Artificial intelligence has many uses at the Commodity Futures Trading Commission, from recognizing fraud patterns to interpreting unstructured data.
CFTC’s mission is to watch out for manipulation of financial markets, and supervised machine learning can help. Cases of spoofing, cornering the market or position limits, for example, are messy in real life and difficult to spot. Sometime they are combined. Therefore, CFTC needs to train machine learning to spot cases in unstructured data sets.
Tamara…
Artificial intelligence has many uses at the Commodity Futures Trading Commission, from recognizing fraud patterns to interpreting unstructured data.
CFTC’s mission is to watch out for manipulation of financial markets, and supervised machine learning can help. Cases of spoofing, cornering the market or position limits, for example, are messy in real life and difficult to spot. Sometime they are combined. Therefore, CFTC needs to train machine learning to spot cases in unstructured data sets.
Tamara Roust is CFTC’s first chief data officer and the director of its Division of Data. She said the agency uses natural language processing, or the ability for computers to process linguistics in the same way as humans, to ingest data from unstructured sources like PDFs and voice data, and to convert it to a structured set.
In its regulatory role, CFTC has another limitation: it cannot go back to someone it is subpoenaing and request data be turned over in a business reporting language such as XBRL or XML.
“Most of the time, they will give it to you however they feel like it,” Roust said during a webinar hosted by ATARC on Wednesday. “I’m a price taker in the economic context of I have to take the data in the way that it’s provided.”
She can push data into a repository, and have an analytics engine call the repository to extract the needed information.
“The difficulties there are you need the authority to operate, you have to have an agreement between you and the provider and every provider, that this is what you’re going to do. I’m going to connect my system that receives structured data to your system that provides structured data, and we need to agree as to what security levels we’re going to have,” she said.
Roust has experience implementing AI in the public and private sectors; she also worked for the state of Illinois from the fall of 2019 to the fall of 2020, and before that at various investment companies. Between those sectors, she encountered difficulties with contractual acquisition tools. She described using a contractual vehicle to bring on a consultant during the pandemic who could tell her what tools were available to set up contact tracing, as there was not an agile methodology in place. This allowed her office to change the culture and the way of doing business.
“So when I was at the state, I needed to get the [COVID-19] contact tracing system stood up, and I was dealing with all these legacy system issues, and I had to create the layers in order to get the contact tracing system stood up … all the context tracing staff have to be able to go into a web browser and enter the data,” she said. “And that means they’re pointing to a cloud instance. And that means it needs to connect to the public health information, which is in the 1970s-era database. And so you need a layer in between to talk to the two sides.”
AT CFTC, system integrators can perform that layering function. Roust said a layer is made first as a temporary measure long enough to remove the legacy system.
In her experience, implementing AI varies based on the size of the organization. Whether public or private, a larger corporation or agency that is more structured and bureaucratic will typically have a harder time implementing AI than a smaller one.
Per the 2020 executive order on promoting trustworthy AI in federal government, CFTC is required to use AI in a way that fosters public trust, maintains privacy and produces results that are explainable. The agency also has to optimize its AI purchases to both budget and procurement timelines, which can be long.
Thinking about the future, Roust said she would like to see more AI in government for process improvements. She likened it to a retail establishment that has not integrated its systems. Using AI in a master data management context, for example, to link rewards systems to ordering systems, would eliminate duplicate logins in most cases.
In the Illinois state government, personal citizen identifiers can carry data across agency stacks from public health, to taxes and unemployment offices, etc.
“When you had … a state citizen come to use any of these services you needed to have, like, one citizen ID so you knew that, ‘Oh, they’re receiving these benefits and their addresses changed.’ So you’re changing in all of the systems at once instead of telling the tax people your address change, telling the health and human services people your disability prescriptions need to be sent to a new address, telling the education people that your grant funds need to go to this address,” she explained. “That is how I would really love to see government get on board in terms of serving our taxpayers better.”
