Automated and algorithmic systems used to decide whether people get loans or jobs have attracted heightened scrutiny, compelling legislators to propose laws requiring audits of those systems. In New York City, there’s already a law requiring companies that provide automated employment decision tools to get their tech inspected, and they have just a few months left to comply before it goes into effect in January 2023.
Hired is one of those companies. Hired’s chief technology officer, Dave Walters, spent the early months of this year evaluating a growing pool of auditing services in the hopes of finding one that would not only help Hired check the AI fairness compliance box, but also offer deeply technical services for a long-term partnership. In the end, Hired chose London-based Holistic AI, which currently is evaluating the machine-learning models Hired uses for matching job seekers with roles that suit their skill sets.
Walters’ story, as told to Protocol, has been edited for clarity and brevity.
We cast a pretty wide net earlier this year, speaking to a number of different companies that were getting into the space. Of course, one of the things we found is generally everybody was getting into the space. This is a new demand, a new area that companies are spinning up to meet. Many of those companies [have come] from academia; they've been in that research area and in that mindset of doing this type of work, but just not necessarily monetizing in this way.
Because of how new it was, there was no single standard approach that every auditor was taking. That is the same with any new paradigm in any industry; there's going to be a certain amount of differential between approaches until some of those approaches get vetted out or rise to the top as the preferred approach.
We spoke with each of them. We started to narrow down that pool. Some of the key metrics we looked at:
- How new was the company: Did they just spin up this year? Or maybe they've been around for a year or two at least — anticipating some of the early adopters in the industry?
- What kind of funding did they have … We wanted to make sure that whoever we partnered with was going to have that level of stability.
- We looked at the auditing, not as a point-in-time audit, [where] we just check this box, go through an audit, then we move on and forget about it. But for us, this is a long-term initiative and partnership.
And as we narrowed down the field to the last couple of potential partners, we also then asked those candidates to provide us with sample reports from audits that they've done previously. We were able to see those public-facing, unredacted reports and see some of the more internal, [redacted] reports to have a better feel for what our end result would be.
There definitely was one that we saw that was not as comprehensive as we would have liked. We had some anticipation that might be the case based on some conversations, but then that helped solidify and crystallize that for us.
Seeing that report and understanding how that's going to translate into your business, and how that's going to benefit you, I think is critical.
Some we did not feel in their proposed approach were going to be digging as deep into the models. We felt very strongly that whoever we partnered with, it was going to be important that they have access to our models and access to our data without creating any security risks, or [personally identifiable information] risks. And so, they are able to fully evaluate the model and the results of the model. We worked with our internal IT group and security teams as well their team. I’ve leveraged some third-party tools to ensure that our working with them is done securely.
The easier path would have been to say, “You know what — we're going to avoid any kind of data and expect that you try to audit us based purely on the models themselves.” But we ultimately felt that was not going to give us the best result.
I would definitely suggest that companies start this process early — now. We’re already halfway through the year, so that is a little bit difficult for companies at this point that haven't started. But maybe for future years, because this will be an annual commitment for New York state to stay current, it’s going to take time to do that research on the possible partners to work with and to vet them out, and to kick off the actual audit and work through that audit.
🗣 How I Decided… 🗣
Wednesday, June 29
Wednesday, July 6
Wednesday, July 13
Wednesday, July 20
Wednesday, July 27
Wednesday, Aug. 3
Wednesday, Aug. 10
Wednesday, Aug. 17
Wednesday, Aug. 24
Wednesday, Aug. 31
Wednesday, Sept. 7
Wednesday, Sept. 14
Wednesday, Sept. 21
Wednesday, Sept. 28
Wednesday, Oct. 12
Wednesday, Oct. 19
Wednesday, Oct. 26
Wednesday, Nov. 2