Lisa Blunt Rochester: Lawmakers Should Look For Ways To Use AI To Detect Market Manipulation

5 months ago

During a Senate Banking Committee hearing on Wednesday, Sen. Lisa Blunt Rochester (D-DE) asked Vice President for AI Models at IBM Research Dr. David Cox about the usage of AI to detect market manipulation.

Transcript

00:00Thank you, Senator Cortez Masto.

00:03Senator Blunt-Rochester.

00:05Thank you, Mr. Chairman, and thank you also to our witnesses for attending today's hearing.

00:12Dr. Cox, I'd like to start with you.

00:16First, I would love to follow up afterwards on the conversation with Senator Reid regarding

00:24the elimination of jobs and any kind of research that you have or know of.

00:31You mentioned specifically lower-skilled workers becoming more productive, but not those jobs

00:37being eliminated.

00:38So as former Secretary of Labor of Delaware and focused on jobs, I'd really love to follow

00:43up with you later on that.

00:45But for my questioning today, AI, we know, can spot unusual trading faster and help catch

00:52problems sooner.

00:53This can make capital markets safer and more transparent, especially if the firms can prove

00:59that these problems really did happen with good records.

01:03For example, AI can look at suspicious trade patterns and alert regulators to investigate

01:09for potential market manipulation attempts.

01:12What concrete metrics should policymakers prioritize to validate that AI improves market oversight

01:20and stability?

01:23That's a great question.

01:25You know, this is going to have to...

01:27Evaluation, I think, in general, is one of the most interesting and difficult parts of

01:32AI.

01:33The top-level capabilities are very exciting, but the question of how do we measure things

01:37is really something that's arguably the most important part of all of this.

01:43And I think that's going to have to happen industry by industry, and it's going to have to happen

01:48with the partnership between public and private and different stakeholders to be able to define

01:53what ultimately, you know, how you're going to measure that progress.

01:57And in business, we do this all the time.

01:58We're measuring KPIs of our success.

02:01It's an art.

02:02And I think it's something that we all have to do together.

02:04So, you know, to dig a little deeper, like what records maybe should firms keep so that

02:11the regulators can replay what happened?

02:14And I'll open these two questions up for the panel.

02:17Thank you for the question, Senator.

02:23So there's two things I think about in terms of your question.

02:26One is we measure a lot of the outputs of these models.

02:30We should be focused also on the inputs.

02:33And so understanding the inputs into the models requires was there good documentation?

02:38Did you have a handle on the data that fed the model?

02:41Who trained the model?

02:43All of those things should be what we have available anytime we put a model in place

02:49that we think is going to drive productivity, efficiency, or effectiveness.

02:52Because right now we are all focused on the output and trying to measure the output.

02:57And as you said, it's a little bit of an art because it doesn't necessarily translate.

03:01A KPI doesn't necessarily translate into your balance sheet, your income statement very neatly today

03:06because people are still exploring, understanding it.

03:09So the other part of it is I think about surveillance and fraud.

03:14I think we would continue to use the KPIs that we look at today.

03:18So for example, for our fraud solution, we focus on operational efficiency and operational effectiveness.

03:24Operational efficiency is false positives.

03:26Is the model allowing us to reduce false positives?

03:30And if it does that, it allows the institution to redeploy that capital in a more effective way.

03:36On the other side of it, is it more operationally effective at identifying instances of fraud or market abuse?

03:45And we've just talked about this before.

03:47As good as we will become with our own AI modeling, and that will be the extra layer of defense.

03:53So AI will be a layer, a fourth layer of defense for us in that regard.

03:57But that's where that public-private partnership is so important because it comes down to the data.

04:01How much data do we have to measure what we're looking at?

04:05And do we have enough data to identify it across the entire financial system?

04:10You know, one of the concerns I have is for those who are smaller companies or first-time investors,

04:19you know, we know that this could maybe lower the cost and be an equalizer.

04:24It can tailor the needs to them.

04:26And we're talking about outcomes.

04:29We know we should measure this with outcomes and not promises.

04:34Mr. Cohen, which results should we track to show AI is truly widening access?

04:42And if you had to publish one metric each year, what would it be?

04:48So I know it's like we don't serve the retail community.

04:51We're not a consumer-facing company in that regard.

04:54But the metric we look at is very simple for us in terms of the market.

05:00It's the quality of the markets.

05:04And we look at the quality of markets in two ways.

05:06One is, are the companies, is the bid-ask spread for these companies what it should be?

05:12So is the, quote, quality for all of the listed companies on NASDAQ tight and available and liquid

05:19so investors can get the best experience, number one?

05:21Number two, we're also focused on the depth of that liquidity through the entire market.

05:27We do not want to just see the top 10 companies in the world be the most liquid and then everybody

05:32else really suffer from the lack of liquidity and the lack of market quality.

05:37So we are putting programs in place to ensure that we democratize market quality.

05:42That's really important to us across the entire spectrum of small, medium, and large companies

05:47so investors can benefit from investing in whatever company they want and not look at

05:53the ecosystem and understand, well, you know, the top side, if you will, the top side looks

05:58much more attractive than the last 100 companies on NASDAQ.

06:01That we think about a lot.

06:04I have questions for the record and I yield back.

Lisa Blunt Rochester: Lawmakers Should Look For Ways To Use AI To Detect Market Manipulation

Category

Transcript

Be the first to comment

Recommended