When we go online, we must constantly choose between privacy and accessing services. We’re at the critical junction where artificial superintelligence is being built and neurotech devices will soon have access to so much personal data. Now is the right time to build neurotech infrastructure to enable autonomy while maintaining privacy. This looks like software, research papers and patents, especially addressing present usability and efficiency obstacles. Our team has a unique skillset across brain-computer interfaces, applied math and design; The skillset, fast learning speed + effort and geographical location mean We Are Gonna Make It!
Scenario 1.
Our chronically online users are already generating massive amounts of personal data. In several years, companies will be able to train a model with your chat history to respond in your DMs automatically just like you. When brain-computer interfaces become accessible to the masses, it would generate even more data: Your attention and mood data from the morning meditation session can be fed into machine learning models. But to be concerned about privacy shuns you away from these services. Should we live off the grid with only our newspapers (and tea) instead? How can we make the most use of our data while maintaining privacy?
Scenario 2.
You are a neuroscientist attempting to study Alzheimer’s disease. You’re determined to recruit a large, quality sample of participants to make a key conclusion. You can even partner with other labs and get more participants. But! You need to comply with strict regulations like HIPAA so you give up on partnering and only work with your own 20 participants. In the already meagre set of data points, you need to anonymize participants, stripping away even more information. There are actually not enough data points to analyze to make key conclusions about Alzheimer’s, not to mention early diagnosis. You collapse on your chair—To collect quality research data while complying with privacy regulations is giving you a headache!
We claim that there’s a technology which can make use of our data while satisfying privacy. It’s been in the literature for many years yet not implemented, for reasons we’ll discuss later. This is homomorphic encryption.
Homomorphic encryption allows computations to be done on encrypted data without decryption. One of the early attempts to compute on encrypted data was known as Yao’s millionaire problem—How can two millionaires find out which one is richer, without knowing how much money they have?
Coming back to the earlier example. Now, in your morning meditation session, the brain-computer interface could gather data about your attention levels, store the encrypted version of this data, send them to the cloud to analyze them en-masse and return the decrypted results. You learn your focus has recently gotten better because of meditation. All the neural data has been utilized for insights without decrypting.
Likewise, the Alzheimer’s researcher could partner with a few other labs. Their colleague enters their data from their 20 participants as they collect, not worrying about safety because it has been encrypted. After data collection, researchers send data to the cloud for secure analyses. Within a flash, each researcher gets a copy of the results from a pool of 100 people. They realize an interesting pattern about dementia.
In these two scenarios, we could indeed analyze sensitive brain data to get produce meaningful results while maintaining privacy.
Hearing the good news, the Alzheimer’s researcher is now bright-eyed and bushy-tailed and they type in “homomorphic encryption” into Google. They discover several Github links in a corner of the internet. Setting up the library took the neuroscientist 5 hours. No nice pictures are guiding the neuroscientist on what to do, just digging through lots of code being scattered around, creating errors that cause even more headaches. Finally, they got it and began encrypting their Alzheimer’s study dataset on their laptop. It ran for half an hour.