U.S. Government Procurement of AI Could Benefit from Unified Contractual Terms, Suggests OMB
federal agencies in the USA are bumping up their AI investments, forking out close to $3.3 billion on AI contracts with private sectors in 2022, a staggering 2.5 times boost from 2017. The catch? The traditional procurement process for IT systems is a slow, bureaucratic mess, confined to established vendors. With the Biden administration eager to promote the ethical use of AI within the federal government, streamlining this messy contracting process is essential, especially for startups unfamiliar with the red tape of government procurement. The Office of Management and Budget (OMB) should kickstart this process by developing voluntary standard terms for AI contracts, enhancing efficiency, and widening the pool of vendors, granting agencies access to cutting-edge systems.
First off, standardizing contract terms would simplify agreement reaching. Federal agencies struggle to clinch AI system contracts with vendors since key terms in licensing agreements aren't universally defined, a reality underscored in the 2022 Global Partnership on AI report. The report champions clearer contracting by establishing standard definitions for terms like input data, processed data, untrained models, and trained models.
Secondly, a standardized set of procurement clauses concerning crucial aspects such as data quality, testing, and documentation would aid federal agencies in creating contracts that harmonize with federal requirements more swiftly. For instance, federal agencies must document regular impact assessments over an AI system's lifetime, test AI systems in specific contexts, and have an independent oversight body scrutinize their performance, among other necessities. The OMB should create standard clauses that align with U.S. requirements, mimicking the European Commission's actions for the AI Act's requirements.
Thirdly, standard contract terms would neutralize the possession of European clauses. The EU's clauses prioritize European interests, with one clause insisting that datasets employed in developing AI systems are relevant, representative, error-free, and complete, a requirement under the EU AI Act. However, the United States does not impose such restrictions, and in many instances, providing error-free or complete data might not be feasible, practical, or necessary. Crafting model AI contracts for federal agencies would bypass accidentally adopting EU regulations.
Fourth, developing standardized contract conditions would make the process more approachable for vendors, particularly small-to-medium-sized businesses. Most federal AI service contracts typically go to Eastern Coast firms, proximity to federal agencies notwithstanding. By creating a broader pool of vendors, federal agencies will gain access to top-tier systems, including those that mitigate risks to privacy, civil rights, and civil liberties.
Developing effective standard terms necessitates a collaborative, multi-stakeholder process to ensure clauses are practical, reasonable, and technically feasible for all parties. The European Commission embraces external specialists for peer-reviewing draft clauses through public workshops and feedback requests; the OMB should mirror this approach for any standard clauses it helps create for the U.S. federal government.
Let's not forget, there'll be a multitude of AI use cases and arrangements that various federal agencies will engage vendors for. The aim should not be to create one-size-fits-all contract language but rather a menu of provisions that federal agencies can utilize.
Image credit: White House
- The lack of universally defined key terms in licensing agreements is a hindrance for federal agencies in securing AI system contracts, as highlighted in the 2022 Global Partnership on AI report.
- To streamline the contracting process, the OMB should establish standard definitions for terms like input data, processed data, untrained models, and trained models, similar to the European Commission's approach for the AI Act's requirements.
- Creating a broader pool of vendors through standard contract terms would allow federal agencies to access top-tier systems, inclusive of those that prioritize privacy, civil rights, and civil liberties.
