Why and how model IP and data protections are critical criteria in evaluating success in AI investment opportunities.
By Alon Kaufman, Ph.D., CEO & Co-founder Duality Technologies
A new gold rush has been ignited, this time with AI. The investment opportunities in AI are faster, bigger, and more lucrative than almost any historical example. But, where should investors dig? What should they be digging for? How can they ensure sustainable outcomes? How do AI and data privacy regulations play into investment decisions?
All AI opportunities today rely on a continuous interplay between the refinement of data and the development of novel AI techniques. Building and consuming AI models of any type requires data for training, tuning, and customization, hence high-quality data is the key for successful AI. Teams building AI applications based on novel AI models must work with third parties (data owners/custodians) to acquire enough high-quality data to build market-worthy products and services. That’s where collaboration in AI becomes critical.
However, the collaboration between such data custodians and AI application builders is becoming increasingly challenging due to the sensitive nature of data and models. This is because both data and models are often considered intellectual property or confidential information. As a result, both parties are hesitant or blocked from sharing assets with the other, creating a major obstacle to growth and success. In the gold rush analogy, one can describe the huge investments in AI foundational models as the infrastructure for new ways to discover and mine gold. And yet, the actual value (the gold) is in the outcomes, insights, and results of using these innovative capabilities on high-quality data.
The Secret is to Unlock Secure Collaborative AI
Currently, it’s major cloud service providers (and few others) who can afford to train foundational models (E.g., LLMs). While investments in such CSPs are public, massive, blue ocean opportunities are found in the services, the applications, and the life-changing symbiosis between AI and humans. Investors are pouring billions of dollars across thousands of new companies, expecting value creation in the hundreds of billions of dollars.
The gold in the AI market is composed of three essential elements:
- Fundamental AI infrastructure
- Scarce and precious proprietary data held by organizations
- AI-driven companies that leverage the interplay between AI infrastructure and the data custodians, doing so in a secure and regulated way
The third element is the connective tissue that protects, expands, and accelerates the critical data flows necessary for innovation and growth. As restated by a recent CNBC article, “One key priority area, based on the latest earnings calls, is AI models-as-a-service, or large AI models that clients can use and customize according to their needs.” Model customization requires collaboration with client data, and is neither practical nor scalable without solutions to ensure model IP protection and data security.
When everyone is invoking “AI,” the true differentiation lies in an organization's ability to harness its proprietary and sensitive third-party data effectively. Whether it be for bespoke AI models or refining customer-centric services, the key to unlocking the full potential of AI requires addressing all three essential elements as described above. This means having an enterprise-ready solution wherein both the model IP and the data used for training are protected in use. This is what we call “Secure Collaborative AI.” Since the infrastructure and data custodians already exist, the largest opportunity lies in enabling these AI-driven solutions.
The Dilemma of Data Privacy in AI Collaboration
As organizations embark on collaborative AI initiatives, they are confronted with a fundamental dilemma: deriving joint value from data collaboration inherently presents challenges in safeguarding privacy and confidentiality, as it requires trust and willingness among all parties. This challenge underscores the need for innovative approaches that enable seamless, secure, and confidential data collaboration. Therefore, both the major opportunities and risks in AI rely upon efficiently addressing security, privacy, and confidentiality concerns. Fortunately, privacy enhancing technologies (PETs) can be operationalized into solutions to overcome such challenges. Advanced encryption methods and techniques such as homomorphic encryption, federated learning, and confidential computing enable collaboration without compromising data confidentiality, data privacy, and model protection.
The Competitive Advantage of Data & Model Protections by Design
Those who remove the above-mentioned friction points will grow faster. Investors are becoming increasingly aware of the importance of protecting data privacy and confidentiality to both reduce risk and boost growth, prioritizing investments in those that demonstrate such commitments.
AI Investment criteria to consider:
- Efficiency in Data Acquisition & Use: AI vendors must efficiently locate and facilitate secure access to data with the governance controls and reporting to prove it.
- Model IP Protection: Vendors cannot risk IP leakage. This is especially important when going to market with models that customers want to customize with proprietary and sensitive data that can’t be sent to the vendor.
- Enhanced Transparency & AI Governance: How is data collected and protected? Used? By whom? Why? Answers to such governance questions must be easy, not manually brute-forced.
- Rigorous Security Protocols: Robust security measures will become the industry standard, with multi-factor authentication, encryption, and regular security audits becoming commonplace, emphasizing protections for data-in-use and the integrity of models.
- Adhere to the emerging AI regulations: A coherent strategy to future-proof compliance requires collaborative efforts across engineering, security, IT, privacy, and business teams as the speed of technologies and regulations continues to accelerate as do the demands of the market.
To be clear, these safety mechanisms are not the goal but the means to unprecedented growth in the AI market and serve as investment decision-making criteria.
The Road Ahead: Navigating the Intersection of Innovation and Security
As we stand at the precipice of a new era defined by collaborative AI, the road ahead is fraught with both promise and peril. The era of advanced models and AI heralds unprecedented opportunities for innovation and collaboration. However, amidst the allure of collective intelligence, organizations must remain steadfast in their commitment to data privacy and security. By embracing a paradigm of Secure Collaborative AI, the AI economy will bloom as leaders unlock the full potential of their proprietary data while simultaneously safeguarding it from prying eyes.
About the Author
Alon Kaufman, Ph.D., CEO & Co-founder Duality Technologies
Dr. Kaufman has over 20 years of experience in the hi-tech arena, commercializing data-science technologies and leading industrial research and corporate innovation teams. Before founding Duality, he served as RSA’s Global Director of Data Science & Innovation.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.