The rise of generative AI has been one of the most revolutionary technological shifts of the 21st century, and with the US being home to the most prominent AI companies, it is unsurprising that the American government has committed to this project more than most.
We sat down with Richard Patterson, Chief Data and AI Officer for the Bureau of the Comptroller and Global Financial Services (CGFS) at the U.S. Department of State, who reflected on the nation’s use of the technology.
For Patterson, there is much that the international community could learn from the US’ endeavours - from its successes, but also, he stresses, its mistakes.
___
There are a multitude of ways in which generative AI is being used within the US civil service. One of the most visible examples of this is the Department of State’s internal chatbot, StateChat. This is used for tasks ranging from completing performance reviews, to understanding the specifics of how new policies will affect the department’s operations, to identifying where to update rules and guidance to resolve contradictions between different pieces of legislation.
Similarly, the Department recently deployed StateInsight, a more powerful tool which brings together both structured and unstructured financial and performance data, to answer questions such as how policies will impact the funding and execution of certain programmes, as well as to better evaluate the efficiency of suppliers. StateInsight was built by Palantir.
The State Department is also experimenting in agentic AI with respect to invoice processing.
With the work of CGFS - which sits at the heart of the State Department’s financial operations - often being highly structured and policy-based, this creates favourable conditions for agentic AI to deliver strong returns on investment, Patterson states.
He say that autonomous invoice processing in particular could alleviate significant backlogs.
“Every summer, there is a group of foreign service folks who move from one post to another, about a third. So every three years, the entire foreign service moves,” he said.
“And so we get backlogged starting in July when those expense reports start coming in. You've got families and they have cars, they're shipping furniture, they're shipping clothes, they're staying in a hotel for a week, they're renting a car while their car is shipped. All these different things.
“Having an application like this makes it much more efficient and effective for [staff] to do their job.”
Yet the deployment of AI brings its fair share of challenges for those seeking to implement it. Concerns among staff about the technology’s reliability, potential redundancies, and costs also result in a reluctance to switch off legacy technologies, limiting the savings that the implementation of generative AI could produce.
While Patterson advocates for the dependability and efficiency of the AI tools, he also warns against the trend within organisations to adopt a shallow view of efficiency as an end in itself.
“This technology is supposed to help everyone, not just organisations’ bottom line,” he said.
He argues that care for the well-being of employees must be at the core of the implementation of any technology, urging leaders to consider how to restructure work to take into consideration people’s quality of life.
“Instead of letting people go, can we cut hours? Can people work less time?”, he asked.
“I think that mindset starts now because it's not going to start later.”
For Patterson, the importance of protecting dignity and quality of life is one of the most important lessons America can learn from certain other countries.
“The US has been over their skis [getting ahead of themselves] in technology”, he said.
“There isn't enough policy to support appropriate use of AI, and instead of our congressional institutions leading in creating appropriate policy, what we end up with is AI being misused, and then it has to go through the courts, and then the courts have to set precedents.”
At the same time, he believes the US approach offers an important positive lesson through its focus on “AI enablement”. This is the process of reworking systems around AI, rather than adding AI to the loop in a way that may not be effective or appropriate.
Many organisations, he argues, are adopting AI reactively rather than strategically.
“It’s like these cartoons that I see all the time”, he said, “where an executive says, ‘we need AI.’ ‘When do we need it?’ ‘We need it now.’ ‘Why?’ ‘Because we need it.’”
That mentality, he warns, risks creating a cycle of overspending without meaningful results.
“There's a real concern that we're overspending on AI without getting the benefits. And the reason is because, yes, we are overspending, and we're under-planning.”
His conclusion is simple. “We need to spend less and plan more.”