How to Accelerate Government AI Initiatives with Data Sharing

The U.S. Federal Government has come a long way since the Executive Order on Maintaining American Leadership in Artificial Intelligence (AI) was issued early last year. A recent study of 142 federal agencies (excluding military) found that 45% have experimented with some form of AI or machine learning (ML) tools, with the top use cases being regulatory research, analysis and monitoring, followed by enforcement.[i]

Federal AI Use Cases by Governance Task

• Regulatory research, analysis and monitoring: Collects or analyzes information that shapes agency policymaking (e.g. analysis/prediction of adverse drug events).
• Enforcement: Identifies or prioritizes targets of agency enforcement action (e.g. facial recognition systems for border protection).
• Public services and engagement: Provides services to the public or facilitates communication with the public for regulatory or other purposes (e.g. chatbots to answer questions).
• Internal management: Supports agency management of resources, including employee management, procurement and maintenance of technology systems (e.g. countering cyberattacks on agency systems).
• Adjudication: Supports formal or informal agency adjudication of benefits or rights (e.g. tools to support patent and trademark determinations).

However, the study also found that only 12% of the use cases could be considered technically sophisticated and AI usage is concentrated in a small number of agencies – about 7% of the agencies were responsible for 70% of all identified use cases.ⁱ

To effectively advance their AI initiatives, agencies need to collaborate with each other and non-federal entities to share data and algorithms that can feed robust AI models and drive innovation. That requires an interconnected neutral data marketplace where data providers and consumers can securely interact, transact and exchange data.

Diversity and trust are key to useful AI

One of the key challenges in developing and training AI models is that they need large volumes of data from diverse sources to be accurate. And, while many federal agencies have an abundance of data to work with, it may not be diverse enough to advance to the next level of sophistication in their AI algorithms. To get there, agencies need to enable data and algorithm sharing and exchange across a broad ecosystem of data providers and consumers, other agencies, and partners, clouds, applications, data scientists and users.

But challenges remain. Major cloud providers typically offer sophisticated algorithms trained on vast swaths of data, but they may not integrate well with AI models developed in-house or sourced from another partner. And government agencies may be reluctant to share data due to security concerns. A MeriTalk survey of 150 Federal, state and local and higher education stakeholders found that 51% identified security as one of the biggest challenges in expanding AI.[ii] Moreover, existing data marketplaces may be architected around limited or simplistic trust archetypes that do not provide the flexibility and security that agencies need to share different types of data sets with each other, their partners and public cloud providers.

Three steps for building a secure AI data marketplace

A secure, neutral data marketplace that meets the standards and requirements of its participants is crucial for government agencies to unlock the power of AI for delivering on their mission. It needs to support different data sharing models and trust archetypes that enable them to bring an algorithm to the data (distributed model), data to an algorithm (centralized model), or exchange both in a neutral exchange location (federated model) where neither party sees the other’s information, but both receive the same report. Key elements that form the basis for a secure AI data marketplace include a neutral location and governance/control framework for members to buy/sell/run their data and algorithms, as well as an interconnected analytics fabric that can support fast, low-latency movement of both compute to data and data to compute.

Three steps that government agencies can take to deploy or participate in a secure AI data marketplace like this include:

1. Interconnect agencies, partners and service providers: Establish a presence on a highly interconnected IT platform with neutral hubs in strategic locations.
2. Exchange data and algorithms, and train AI models: Share/use data and algorithms to achieve agency missions.
3. Join or establish a data marketplace: Establish a presence within a data marketplace that enables secure data and algorithm exchange.

By following these steps on Platform Equinix^®, government agencies can gain access to high-performance compute resources with high-speed, secure interconnection to multiple clouds, private data centers and data brokers. This provides benefits such as:

• Simplified and secure data/algorithm exchange on one trusted global platform interconnected physically and virtually across 56 metros on 5 continents.
• Support for multiple data sharing and trust archetypes (models) to accommodate different dataset requirements.
• Support for federated, geo-distributed and heterogeneous analytics to bring data and algorithms together.
• Full control of data through multi-zone security architecture, covering control, data and storage planes.
• Data traceability and integrity via blockchain-based lineage tracking.

To gain a deeper understanding of the data marketplace and how it can help accelerate AI-based innovation for your agency, schedule an interactive virtual Digital Edge Strategy Briefing.

[i] David Freeman Engstrom, Stanford, et. al., Government by Algorithm: AI Use by Federal Agencies, Feb 2020 as cited in Congress.gov, Senate Report 116-225 – AI in Government Act of 2019, 116th Congress (2019-2020), June 2020.

[ii] MeriTalk in collaboration with the United States Geospatial Intelligence Foundation (USGIF), Mapping AI to the GEOINT Workforce web page, report and press release.