The Prompt, Ep 3 — Data Privacy in AI

[Music] foreign [Music] welcome back to the prompt on today's episode we will be discussing something totally not controversial and that doesn't require a lot of conversation which is security and I was kidding about everything I just said um super chill super chill topic I think it is marketers I don't know about all of you but I get spooked when I hear data privacy when I hear gdpr when I hear things like how are we dealing with security when when reaching for these new tools and so Alton I was stoked to see that you published um a blog post with our head of security John um and I love a tldr on how your conversation went with him um and what you can expect from this episode is hearing the three of us sort of just talk through that and what marketers should be looking out for and how we should be speaking internally about security as we sort of start championing these AI tools internally yeah totally it's a conversation I was excited to have because I have some like heavy opinions on data security overall but I also don't know anything about data encryption and the technical nuts and bolts of how data protections work and stuff so being able to talk to him shed a lot of light on some things that users and businesses can think about when they're using these AI tools and I don't know just gave me some food for thought on data privacy overall so some of the the biggest takeaways were him offering some insights on from the from the business side people being more aware of the specific tools that they're using and whether or not they can get into an agreement with these providers on if the providers are going to use their data use the use user data in any way so are the inputs that I type into this particular AI tool going to be used to retrain the tool are they going to be kept on a server somewhere in use for another reason that I may not know about or is there an agreement in place that says it's not going to be used at all it'll be held on our server for 30 days for instance and then deleted or it's not going to be used in any capacity it just disappears The instant that you type something into it so yeah that was a recommendation for for companies on the user side he recommended people not entered sensitive personal identification information like email addresses phone numbers you know home addresses things like that and a lot of common sense things when you when you kind of think about it to a certain degree um but yeah in in light of those two pretty pretty major insights I wanted to ask both of you all as as marketers and um people who build campaigns for things and build a lot of content that like sometimes might have some sensitive um proprietary information in it so let's say hypothetically you don't work at an AI company um but you you have a major product launch coming up you want to use AI to help get some content out the door a little bit faster a little bit easier just streamline things a little bit more but you're you're interested in protecting that proprietary data what are some of the steps that you would take to make sure that that data is protected do you do some shopping around to see which tool has the best agreements um do you adjust the data that you put in and be kind of careful about talking about these things versus these things what what are your your strategies in there oh man my toxic trait's gonna come out immediately where I'm like shouldn't my organization have done that for me like I'm a marketer I'm not a security you know what I mean procurement person like I feel like I would love to go into that being like hey I want an AI tool for my org here are three that I think are great for businesses okay security procurement people like pick the best one for me and let me get it implemented um I think you make a great point on like AI has been winning right now in the media in our hearts in our laptops like everywhere everyone's using AI because of the productivity gains it gives us immediately and so if I now have to stop and wait for that and and have to start like thinking about what can I put in what shouldn't I put in like I don't know it kind of bites into those productivity gains for me so that's what's coming to mind immediately yeah yeah for sure I'm definitely with you on like please someone higher than me help me figure this out and not only like which tool to use but also like what can and can't I say like what is considered proprietary um company information that I shouldn't use in that tool I think that would be my biggest question as like a Content person like what am I allowed to write about what am I allowed to tell my agencies and Freelancers to write about um I think the other big thing is like this is this to me is like the next thing that companies are gonna have to figure out because right now everybody's so interested in Ai and they're like everybody's so curious about it so I feel like team like the content team is using Jasper the sales team might be using chat gbt another team might be using something else your Freelancers are using whatever free tool they can get their hands on and so I think also it would be important for me to a understand what tool we're using and then figure out what's the right way to like thread that through the entire company and communicate that to any outside you know agencies Consultants Freelancers that we're using so that our data does stay safe because it's not just you it's across your whole company it's the people that you're you know the outside resources that you're using um so yeah I think I would just I that would be my biggest thing is like what tool is safe what can and can't I say to it and then also like how do I make sure that all of our resources are using the same thing so that the data is actually safe because yeah I think we're not only are we in a situation now still where uh people are just kind of using whatever and not telling their company but even you have Freelancers and agencies who are like still not comfortable admitting that they even use these tools which is very problematic when you're trying to preserve your data um so yeah I think just that's like the biggest thing that would be on my mind if I'm trying to implement this as at not an AI company totally I think those are all like really great insights and it reminds me of um the story I was just reading about Walmart in February Walmart issued a memo to its employees saying be careful what you put into chat GPT specifically and then shortly after that they release a set of guidelines around like okay if you want to use this tool this is how you're supposed to use it this is what you like should and should not be putting into it um things like that and I think those guidelines are super important um because I mean we have guidelines like that when it comes to watching out for phishing scams and just general protections across the board for like just the data and things like that on on our computers and in our emails like why it would make perfect sense to have guidelines like that at companies on how to use AI on what tools would make the most sense for like this preferred our particular business based on the data protection services that they may or may not use and and stuff like that so yeah it's it's about I think trusting that your organization is building the guidelines necessary to keep the data safe to make sure that it's being that the AI tools are being used in in the right way where you're not going to run into any issues down the line of like data leaking out or you know things not being used in the way that they should be yeah and I think that's something that kind of like keeps me up at night as a Content marketer is I think for those leaders to create guidelines and teams to create guidelines it's kind of on companies like us to First educate them on how to do that what do those guidelines need to incorporate like what does that even need to look like um so yeah I think that's something that I also think about a lot is like oh it's actually on us to like get this ball rolling a little bit and help these people out um I don't know just that's concierge on my mind these days yeah I feel like add that to the list of things that keeps me up at night um but I have a question for the two of you and it's sort of based on a real world example that I'll keep Anonymous but say you're working with a Content agency they write your SEO stuff they do some of the editorial work for you and you're about to renew your contract with them um and this is like current day and age right like while like the AI discourse is happening and they don't mention anything to you about AI like they haven't discussed it during any of their pitch meetings any of their original meetings they also haven't said they don't use it they haven't said they do use it it's just never come up and the renewal is on your desk are you signing are you saying wait we haven't had a conversation about AI or are you saying hey I feel the type of way that you haven't even brought up AI yet like what's your take on it yeah that's an interesting question I think for me I'm definitely bringing it up because like there's no way you can move forward at this point I think without at least asking questions like how are y'all using it for your other clients uh what tools are you using what are the pros and cons that you're seeing I think I would probably initiate that conversation if they did not um but I definitely would a little bit you're right I would a little bit be like why aren't we talking about this like it's such a huge thing and I know that you're using it or at least thinking about it um so yeah it would be it would be like uh kind of weird if they didn't bring it up first but I would absolutely bring it up there's no way I could I could move forward without it I think yeah same I would 100 have to ask like even if the renewal is right there on the table I mean I'm making a phone call and trying to get some some insight as best as I can because it's it is um it's something that for the most part anyone in content right now I would imagine is like at least aware of aware of its impact um aware of his capabilities and things like that so yeah I would love to get the perspective on their their views toward it how they use it um and even offer my guidelines on like this is how I would like to see it used this is how I would not like to see it use in the content that you give me um because I also feel like as content professionals it it might be a little bit easier for us even to say like to or to see okay AI there's a decent chance AI wrote this because it's kind of like surface level or like not as deep a nuanced as I would expect it to be um so it would be like imperative to a certain degree that we that we ask those questions yeah yeah I agree asking them from a security perspective what tool they're using then because they then they do have like you know proprietary information about your company or a product that you maybe haven't announced yet or whatever like that's where it gets dicey like how deep are you getting how much transparency are we like demanding of freely insurance and agencies I think like I think what's going to need to happen is it's just gonna become baked into your brand guidelines like kind of like Alton was saying like it as content professionals it's kind of on us to set those guidelines too and the hope would be that your marketing leadership and your team has decided on a tool that you trust and that feels safe to you and then you can then go give any Freelancers or agencies access to that tool so that anything they create for you has to be run through that tool specifically um I don't know how you would monitor that or even like know if they use something else I think that is where it would get tricky for me um because yeah it's like we can give them access to Jasper but they could be using anything they want really without us knowing so I think that's where it would get dicey but I do think that like overall it's just gonna have to become part of your brand guidelines and your will you just have to make sure that anybody you're working with outside of the company has access to the tool that you actually trust yeah trust is going to be major in those instances because you can give someone your information like let's say some Pro information about a new product that you're releasing you're giving it to a freelancer having them build some content for you like you said they could use once it's in their hands they could use any tool they want to build that content even if you tell them use Jasper specifically because it has agreements to where it won't do anything with the data once it has it ingested you don't necessarily know that so there's a trust Factor there but I think on the company side when you're given that information there's a trust factor in like the I think there's a level of um how can I describe it sort of like being optimistically cautious and saying like Okay maybe you don't give them all of the company information that you want to share which is the same way that you wouldn't there's like a lot of security things hurdles that you have to get over just internally in general like not releasing information too soon if you're gonna IPO or anything like that like you wouldn't give that information necessarily to um a freelancer so it's just about being a little bit not like a little being a little bit more cautious and you know in those aspects because there's still there's that gray area where you can give some information but not all of it still get a lot done but you're not also just hoarding all your information because that just slows down content production and things like that yeah I think that's wondering like what we're afraid of like I get theoretically what we're afraid of right like we're giving our information to a black box and then we don't know how this black box then generates outputs for other people who may be talking about a subject related to the information that I gave said black box but sometimes I'm just like does chat CBT really care about me and like my marketing campaign right now where I gave it a tidbit of like proprietary information or you know what I mean those are like sometimes the conversations that happen in my head and as marketers that's like what I try to do a lot is like take the like big headline and dissect it down into like what it actually means for me um and I think a concept that's really interesting as we're talking about these models being black boxes like they're not able to articulate how they came up with the output we just know that they did um is this idea of like an emphasis on explainability on like um it's it's like something people are talking about right now where it's like we really need to get to the model the models in a place where you can say well how'd you come up with that output like what were this series of computations you made or decisions you made that made you decide that this output was the best and not only could it you know help with some of those security concerns that our minds run wild with when we just don't know how it's coming out with this stuff but also in terms of like things like bias and Prejudice and the decisions that it made to come out to that output it would be a lot easier to screen for like this is bad if we were able to know kind of have that explainability factor or that breadcrumb factor yeah no that makes total sense yeah it reminds me of um Alton just wrote a really interesting article on like bias in AI specifically for the lgbtq um community and it was interesting because like it is sort of this weird thing of like you don't know how it's coming up with information that it's coming up with and so there is a privacy concern where like to fight the bias you need to be training it on lgbtq peoples and like their information their opinions like you need to be training it on that stuff so that it has that information to then output but then that's like that's a security concern because what if I'm at work and I don't want people to know that I'm you know necessarily out or something like that then it becomes this weird personal information thing and then I just this weird this weird battle between like privacy but then okay well how do you actually like train these models you know with with taking that into account and doing it safely um way above my pay grade but it is so interesting to think about and and it's it to me it's exciting to know that like behind the scenes of what I do every day um there are people out there like thinking about this stuff and things like explainability and like um I don't know I'm just excited to see kind of what comes from that stuff yeah and Alton I'd love to hear some highlights from that piece on lgbtqia and Ai and like how you thought about that for maybe folks that didn't get to read the piece yet sure yeah one of the the topic Krista mentioned was one of the things that I thought was really really interesting because it in order to remove that bias you have to introduce more um information about like not only the lgbtq like IA Community but people within it and those the level of nuance within like everyone within that community and intersectionality and things like that but you you can't get too specific to out people so I thought that was really really interesting and I love um the gdpr like General data protection rules in Europe specifically because they make it a lot easier for people to opt in and out of having their data used for things like AI model training so I I thought that that was like a cool parallel between those two stories but it was also interesting reading about um and reading about research on there was one research study that was done where some researchers built um this data set called win a queer that tested models for bias against [Music] um lgbtq individuals and like out biased within the outputs I should say and what they did was they built a like data set off of it was around a little over 2 million tweets from members of the lgbtq community and use that to basically fine-tune um Google's Bert model that came out in 2018 which they said was like off the shelf inherently a little like kind of biased um against lgbtq people so with that fine tuning they were able to take the language of people in that community and make the model way less biased which I thought was awesome like being able to take the exact language that these people are using and Infuse it in this thing to where it is way less harmful destructive like way more inclusive of like histories the colloquialisms everything um it just really opened my eyes to the possibilities of that happening across the board and I mean that is already happening in different instances of models like training models to be better in assessing financial data or Healthcare records and things like that so if you can do that in general with you know more techno with with other areas like finance and Healthcare there's no reason why you can't take but a model and Infuse it with more inclusive language and um information about people's backgrounds and things like that so I thought that was great and it trans transcends just generative AI models for Content like there's a big a big section in that article I wrote about um image recognition and how these models can definitely Overlook people who don't look um like a quote-unquote traditional like male or female like what happens if someone is a trans individual what happens if someone is presenting as other something other than a male or female it like looks at your facial structure and can even assess your tone of voice in certain instances so being able to fine-tune models with more inclusive data I think will be a game changer in terms of of curbing that bias and one of the the like biggest takeaways as far as the solutions um to curb in that bias is ultimately just getting more involvement from people in the community in addition to like the fine tuning but just overall in the development stages like when you're building the data when you're when you're building the model itself when you're testing it after it's released when you're getting input on it like get input across the board from members of the community and that will go such a long way to um curb in the bias basically yeah and you know what's what else is resonating with me is I mean it comes down to like every other inclusivity issue we talk about but it's not just getting feedback but it's like actually employing those people to be developing these things and developing the methodology um like I just think it's to even prioritize something like that like in a wide road map of things you have to prioritize while you're building models or building even an application layer and like correcting for some of these things like it it is you know you do need members of this community to be actively employed and engaged in the work otherwise it like I don't think it ever can resonate the way that it should um so that's a point I've been thinking about but um you know I was on a another podcast recently and it was kind of like a free riff session and I loved what we came out with there which was um like the good that AI could potentially do that we don't always talk about like we're so good at talking about the hypothetical bad things that AI could do which is great like we Riff on that all day on like all of the potential like tragedies that are about to happen but some of the good things that we were ripping on were actually really good it was like the idea of even like founding companies right like our boss Megan says this all the time is that like she believes the only thing that has really ever held us back from us and our ideas is our ability to communicate them um effectively and if you think about like the folks that maybe don't have access to some of the community and the influences that allow them to understand hey I can start my own business hey I can build my own model one day or build my own company to be a company that I'd be really proud of and like using tools like AI to be able to get there um is just huge so we were just riffing on that all day on like like what is the future going to look like now that age is no longer a limiting factor like you can get all the access to the information you need now that you know proximity to whiteness or privilege or wealth isn't the determining Factor on if you could be like a founder or not um and all the companies and good that might come out that way yeah I love that I think that's such a good point yeah and I think like I think we're gonna see that we're also very good at staying in our like Tech and marketing bubble but yeah I mean I can see that having implications across industries of of all kinds I know that we've talked a bit about um we talked a bit about like the entertainment industry in in one of our last episodes um and I I know that's also a hot uh button topic and people are scared of it but I think it could do amazing things for TV writers and screenwriters um so yeah it'll be it'll be interesting to see how that plays out yeah I see a lot of that being the like the idea of individual people figuring out the use cases that work best for them like those screenwriters and people like that and even me like my use cases for it as far as it helping me to um like edit content for grammar and Clarity and things like that when I don't feel like bothering one of you or you know somebody else on our team um but then also like kind of back to the idea of trust and trusting that other people are not using it for things they really shouldn't be using it for or that it's not being used in a way that is that someone else's use case is not doing any harm to someone else that it is taking a job away from someone who otherwise should be doing this job like taking a job away from a screenwriter for instance because someone at a major like Film Production Company is like I'm instead of having five writers on staff we're just gonna have one who's just gonna prompt like all of these scripts to be made edit them and then release them you know like give the tools to the five screenwriters and have them do even better work than they maybe were doing before or faster work um so yeah it's I think it's like really really pushing up those positive individual use cases and just trying to educate people and encourage them to like not pull back on the negative use cases and yeah the ones that could potentially harm other other professionals and whatever like sphere of business they're in yeah it really brings me right back to what keeps me up at night that education it is on us to to like help people understand the shortcomings like no no you don't need to replace your screenwriters or your freelance writers like you need to give them these tools because they like they can use them but you still need them um I think we're in like a really kind of slightly messy adjustment period where um you see people who are are trying to figure out how to use them but I do think it's it's for us to step in a little bit and help guide that so that we don't have any of those situations where you're like you were talking about often where people are um unjustly losing their jobs or getting replaced when they really shouldn't be um but yeah absolutely well I guess as we wrap up I'm something I'm curious about myself and none of us we none of us may have the answer to this but I'm curious like if I am at a company wanting to use this stuff it's like what does it even mean for a tool to have good security like is there do we even have like rules around like what is a good security program even look like like if I'm a director of content what do I need to even be looking out for yeah I would assume like let's I mean I know stuff we do like we encrypt data we make sure that like nobody can get in and steal data um I think like there's probably some stuff around um we don't use it to train or anything like that like like is that stuff that makes sense to y'all like that company should be looking out for because I feel like that's a big question that that I see I think all that is spot on yeah encrypting the data and just there being some verbiage of there being an agreement to where we will not like use this data to either train our model or we will not use this data like for anything sales related even or at the very least like giving that data to other companies for whatever reason um which already happens too much as far as like you sign up for a newsletter for one company like a fashion company or something like that because you're buying clothes and the next thing you know you're getting emails from like three or four other fashion companies um so just looking out for language in the agreements that keep stuff like that from happening um and John in the interview made a really good point that from the security side for people working on actually making these tools these AI tools more secure that he personally approaches it and like guides his team from the perspective of the security principles for securing something like this are kind of the same for financial institutions right because finite like our money is Within These Financial assist systems that data has to be as secure as secure gets but those principles can apply to pretty much everything and there's really no reason why they shouldn't when you really think about it but it's just a matter of making sure that those protections are in place and that I guess you're kind of reading the fine print on what their encryption is like and just the more common sense stuff um like that data being shared and I mean you can really get technical and nuanced when it comes to that stuff but I think it's important for the companies that are doing these protections and putting these things in place to kind of explain it in common sense language for the lay people like us to explain really well too so I think reading through the fine print as much as you can in instances like that where you're trying to use a new tool will will go a long way I I think to look for some of those like more common sense security protections yeah sounds like I need to have John help me put together a security checklist to live on our blog yeah let's just send him like the company like any company we're giving our email address to now hey John can you review this really quick just to make sure they don't sell my email address for me please yeah it's so funny I think I'm the only one and again I'm just revealing all my red flags on this um podcast I love when I get retargeted I'm like great I don't like shopping I did do Google Search and I would like to be haunted by you know one brown sweater that is oversized and you can just forget me with every single brand I was about to drag you because as that's like a red flag but I get I am so like convinced by Instagram ads I don't know the last Instagram ad I've gotten where I haven't like pulled the trigger and actually bought the thing and then the swipe up Apple pay forget about it yeah I bought the moment I looked at it yeah it's too easy I'm not even that haunted anymore when I'm like having a conversation with somebody and then something from that conversation pops up as one of those Instagram ads I'm like all right well that makes sense before they used to like freak me out and I'm like no worries I know I'm always like I know it's happening like I know what they're doing to me I'm a marketer but at the same time I need the platform Adidas like I just need them I I didn't know I needed them until this very second but here they are I mean the models built like the precedent is built for like taking someone's interest or search and then retargeting the hell out of them and making a ton of money by doing so by selling like that data point so I could see a world where and we start we're starting to hear about them start popping up I don't know if they have you know enough juice to actually have a viable product they could be out there like the the fashion AI startups and and the beauty ones and things like that where you're able to like I don't know in some world have ai put together outfits for you or put together some sort of like hey this is my style spin up five different outfits with links from the internet on where I could buy those things and then that data goes and then you're being retargeted but again that feels like efficiency to me um uh in my head it's like the scene from Clueless where Cher is using her little AI computer thing to put together outfits that she should wear that's what we need we need to go back to 1995 they had it they had it figured out then I love that well leave it to the three of us to start out with a very serious security topic and end on the topic of Clueless um but I appreciate you all for joining us on this episode and we will see you on the next bye guys thanks y'all see you next time foreign [Music]

As found on YouTube

Get Your Resources Here:

You May Also Like