Learning Library

← Back to Library

Trustworthy AI: From Dating to Hiring

Key Points

  • The speaker outlines four trust pillars for personal advice—unbiased recommendations, privacy of shared data, adaptability to evolving preferences, and transparent reasoning behind selections.
  • These same pillars define “trustworthy AI,” which is essential when businesses rely on AI advisors for critical decisions like hiring.
  • New regulations (e.g., New York’s upcoming law) will require firms to demonstrate their AI models’ trustworthiness after a year of production, prompting a need for formal governance and accountability.
  • Implementing trustworthy AI involves a structured process: defining business goals and target outcomes, assembling cross‑functional teams (HR, compliance, privacy), establishing governance structures, and rigorously developing and validating the model.

Full Transcript

# Trustworthy AI: From Dating to Hiring **Source:** [https://www.youtube.com/watch?v=V7kWAZ-dV0w](https://www.youtube.com/watch?v=V7kWAZ-dV0w) **Duration:** 00:05:28 ## Summary - The speaker outlines four trust pillars for personal advice—unbiased recommendations, privacy of shared data, adaptability to evolving preferences, and transparent reasoning behind selections. - These same pillars define “trustworthy AI,” which is essential when businesses rely on AI advisors for critical decisions like hiring. - New regulations (e.g., New York’s upcoming law) will require firms to demonstrate their AI models’ trustworthiness after a year of production, prompting a need for formal governance and accountability. - Implementing trustworthy AI involves a structured process: defining business goals and target outcomes, assembling cross‑functional teams (HR, compliance, privacy), establishing governance structures, and rigorously developing and validating the model. ## Sections - [00:00:00](https://www.youtube.com/watch?v=V7kWAZ-dV0w&t=0s) **Seeking Trustworthy Dating Advice** - The speaker outlines criteria—unbiased, transparent, private, adaptable, and explainable—to ensure any advisor, including AI, provides trustworthy guidance in finding a boyfriend. ## Full Transcript
0:01i'm looking for a boyfriend 0:03and 0:04i've been asking around i've been 0:05soliciting some advice but how do i make 0:07sure that that advice is trustworthy 0:09especially when it comes to something 0:11this monumental 0:12so i've been thinking and there are a 0:14few things that i would like out of this 0:16advice in order to trust any advisor in 0:19making this decision 0:20so first i want to make sure that 0:21they're unbiased right i want to make 0:24sure that they're open to everything 0:26i certainly am not ruling anything out i 0:28want this to be a fair process 0:30i also know that in order to get the 0:33optimized outcome i'm probably going to 0:34have to share a good amount so it's 0:36going to be a give and take a 0:38transparent process 0:40but if i do share that data i want to 0:42make sure that it's not being used for 0:44additional purposes that they're not 0:46telling other people that they shouldn't 0:48i want to make sure that my 0:50preferences are private 0:52next i want to make sure that they're 0:54robust so 0:56my dating history maybe not that 0:58indicative of my current preferences 1:01i want this advice and this advisor to 1:03be able to adapt with me to take in 1:05these new parameters uh as we learn and 1:09grow together 1:10the last piece says when we finally do 1:13get to a point where some lucky fellow 1:15is selected i want to understand what 1:17was their thought process behind it why 1:19do they think that we'll be a good fit 1:21these are the different pillars of trust 1:24that would be important for me for 1:26making this decision but they are also 1:29crucial to any business 1:32having any ai advisors help them make 1:35decisions and this is how we define 1:39trustworthy ai 1:40but now that we have these models 1:43or these different traits we don't stop 1:45there right we actually have to create 1:47the model we actually have to create the 1:48selection process 1:50so we're going to switch to something 1:52that's a little bit more contextualized 1:54in business terms and this is hiring 1:57practices right now so many fortune 5 2:00fortune 10 companies at this point do 2:02use hr 2:05ai models to help them with their hiring 2:07practice for selecting top talent but 2:10this has proved to be tricky to say at 2:12the least 2:13and there's legislation that's coming 2:16out at a federal level in new york it's 2:18rolling out of the first of the year 2:19that says that you have to prove that 2:21your model is 2:22trustworthy 2:24by january 1 and it has to have been in 2:27production for over a year and prove 2:29this how do we do that we start with 2:37the model proposal 2:40so at this point we need to make sure 2:42that we are selecting the right business 2:44outcomes we want to know who we're 2:46selecting for why we're selecting for 2:48what different data sources we also want 2:50to make sure that this is a team sport 2:52hr department's involved compliance 2:54maybe privacy if you're worried about 2:56that but this is the part where you set 2:58up the governance structure inside your 3:00model organization so that there is 3:02accountability throughout 3:04this is a crucial step 3:07the next is we want to make sure 3:15that we're doing 3:17the model development correctly so this 3:19is when we're pulling in our 3:21developers our engineers they're 3:23creating the model as we have defined it 3:25previously 3:27after that we're going to do a 3:29pre-implementation review 3:33and during this piece we're going to 3:35make sure that our model here doesn't 3:37include anything around bias right we 3:40want to make sure that we are selective 3:42of all groups getting back to this 3:44fairness piece we also want to make sure 3:45that it's a transparent process we're 3:47not sharing any data it's private and 3:50it's going to be adaptable and 3:51explainable so once we feel like we get 3:53this piece done 3:55that's when we're going to come in here 3:56and we're going to involve 3:58the model approval 4:02and the model approval piece is going to 4:04bring in an overall governing structure 4:06again so remember right we're setting it 4:08here but then we're actually having 4:09people come in we've created our model 4:12we've proven that you know it's got all 4:13of these different components of 4:15trustworthiness and then we're saying 4:17we're good to go we're going to put this 4:18in production 4:20typically we're seeing that 4:21organizations at this point are 4:22involving you know a board an ethics 4:25board this is something that's being 4:27outsourced into more and more divisions 4:29within your organization at higher 4:30levels because this is not a technical 4:32discussion this is the technical piece 4:34but in terms of implementing the model 4:36and creating trustworthiness that's a 4:38line of business opportunity here so 4:41once this gets the approval then we go 4:43and we put it into production but 4:46at certain intervals we're doing the 4:48compliance and validation testing so 4:50we're not done there at certain 4:52intervals that we've set here right 4:54we're going back and we're saying okay 4:56is this fair is this is there any bias 4:59has there been any drift is this still 5:02achieving what we need it to so in this 5:04way you can see that this is going to be 5:06an iterative process where we're going 5:08back and forth to ensure that your model 5:11will always be trustworthy 5:13thank you 5:15if you have any questions please leave 5:17them in the comments below also please 5:19remember to like this video and 5:20subscribe to our channel so we can 5:22continue to bring you content that 5:24matters thanks for watching