The series, “Accelerating the Mission with AI and Security,” produced by Scoop News Group for FedScoop and underwritten by Google for Government, invited leaders to share where they hope to see the most significant return on investment for AI implementation in the coming year.

Artificial intelligence to meet core mission needs

Workforce augmentation was a highly discussed use case for AI implementation in the series.

FEMA’s Office of the Chief Financial Officer is one agency that has been strategically working on a generative AI tool to improve mission efficiency.

Christopher Kraft, Assistant Administrator, Financial Systems for FEMA’s OCFO shared that his office is developing a proprietary generative AI tool – owned and operated by FEMA and DHS – to generate draft responses to budget requests that his team can review for accuracy.

The Department of Labor CISO Paul Blahusch discussed how his agency is leaning into AI with a dedicated AI office inside the Office of the CIO to help develop and implement tools and techniques to streamline workflows, which can translate into cost avoidance and improved programs. He referred to three AI implementation areas his agency is focusing on, including cybersecurity, back-office support, and assisting constituents in accessing services more quickly.

For agencies like the U.S. Patent and Trademark Office, using AI as an augmented assistant has been developing even further over the past three years, according to CISO Jamie Holcombe, providing each examiner with an augmented intelligence system next to them.

“So, during its searches, it can bring up not just one thing but a myriad of things that pertain to the uniqueness of that patent application or trademark registration. So, you really have to think that the examiners don’t want one thing, they want a plethora of things to say, ‘yes,’ it is unique and novel, or ‘no, it’s not,’” Holcombe explains. “AI and generative AI has helped in that regard because each examiner has a customized version that just applies to them.”

Many leaders see generative AI as a way to improve standard workflow procedures. Department of Commerce CIO Andre Mendes, said that for tasks that are incredibly onerous, his department is looking at how AI can be used to break through some of the clutter.

“In HR processes, for example, position descriptions are not really that exciting, but at the end of the day, consume an enormous amount of people and time and resources, and where we can, I think, leverage AI to dramatically improve and optimize those environments,” he explained.

Improved security for federal data

Agencies like U.S. Citizen and Immigration Services (USCIS) are far along in their cloud migration strategies, which means that data security strategies must now shift to account for an explosion of digital resources.

“All the immigration data that has to be cataloged and identified and tagged is a monstrous task. And frankly, there is no easy button to push when you’re talking about the volume and scale of data that we have, and the amount of change that it goes through on even a daily basis,” shared USCIS CISO Shane Barney.

“We have, from a cybersecurity perspective, in my plans I am building, what we’re referring to as a security integration platform, which is an open source-based platform, and it has a whole AI/machine learning piece built into it based on open-source principles and practices, as well as some software platforms that will be integrated into the security program. And more on the threat hunting side of things where we’re looking for those abnormal changes in the environment that could indicate a breach.”

His agency leadership is waiting on further White House guidance on AI implementation but is working on foundational principles that can help the organization move forward with implementation plans quickly, referring to an open cybersecurity schema framework USCIS has been working on.

“I see it as the future. It’s the way we have to handle it; the future of cybersecurity is data,” said Barney.

This sentiment was echoed by other leaders who want to improve how they manage, store and analyze data to strengthen their agency’s security posture. Centers for Medicare and Medicaid Services (CMS) CISO Robert Wood said that his agency is building a security data lake to minimize data silos.

According to Wood, generative AI models could play a more significant role in empowering the government workforce to ask plain language questions to get actionable insights from data if properly structured and react more quickly to security threats and vulnerabilities.

Other participants who shared their insights in this series included:

• David Shive, CIO, U.S. General Services Administration
• Guy Cavallo, CIO, Office of Personnel Management
• Dr. Kelly Fletcher, CIO, Bureau of Information Resource Management, Department of State
• Josh Lehman, Director, Office of Business and Customer Assurance, U.S. Food and Drug Administration
• Russell Marsh, Director, Cybersecurity Operations, National Nuclear Security Administration

This video series was produced by Scoop News Group, for FedScoop, and sponsored in part by Google for Government.

The post Federal leaders on accelerating the mission with AI and security appeared first on FedScoop.

Jamie Holcombe, chief information officer in the patent office, said that when it comes to adoption of emerging technologies like AI, leadership within the federal government needs to change the culture and incentive structure of the workforce.

“Culture, culture, culture. I don’t care about the tech. We can solve it. We prove that you can. But it’s the culture, if they’re willing to receive [new tech]. Unless there’s a burning platform, a lot of people will say, ‘I’ll [have] somebody else do it, I’ll figure it out later,’” Holcombe said during the Google Public Sector Forum hosted by Scoop News Group in Washington, D.C.

“Especially with government bureaucracy, we have so many people that are just incentivized to sit there and punch their card, turn the paper or, you know, sign this and put it over here, I did my job,” he added. “We need to change and challenge the status quo, we need to get a sense of urgency and incentive for our government workers.”

The USPTO in 2021 sent its top engineers to Google to be certified in TensorFlow and develop neural network feedback loops for patent examiners to rate algorithms, as well as to apply machine learning and AI to patent classification, search and quality.

Holcombe was highly critical of the federal government’s approach to innovation and long-term modernization during the Google forum.

“Our budgeting process is stupid. Our procurement is stupid. Everything we do in the government is pretty stupid, when you compare it to the commercial world, right?” Holcombe said. “There’s so many lessons that no one is willing to take. Who in their right mind would run a commercial enterprise or operation using a budget that was conceived three years ago?”

He also highlighted key differences between how the federal government and the private sector operate when it comes to emerging technologies like AI.

“If we ran our government like we ran Silicon Valley, we’d be much more efficient,” Holcombe said. “Can you imagine the Silicon Valley guy saying, ‘Wait a second, I have to fill all my compliance things before I prove my product works in the marketplace? What are you freaking kidding me?’” 

Google Public Sector Managing Director Aaron Weis pushed back on Holcombe regarding government compliance.

“We do compliance for the government, so we actually have to get all our Google products accredited,” Weis said. “So we’re gonna fill out all the government’s compliance forms, but we might use AI to do it now.”

The post USPTO CIO says AI adoption is held back by government culture and bureaucracy appeared first on FedScoop.

USPTO’s Office of the Chief Information Officer seeks a red-team vendor that can simulate attacks on its networks “utilizing current threat actor methods and resources to evaluate mitigation effectiveness all the way up to Advanced Persistent Threat (APT), Nation State (NS), and Non-Governmental Organization (NGO) threat actors,” according to a request for information the agency issued this week.

“[T]he United States Patent & Trademark Office faces some of the most advanced and persistent threat actors in the world,” the RFI says. “Therefore, USPTO is seeking market research information about partners with the necessary capabilities, experience, people, technology, and drive to join our team as a partner in helping to defend against this ever-evolving challenge.”

The Patent Office plans to use what it calls the alternative competition method under its agency acquisition guidelines. As such, the agency is searching for large and small businesses that can meet its pen-testing needs, and if it deems there is an adequate market, it will create a pool of eligible vendors and invite them to bid for the contract.

Because of the sensitivity of the work, USPTO will limit competition of the contract to only domestic U.S. companies. “For security purposes, due to the sensitive nature of the materials, the RTPTS RFI materials shall be disseminated only to verified domestic United States of America contract entities (NO-FORN) only after execution of the attached Non-Disclosure Agreement (NDA) by responding contract entities,” the RFI says.

After companies attest that they are U.S.-based by Jan. 11, they will be sent a more thorough package of RFI materials through which they can detail their services and past performance.

This RFI comes as Patent Office CIO Jamie Holcombe is pursuing a sweeping move to a zero-trust security architecture. In November, Holcombe told FedScoop his office is considering the adoption of encryption-in-use technology to protect data as it builds out its zero-trust security architecture.

The post Patent Office seeks penetration testing services appeared first on FedScoop.

Traditional encryption protects data at rest or in transit but not when it’s in use by on-premise or cloud applications, and disk encryption solutions degrade performance and can lock users out.

Encryption-in-use secures only underlying sensitive data, regardless of location, and analyzes requests in real time to block suspicious ones. According to Holcombe, it could help USPTO protect sensitive claims information because the technology is less likely than traditional forms of encryption to degrade performance.

“I have an obligation to disseminate all public data as best I can, but the things that I need to keep secret are the claims that the patent applicants file with us,” Holcombe said. “And it’s only good from the first application date to 18 months later, then something has to happen to it.”

Until then claims are USPTO’s version of “top secret,” he added.

The companies developing encryption-in-use are mostly startups, but Holcombe isn’t interested in those adding it to USB devices. He wants the capability in the data center.

“That’s where your cloud storage companies come in because they’re buying that technology from these little guys, but I want to get it before it’s sold to them,” Holcombe said. “If it comes wrapped with [Amazon Web Services], that’s fine.”

USPTO operates on a three-year, procure-and-replace cycle and is working with different tech companies to satisfy all the pillars of the federal zero-trust strategy: users, apps, data, network and devices.

The agency is trying to mature multi-factor authentication to protect users and working with Venafi on a device management solution. USPTO has a partnership with Netskope for secure access service edge. 

“We’re looking to spread that because that’s just one solution of many for the [zero-trust architecture],” Holcombe said.

The post USPTO eyeing encryption-in-use technology to secure claims data appeared first on FedScoop.

Speaking with FedScoop, Jamie Holcombe said departments continue to struggle with writing IT contracts that balance the need to obtain new technology as quickly as possible while also avoiding vendor lock-in.

“[H]ow easy is it to bring in [new technology] from an emerging technology point of view? If it’s not done within 90 days, don’t talk about right? You can’t spend years, because by the time you’re putting something in, it’s been overlaid with something else,” Holcombe said.

He added: “[You’re wrong] if you’re not writing your contracts for errors and omissions. If you’re not writing your contracts to be changed in the future, you’re also wrong.”

Holcombe spoke with FedScoop after the agency earlier this month agreed to a secure access service edge (SASE) contract with technology vendor Netskope. The new contract was awarded using the NASA SEWP governmentwide acquisition contract and is the first such contract to be agreed upon by a federal agency.

Secure access service edge is a framework for network architecture that brings cloud-native security technologies together with wide area network capabilities to connect users and systems. The contract could be worth $4 million and could last up to 19 months.

Holcombe added that using alternative acquisition pathways such as governmentwide acquisition contract (GWAC) vehicles can often be the most effective option for bringing services to government in a way that can be adapted as mission demands change.

One of Holcombe’s first moves after his appointment as CIO in 2019 was to push the agency to use more GWACs for IT services acquisition vehicles, which can allow departments to obtain greater value for money by using the purchasing power of multiple agencies when negotiating with vendors.

The post USPTO CIO: Pace of tech change continues to pose challenges for agencies acquiring IT services appeared first on FedScoop.

USPTO created the machine-learning algorithms to increase the speed at which patents are examined by importing relevant prior art — all information on its claim of originality — into pending applications sent to art units, said Jamie Holcombe.

Filtering data into haystacks allowing patent examiners to more easily find what they’re looking for — the needle — is the new paradigm for search algorithms, Holcombe said.

“The ability to search, especially the big datasets, gets me so excited,” he added, during an ACT-IAC event Tuesday. “Because that means we can unleash that power to anybody who can get on a computer and access the net.”

Patent examiners previously had to scour three to four pages of single-spaces, text searches for relevant prior art assembled based on word relevance. Now examiners can search concepts like “chemical adhesion” and receive all the relevant prior art they need in one place.

Authorities to operate

One hurdle Holcombe faces as he attempts to test new innovations to modernize USPTO is that often they haven’t been approved by the Federal Risk and Authorization Management Program (FedRAMP).

Small businesses have a hard time ponying up the money needed for authorization, and the federal government is “overly oppressive” when it comes to compliance, Holcombe said.

If a new technology has potential, USPTO temporarily tests it in a sandbox to determine the minimum requirements needed to issue an authority to operate (ATO).

“We’re not exposing everything to it,” Holcombe said. “But I’m giving it enough time to grow to the point where it can comply with the minimum amount of bureaucracy that it has to comply with.”

The most important thing to Holcombe when issuing an ATO? That data at rest resides in the U.S.

Data in motion around the world can be encrypted, but the rest of the world is the “Wild West” when it comes to protecting stored data in accordance with other countries’ regimes and authorities, Holcombe said.

The post USPTO chief information officer most excited about new search algorithms appeared first on FedScoop.

USPTO sent its top engineers to Google on the East and West coasts to learn more about ML and TensorFlow application programming interfaces.

Now those engineers are using Python with TensorFlow to apply ML to patent classification, search and quality.

“We immersed them in the culture, and they got Googly,” Holcombe said during an ACT-IAC event Wednesday. “They got certified in TensorFlow, which is the open-source library for a lot of neural network feedback loops.”

USPTO has patent examiners use those feedback loops to rate how well ML algorithms are classifying patent applications to the art units and examiners that evaluate them, as well as search for algorithms in the system.

Despite having 250 years of historical data to train its algorithms with, USPTO relies primarily on daily feedback from examiners to ensure they’re working.

The agency is also hiring vendors to classify patent applications and comparing those classifications against its own algorithms. Having vendors and patent examiners working in tandem further refines the algorithms, Holcombe said.

“There always could be a black swan, and that’s what you’re trying to prevent in the curation of your data moving forward,” Holcombe said. “Black swans have to be cared for and handled and managed appropriately, or else it will break the system.”

USPTO remains in the early stages of ML use, in part because it’s still cleaning its data. While the agency uses robotic process automation (RPA) for clerical and administrative processes, it’s still getting used to the technology before applying it to patent and trademark workflows, Holcombe said.

Holcombe said he doesn’t believe AI yet exists, but there’s a spectrum of automation technologies with RPA at one end and more advanced neural networks at the other.

The post Machine learning speeding up patent classifications at USPTO appeared first on FedScoop.

Details haven’t been finalized, but Chief Information Officer Jamie Holcombe tasked human resources with overseeing what he calls an “organizational shift.”

Allowing people to serve on projects — called product teams at USPTO — of their choosing “seems to be almost impossible” in government, Holcombe said. He said it could lead to “heartache cases,” where no one picks a particular team, but the CIO has a plan for that, too.

“When we do have holes or missing folks, competencies and so forth, that’s where I’d like to contract with vendors to fill in those holes,” Holcombe said, during an ACT-IAC webcast Wednesday. “So it’s an amazing thing, the organizational change that will occur.”

Selecting their own projects allows employees to put their best foot forward while fostering greater accountability, he said.

USPTO’s new product catalogue consolidates the number of IT projects underway from more than 150 to just 30 across four product lines: patents, trademarks, enterprise, and infrastructure. Projects range from new software for internal use to products for patent and trademark applicants.

USPTO’s new product catalog. (USPTO)

Each project has a product owner and technical lead, and the agency’s current focus remains on ensuring product teams know their roles, Holcombe said.

“Once that occurs, then we might get some shifting based on wants and desires for individual product delivery,” he said.

Productivity at USPTO is up since the coronavirus pandemic hit, in part, because people aren’t taking vacations. The agency serendipitously moved to the cloud in January, ahead of mandatory telework, and “easily” scaled from about 8,000 teleworkers to 13,000 as a result, Holcombe said.

Teams are retraining machine-learning algorithms in a supervised neural network to more precisely classify patents, with similar technology for trademarks and fraud detection in the works.

New services include an image file wrapper archive — improving user access to high-value data assets, images and metadata — and Patents 4 Partnerships, a searchable repository of patents and applications available for licenses that USPTO stood up in 10 days during the pandemic.

“I think this helps the economy and it helps other entrepreneurs look at things that are a possibility for solving some of these problems that we have,” Holcombe said. “Also the robust IP marketplace can monetize your IP, and that’s what we’re trying to influence.”

The post Patent and Trademark CIO wants IT staffers to pick their product teams appeared first on FedScoop.

When Jamie Holcombe joined USPTO a year ago, one of the agency’s major applications still ran on an HP 9000, affectionately named “T-Rex” because the Unix-based machine was a dinosaur. Holcombe had T-Rex replaced as part of ongoing, multi-year efforts to stabilize and modernize critical IT systems and infrastructure supporting more than 8,000 patent and trademark examiners nationwide, around the clock.

Filings are at an all-time high — new patent applications reached 665,231 in fiscal 2019 and trademark applications increased 5.4%.

“Based upon our fees and based upon the velocity of the patents and trademarks that come through, our economy either flourishes or it doesn’t,” said Jamie Holcombe at an ACT-IAC event Thursday. “If we can get dependency — the time it takes to award a patent — down significantly, that frees up venture capital.” The office, unlike most federal agencies, is fully funded by the fees it charges.

USPTO has one of its 14 agile DevSecOps teams working on a replacement for a patent examiner search tool created in the ‘90s. Those teams are collocated and span the full development pipeline, with plans to increase their number to 30 or 40 in 2020.

As for the new search tool being developed, it leverages machine learning algorithms and will eventually make use of artificial intelligence.

“My opinion and my opinion only — I am not speaking for PTO — is there’s no such thing as artificial intelligence. There really isn’t,” Holcombe said. “I have not met Mr. Singularity, yet. Skynet doesn’t exist.”

For now, Accenture and Google have partnered with USPTO to identify search algorithms for “supervised learning” sessions with patent examiners. The algorithms learn what’s good and bad to improve searches, as well as how applications are classified for routing and evaluation — a “top priority,” Holcombe said.

Machine learning may also be able to improve trademark image searches and flag improper trademark activities and fraudulent imagery.

“I’m hoping that machine learning can help us find some of those data sources that we’re not using,” Holcombe said. “And I’m also hoping that we can simplify the ability to search by giving citizens search, so they can actually get in and figure out if the patent exists or not.”

Acting more like USPTO’s own customers

In the past, USPTO worked through about 200 to 300 systems across 150 to 200 projects at any given time. That number has been consolidated into 30 products across four business lines: patents, trademarks, enterprise and infrastructure.

The change comes in response to vendor complaints that USPTO did too many small task orders and instead needed more agile contracting focused at the portfolio, rather than project, level.

“We were stifling innovation at the agency for innovation,” said Kristin Fuller, chief of the Procurement Systems Division.

Functional testing is now included in development task orders, as are better metrics. And USPTO is moving to a new paradigm where — instead of every project having multiple orders for infrastructure, development, operations, cybersecurity and testing — there will be one contract per product addressing those various areas.

The timeline for implementing the product catalogue isn’t finalized and will coincide with how USPTO coordinates contracts, said Debbie Stephens, deputy CIO.

Ultimately, USPTO plans to remain in compliance with the Federal Acquisition Regulation but increase contracting speed, Holcombe said.

Even though contracts will be larger, USPTO will remain mindful of small businesses when making awards.

“We do have to be sure that we’re not a bull in a china shop, and I say that because I could have come in and wrecked everything, blew it up and started again — the Army way,” said Holcombe, a former Army officer. “Instead I’m trying to be more like the Air Force and actually have some thought about it before I blow it up.”

The post Faster patents require faster IT and contracting, and USPTO’s CIO has goals in place appeared first on FedScoop.