Governments today face a growing issue: Artificial intelligence can help realize massive social, economic, and security achievements, but for them to proceed without AI-specific procurement processes is to fall increasingly behind. Governments are constrained by archaic regulations and practices that hinder rapid acquisition of cutting-edge technologies.

In my recent testimony before the U. S. Senate, representing CrowdAI—a small business I led that was recently acquired by Saab— I called attention to these complexities while advocating for the federal government to modernize its acquisition rules and processes, lest they be left behind by both our Allies and adversaries.

While it won’t be easy, change is critical. Here are five ways for government to build effective roadmaps for successful AI procurement:

1. Contextualize data

Despite what some businesses tout, commercial AI does not have plug-and-play solutions for governmental projects. In reality, there is only nascent constructs -- building blocks that are adaptable (and potentially incredibly effective) but their efficacy hinges on precise data for model conditioning. Without the correct data, AI models cannot produce the desired outcome.

While governments have, and should, protect their data, sources, and methods, it is impossible to build effective AI technologies without access.

To make AI and automation achievable, governments must begin curating specialized, high-quality datasets that are not only compliant with security and privacy regulations, but conform to common standards for machine learning. This task is complex, especially considering international data-sharing agreements and the complications that arise from global partnerships. However, ignoring this step will lead to inefficiencies that will frustrate adoption of AI-powered solutions.

2. Perpetually improve

The contracting models of the past are fundamentally ill-equipped to cope with the dynamic nature of AI technology. Just like your smartphone needs regular updates to improve features, fix bugs, and prevent vulnerabilities, AI models are similarly iterative technologies that rely on dynamic, ongoing adjustments.

For instance, our collaborative projects with the California Air National Guard to map wildfires in real-time revealed the necessity for continuous model adaptation as fires moved from forested to urban areas.

While we know what needs to happen for long-term success, most government contracts fail to include ongoing updates and maintenance required for these projects.

Governmental contracts must go beyond making a simple, one-time purchase and entrench the iterative nature of AI into their frameworks. A contract must be a reflection of the technology it is purchasing: adaptive, agile, and perpetually improving.

3. Rigorously evaluate

The democratization of AI through open-source models presents a dual-edged sword. While it expands access, it also complicates vetting, particularly for government agencies that may not have the in-house technical acumen to validate a model’s suitability for intricate missions.

Compounded with contracts that fail to facilitate iterative development, the result is significant risk exposure to governments from unqualified contractors and failed projects at taxpayer’s expense.

To mitigate this risk, governments need to invest in building or acquiring evaluation expertise. Procurement protocols should encompass not just initial benchmarks, but periodic reviews that scrutinize alignment with evolving mission objectives, data privacy and security norms, and social-ethical constraints.

4. Track and prioritize

Smaller businesses, with their nimble structures and innovative dynamism, often find themselves stymied when scaling their solutions within the confines of governmental procurement. While small businesses may be agile, they can also wither under lengthy government procurement timelines without support.

Contractual milestones within research and development contracts for transition to operations would help innovators scale up and provide lasting value to government organizations. Some programs like the Small Business Innovation Research (SBIR) have initiated steps in this direction. And SBIR’s sole-source award policy enables small businesses to compete with large prime contractors. However, it is not enough.

SBIR faces growing challenges. Over the last few years, we’ve observed the quantity of SBIR awardees balloon, the size of awards shrink, and the amount of transitioned projects slip.

To support SBIR program success and the businesses that pursue these awards, governmental offices should track and prioritize the quantity of projects transitioned to programs of record, not simply measure the quantity contracts or funds awarded. What’s more, departments and agency that participate in SBIR must update transition mechanisms, the contracts and funding, that will account for a capability needing significant ongoing improvement such as AI.

5. Support international frameworks

Over the past 75 years or so, we have come simply to accept that policy lags technology. The pace of change is too fast, we say, for regulations to be contemplated, let alone promulgated. However, the gap has become too wide and too consequential to ignore. The evolving tapestry of international entanglements, underscored by technological dependencies and data alliances, has rendered traditional procurement frameworks inadequate to the point of being dangerous.

To move forward, we need more than a national AI strategy. We also need to support international collaborative frameworks that consider the multi-faceted impact of AI, from cooperative security to ethics, if we want to create and sustain successful frameworks for advanced technologies.

Conclusion

Governmental agencies need to drastically evolve their procurement methods to meet the unique challenges posed by AI.

For governments to transition to the effective procurement and use of AI technologies, which carry the potential for extraordinary benefits, we must look at the bigger picture. Yes, we need contract adaptations. But governments also need a more profound systemic overhaul that aligns technological advancements with strategic foresight, legal rigor, and ethical considerations.

By doing so, we can transition from the present state of disjointed engagements to a future where AI serves as a robust, reliable partner in achieving the collective aims of society.

Devaki Raj was the CEO and Co-Founder of CrowdAI. On September 7, 2023, Devaki joined Saab, Inc.’s newly established strategy office as the Chief Digital and AI Officer. On September 14, Devaki testified before the U.S. Senate Homeland Security and Governmental Affairs Committee on AI acquisition and procurement.

Share:
In Other News
Load More