. Perhaps you have experience with similar projects? Would you be interested in joining a group like a Discord or Slack server with the focus of developing integration to fruition? Are you perhaps already making progress on integration? Do you have a more promising test case than avian bio-sensors in mind? - I would like to hear about it!