IP pro­tect­ing AI/ML Vault

Back­ground

A study by Oxford Uni­ver­si­ty and Yale Uni­ver­si­ty indi­cates that AI will out­per­form humans in many ways and will auto­mate all human jobs in the next 120 years. By 2024, AI will be bet­ter than humans at trans­la­tion, will write best­selling books by 2049, and will per­form surg­eries by 2053. Machine learn­ing (ML), the pro­fi­cien­cy of a machine to mim­ic human abil­i­ty to accu­mu­late knowl­edge and use it to dri­ve insights, is gen­er­al­ly con­sid­ered the basis of AI.

Data is the dri­ving force for AI

Although AI might depend on its machine learn­ing abil­i­ties, we need to take a step back and real­ize ML doesn’t hap­pen in a vac­u­um. ML is dri­ven by big data, with­out which it can’t take place. Effec­tive­ly, there­fore, AI depends com­plete­ly on the amount of data we can cap­ture and the meth­ods we use to process and man­age it. For this rea­son, we need to pay more atten­tion to data cap­ture, trans­port, pro­cess­ing, and stor­age if we want to real­ize the promise of AI in the future.

Data Cap­ture is pivotal

Cap­tur­ing data is essen­tial, whether it’s for soft­ware-based AI appli­ca­tions, smart robots based on AI, or machine learn­ing. When AI prod­ucts were ini­tial­ly designed, devel­op­ers spent huge research and devel­op­ment resources col­lect­ing human behav­ioral data, both on the indus­try side and the con­sumer side.

The AI/ML Busi­ness Dilemma

AI/ML SaaS star­tups, com­pa­nies, and enter­pris­es share a com­mon under­stand­ing of the val­ue AI/ML can bring to auto­mate human and busi­ness-cen­tric process­es. Yet there is a strong mis­un­der­stand­ing of how AI/ML can be imple­ment­ed at the client with an align­ment of interest.

Con­sid­er, for exam­ple,  com­pa­ny A, a spe­cial­ist in deep learn­ing, has trained over years in a con­vo­lu­tion­al neur­al net­work for PDF doc­u­ment recog­ni­tion. In fact, com­pa­ny A is a world leader in this domain and can help com­pa­ny B to auto­mate finance and account­ing by scan­ning invoices.

While both com­pa­nies agree that A’s tech­nol­o­gy is ben­e­fi­cial for com­pa­ny B, they have con­flict­ing inter­ests regard­ing the deployment:

• Com­pa­ny A’s core intel­lec­tu­al prop­er­ty is the PDF net­work. Clear­ly, A is afraid of copy­cat­ting the mod­el as the core busi­ness val­ue. Infringe­ment is some­thing com­pa­ny A can detect, but it can hard­ly coun­ter­act giv­en the mere­ly infi­nite resources com­pa­ny B has.
• Com­pa­ny B’s inter­est lies in the pro­tec­tion of data. B is afraid that sen­si­tive data is leaked via pro­cess­ing invoic­es like cus­tomer names, address­es, and trans­ac­tion details which has severe con­se­quences for busi­ness and cus­tomer relations.

So far the par­ties had to agree to run the net­work in A’s or B’s envi­ron­ment. Nei­ther of the two choic­es meets the inter­ests of both par­ties, which com­pli­cates the busi­ness rela­tion­ship from the first day on.

Solu­tion: AI/ML Vault

enclaive’s con­fi­den­tial com­put­ing tech­nol­o­gy can help here. For the very first time, it is pos­si­ble to find an equi­lib­ri­um between AI/ML experts and data own­ers. What com­pa­nies A and B need is an AI/ML net­work vault.

How it works com­pa­ny A pro­vides com­pa­ny B with an encryp­tion of the network com­pa­ny A ini­tial­izes the AI/ML vault with the decryp­tion key com­pa­ny B queries the AI/ML vault with the encrypt­ed net­work and data of its choice

AI/ML Vault Benefits

Using enclaive’s con­fi­den­i­tal con­tain­er tech­nol­o­gy to real­ize the vault has the fol­low­ing benefits:

Rec­om­mend­ed Con­fi­den­tial Containers

Rust

Python

Node­JS

Java

Go

C++