5 questions to help government agencies implement AI

Following in the footsteps of the private sector, the federal government is starting to embrace the transformative power of artificial intelligence (AI). As of December 2023, the Government Accountability Office (GAO) reported nearly 1,200 current and planned AI use cases across 20 federal agencies.

The Office of Management and Budget’s AI Intelligence Community of Practice has already attracted 12,000 members across more than 100 federal, state and local government agencies.

As AI takes root, IBM continues to build on its deep experience with the federal government by helping agencies understand how to implement this new technology. At the foundation of our AI efforts lies our partnership with Amazon Web Services (AWS), which enables us to deliver innovative AI solutions built on AWS technology stacks.

Together, IBM and AWS are helping federal agencies navigate the complexities of AI implementation and identify areas of opportunity—from proofs of concept (POCs) to fully operational systems—while helping to ensure compliance with stringent security and data governance requirements.

5 key questions can guide federal agencies as they implement AI:

Many government agencies often take a justifiably cautious approach to applying AI. Implementing new technologies is not without risk and AI is a powerful tool that requires guardrails.

A good place to start is with backend processes that focus on synthesizing information to accelerate decision-making processes. As a first step, we guide agencies to look for workflows that have the largest backlog in the organization. What presents the biggest roadblock to mission-critical work? That’s often a good place to start with AI.

From there, we identify the core business process and develop an AI solution that can help drastically streamline the workflow.

In the short period of time that government agencies have been exploring AI, distinct use cases have already emerged. If you’re an agency looking to streamline processes, here are 4 powerful AI use cases to consider:  

• Delivery of personalized content: Federal agencies are inundated with content that guides their mission delivery, such as regulations, policy guidelines, executive orders and other data sources. AI helps constituents and employees navigate content, answer queries, summarize complex texts and consolidate data from various sources. It rapidly accesses multiple databases to deliver personalized content tailored to user needs.
  
  
• Anomaly detection: Agencies are using AI to detect anomalous activities, including fraud, waste, abuse, improper payments and cybersecurity breaches. AI helps staff gather and review evidence faster, enabling quicker action.
  
  
• Natural language processing (NLP): Agencies process structured, semi-structured and unstructured data. AI’s NLP capabilities help organize, summarize and process this data from various sources, creating enormous efficiencies in activities such as benefits claim processing and enabling multilingual support for diverse constituents and communities.
  
  
• Preventive maintenance: AI empowers agencies to analyze supply chains for potential disruptions and failures by proactively monitoring maintenance needs in machinery, infrastructure and vehicle fleets.

Being open to pursuing POCs and experimenting with AI technology is a good start, but many agencies get stuck in this phase when their efforts stall. This can negatively impact staff who already see its potential in their day-to-day tasks.

IBM helps agencies move beyond pilot phases by developing comprehensive roadmaps. These roadmaps consider the necessary technology stack (including infrastructure and third-party software), cost estimates and guidance on acquisition for scaled deployment.

We work with agencies to determine the final operating model, including operational responsibilities, AI model support, ongoing monitoring and maintenance and strategies to address model drift.

For public agencies, security is paramount. Thankfully, AWS builds security guardrails into its AI products. For example, Bedrock, which is used to build generative artificial intelligence (gen AI) applications, never uses customer data to train its models.

Instead, individual agencies retain secured control of their unique data and models, using their own data to train their models. Our use of AWS resources enables us to minimize data security risks while enabling agencies to efficiently meet compliance needs.

IBM’s partnership with AWS uses the compliance and security benefits of AWS GovCloud, which is FedRAMP compliant and offers Joint Authorization Board Provisional Authority-to-Operate (JAB P-ATO) and multiple Agency Authorizations (A-ATO) for high-impact levels. We also support stricter data residency requirements within GovCloud.

Maintaining data quality is crucial for secure AI implementations. IBM® InfoSphere® Information Server for Data Quality uses NLP to help cleanse and classify data, especially within agency enterprise resource planning systems.

Agencies partnering with IBM gain access to 3 AWS gen AI technology stacks catering to varying maturity levels:

• Infrastructure for foundation model training and inference: AWS provides the infrastructure and tools to build, train and deploy large language models (LLMs) and other foundation models (FMs) efficiently and cost-effectively. This includes GPUs, AWS Trainium, AWS Inferentia and Amazon SageMaker.
  
  
• Tools to build and scale LLMs and other FMs: Amazon Bedrock simplifies building and scaling gen AI applications using LLMs and other FMs. IBM has also strategically positioned Granite models on Amazon Bedrock and Amazon SageMaker, broadening their accessibility and utility for enterprises.
  
  
• Applications that use LLMs and other FMs: AWS offers applications powered by LLMs and FMs, including Amazon Q for Business, Amazon Q for Developers, Amazon Q in QuickSight and Amazon Q in Connect.

IBM also offers the IBM watsonx™ AI and data platform, which uses AWS’s GP2 node for processing. Our watsonx platform can be provisioned as a fully managed solution in AWS and includes these components:

• IBM® watsonx.ai™: Our AI studio enables customers to train, validate, tune and deploy gen AI models.
  
  
• IBM® watsonx.data™: Our tool for scaling analytics and AI workloads for all data.
  
  
• IBM® watsonx.governance™: Our tool that helps ensure that AI workloads are responsible, clear and explainable. IBM watsonx.governance as a Service is available on AWS, and also integrates with Amazon SageMaker to reinforce AI governance, bias mitigation, and compliance.

One of the benefits of IBM’s close work with AWS is that watsonx enables agencies to combine data sources, including IBM® Db2®, AWS S3 and Aurora databases. Accessing data from various sources provides agencies faster insights without the added cost and complexity.

IBM Db2 is also offered as a managed service through AWS. With access to AWS’s Nvidia GPUs for accelerated computing instances, agencies can use Db2 to train and deploy their own LLMs.

Read IBM’s viewpoint: 5 key areas for governments to responsibly deploy generative AI - IBM Blog

IBM Consulting®, with a long history of serving government agencies, has achieved the AWS Generative AI Competency, further strengthening our federal expertise. Partnering with AWS, we’re developing innovative, AI-driven solutions to address common government challenges, helping agencies securely and efficiently fulfill their missions. These proactive solutions enable faster and better constituent service while future-proofing agencies.