00:00:00
Speaker 1: You've got a spreadsheet full of customer names and phone numbers, or a blood test result you don't quite understand, and you're about to paste the whole lot into AI. Dan you pause, should you? I mean that's a fair hesitation, because once something's uploaded, it's sitting on a server somewhere and getting a back isn't simple. In this episode, Neo and I work through what's actually safe to give AI and what's better left out. You'll learn how to check your settings so the free tools aren't training on your data, how to de identify sensitive things like health results and contracts, the rules that change once you're at work, and the one question to ask yourself before you upload anything. By the end of this episode, you'll know how to get the same great results without handing over more than you need to. Welcome to how IAI with me Doctor Amoutha Imba and Neo Applan, head of Inventium AI. Each episode we share one practical way to use AI better at work and in life. No fluff, no tech jargon, just things you can use straight away. Neo. One of the most common questions I think we get in our workshops and keynotes that we do at inventium AI is what shouldn't tie upload to AI. So that's what we're tackling today, and let's start with whether the answer depends on what platform you're using.
00:01:43
Speaker 2: And whether you're paying for it, and whether you've locked it down as well. So if you've got a free version of some of these tools I'm talking about Gemini and Claude and Copilot and things like that, do go into your settings to see whether you can click off the thing that says use my data to train the model and things like that. Often on the free versions you can turn that off. Sometimes, say with Gemini, you cannot, so find the tool that you're using and just automatically do that. They do not need to use any of the information you're putting in there to train their models. But with that, once you've got that, or maybe you're paying for the tool, then yeah, you can absolutely upload information up there and it will be private between you and the company. So private between you and say claud or Anthropic as an example, And that's great. You can do that, and certainly you can upload files and it's going to be private between the two of you. But here's the thing, do you really want your say, health records. If you're going to upload health records, do you want your health records to be sitting in a server of Anthropic in America. That's up to you. Maybe the answers yes, maybe the answers no. Do you want to have a conversation with AI and say, here's all the psychological problems I'm having at the moment, and for that to be then available in the cloud and maybe someone might be able to access your computer and even see your previous chats. That's up to you as well. What I personally like to do is I like to de identify that. So I've had blood tests before, and I've wondered, is you know a seven point three million miles per liter are good or bad for that particular thing? Just give it that, don't give your whole blood tests. Just say this is a number should I Is that interesting? Or is that above or below what it should be?
00:03:27
Speaker 1: I want to get into how it's different between the platforms. Let's assume that we've got a paid subscription, what then should we be thinking about in the context of work and what we should upload work.
00:03:44
Speaker 2: It's a bit different to home because if it's your stuff, then of course you've got a choice to upload it to these services or not. Work is really different because you're then beholden by whatever your company is saying you need to do. So, if you've got a company and they're saying they're going to go copilot, for example, that means you can't use your home copilot where you're logged in with hot mail to send your company information too. You can only use the work copilot when you're logged in with your work account. So that's really important, And a lot of companies are saying used to allowed to use this one tool, not many tools, because they don't want all of their client customer data just to be spread everywhere, or their company secrets and things like that. So the basic rule there is, if your company has given you one, two or three tools, only use those one, two or three tools and be logged in with the accounts that your company has let you be logged in with.
00:04:34
Speaker 1: Now, a lot of clients that we do work with have an AI use policy or an AI strategy, but there are still quite a lot of organizations that do not. So just say I'm working for an organization, maybe they do or they don't have an AI use policy, but I have no idea what it is. What should I be doing in terms of what should I feel comfortable giving to AI at work and we'll talk about personal stuff in a moment, and what should I not be uploading?
00:05:02
Speaker 2: No policy means you can do what you want, right, there's no policy, no, obviously not. The first thing I'd look at is your privacy policy, believe it or not, because how you deal with your client, customer data and even partners and suppliers and whatnot, that'll be covered in your privacy policy. And with AI you do not get to get out of jail free card with privacy, So you've got to secure your customer, client extra data, just as you would with any other piece of information because you're using this as any other tool. So look at that first. If there is no policy, if there's a policy, what we found in some companies, and we've have very large companies where people read the policy and they one department thinks you can do it whatever you want with it, and another department reads the policies. Oh no, I can't use much AI. So I encourage companies to actually give really good examples and guidance on what you can and can't do to adhere to that policy because it is hard to understand, like don't be an idiot is effectively what most of the policies say. But there's my version of don't be the idiot's very different to your version of don't be an idiot. So a couple of things i'd say here is if you're in a regulated industry, so if you're in healthcare, finance, all of those kind of things, then there are absolutely some very big, hard and fast rules there and you will definitely need to follow those. So some of them are data to sovereignty. In other words, all of your data needs to sit in an Australian server and things like that. Under certain information bits of data and whatnot, there will be other things there which is kind of like what your company thinks and expects. So, for example, tax file numbers in Australia, they're like the social security number, like this is very very important information and it's not something you want to just spread around. And so some companies are saying, look, there's some information that doesn't really belong on AI, and I'd say that tax file numbers is definitely one of those. I'd say that people's bank account details doesn't really need to go to AI. I don't understand why I need to have a conversation with AI about those kind of bits of information health records, client, those client things like that. Those things probably don't belong there. Even if you could put them on the policy, I'd probably say not, it's probably a good idea just to leave them alone. But really find it what the policy is. If there is a policy, then absolutely try to follow it.
00:07:21
Speaker 1: What about things like you know, if you've got a big customer database like names and email addresses and phone numbers, should I upload that to AI? Yeah?
00:07:32
Speaker 2: I had asked a question, why are you doing that? So if you're doing like analytics, then do you really need the name or the phone number for that? Maybe you just need that there is a record with these people, and maybe their postcode is the only thing you need to get to be able to get some analytics out of it. So what information can you redact, effectively remove or strike out? I use it say with contracts. So if there's a contract there, then AID does not need to know who the contract is between me and the other company. But I'm quite happy to get excerpts of the contract and get it to review that. No problems whatsoever. So it's not a big task for you to go through and just remove some things. It or be columns in a spreadsheet or rows on a word document or something.
00:08:17
Speaker 1: So on the work front, NEO, what do you think about with what you're uploading to inventis AI tools, Like what would surprise people about what you're say redacting versus what you're really comfortable uploading.
00:08:33
Speaker 2: I generally don't like to put any non work personally identifiable information there. So I'm comfortable, for example, it to look through my emails because I'm in copilot already touches my emails and it's already there, and so picking up that I sent an email to Amantha Ember on the fourteenth of April. Fine, happy days, no problems there. These work details are great. But if I had client information, and we don't, we're really lucky. We don't have people's home addresses. We don't need your home address to give you AI training. So but if I had home addresses, that wouldn't go there. Certainly with staff stuff, I know we don't do this, but we don't have HR related you know, tax and pay type things going into AI that just doesn't belong there. At all. So those kind of really personal they call it PI for Personal Identifying Information and SPI, which is sensitive personal information, whether it be like health records, union membership, those kind of things. As a general rule, ask yourself like, do I really need AI to have this to be able to then get the information out of it? And I'd say good, ninety percent of the time the answer is no. And for that last ten percent of time, it's like, can I get a similar kind of output by redacting it? Other things that I don't generally put in there are things that have the company's details. So if I'm working on things for some clients and I'm giving a lot of background to what the client's back problems are and the issues and all those kind of things, I just say I've got a client that has this. I don't say Acmeincorporated or road Runner Industries or whatever. Yeah, I'm showing my age here. Yeah, But I don't need to give it all those names to be able to get that information out. On a personal level, if you're going to use AI for coaching, counseling and things like that, or health records, it doesn't need to know that it's actually your health record or your coaching or counseling things for some of these sensitive things. So redacting is a really good way to go.
00:10:32
Speaker 1: So let's actually shift into personal So what things would you on your personal AI that you use? What do you deliberately not share or redact there?
00:10:44
Speaker 2: I've given it health things before, but I don't give it the entire piece. So I will say, here's a blood test result, not necessarily saying mine. I'm saying is it above or below the iron levels that I need to have for those kind of things. For bank statements, I don't give it bank statements. But what I have done in the past is I've done an analysis on things where I haven't got the dollar figures or the actual bank statements, and instead I've said, here's a bunch of transactions. I want you to categorize. Those transactions might be how much my spending on food versus how much am I spending on entertainment. So those kind of things are fine because if I'm spending twenty dollars on that or forty bucks on that, I'm fine with that. But it doesn't need to know my bank account details.
00:11:30
Speaker 1: And then how about with things relating to mental health and using the AI as a you know, a therapist like tool, because I know that you've built an agent that is like that for you. What do you choose to share and what do you not share? There?
00:11:48
Speaker 2: That's a wider discussion, and I love using it for this, and I want people to think of AI as a potential buddy or confident or someone you can dump a bunch of stuff on and it doesn't complain. But it's also not great if you're in a moment of crisis. So first off, if you're in a moment of crisis, then definitely humans are the absolutely best for those kind of things. But it's great for Hey, I had this situation, do you reckon I managed it well or I'm second guessing myself on a you know, help me see it from different angles. AI is amazing for those kind of things. And I would say that that's not highly highly personal, Like I'm quite happy to tell other people that I had a moment of questioning and whatnot, But it's not that I'm I'm having a horrendous time and I'm you know, I need serious, serious help. Those kind of things I would definitely get humans to help me with. But you will find that some people open up more and more to AI, and it's a slippery slope because you feel it's really it knows a lot about you, and it's quite helpful on those kind of things. It's not a problem from a privacy kind of perspective, but it's more about how can you use AI so that it gives you the right steer because a I does like to agree with you and go, oh, you're so right. I don't want someone say you're so right if I'm absolutely crazy and about what I've said, like it's a stupid idea, Neil, I want someone to tell me that was a stupid idea. Of course you should have given someone the seat or whatever it is. So having it as a challenger sometimes is really useful. So yeah, for counseling things, your miling is is going to be varying, but really watch out for those yourself.
00:13:21
Speaker 1: If someone is just going to do one thing differently based on just being more sensible with what they upload to AI, what would you recommend out of everything we've chatted about?
00:13:32
Speaker 2: Question? So before you upload something, before you give something to AI, ask what am I trying to get out of it? Out of the AI? What kind of benefits am I going to get out of it? And then question does it need to have all that information in order to give me some or most of those benefits? So often you can remove the heading, the title, the customer, the client, that row of everyone's phone numbers, and you can still get some amazing insights out of it, and it doesn't take that much extra work. So just question it doesn't need all of this stuff, and if it doesn't, just remove it and you'll probably find you're getting a great result.
00:14:09
Speaker 1: I love that. Neo, thank you so much for helping us all use AI a little bit more safely and securely. How AI was hosted by me Amantha Imber and Neo Applan. A big thank you to Martin Imber who does our sound editing, and Jim Rubio for production support, and thank you to John Kilby who composed the theme music.
Speaker 1: You've got a spreadsheet full of customer names and phone numbers, or a blood test result you don't quite understand, and you're about to paste the whole lot into AI. Dan you pause, should you? I mean that's a fair hesitation, because once something's uploaded, it's sitting on a server somewhere and getting a back isn't simple. In this episode, Neo and I work through what's actually safe to give AI and what's better left out. You'll learn how to check your settings so the free tools aren't training on your data, how to de identify sensitive things like health results and contracts, the rules that change once you're at work, and the one question to ask yourself before you upload anything. By the end of this episode, you'll know how to get the same great results without handing over more than you need to. Welcome to how IAI with me Doctor Amoutha Imba and Neo Applan, head of Inventium AI. Each episode we share one practical way to use AI better at work and in life. No fluff, no tech jargon, just things you can use straight away. Neo. One of the most common questions I think we get in our workshops and keynotes that we do at inventium AI is what shouldn't tie upload to AI. So that's what we're tackling today, and let's start with whether the answer depends on what platform you're using.
00:01:43
Speaker 2: And whether you're paying for it, and whether you've locked it down as well. So if you've got a free version of some of these tools I'm talking about Gemini and Claude and Copilot and things like that, do go into your settings to see whether you can click off the thing that says use my data to train the model and things like that. Often on the free versions you can turn that off. Sometimes, say with Gemini, you cannot, so find the tool that you're using and just automatically do that. They do not need to use any of the information you're putting in there to train their models. But with that, once you've got that, or maybe you're paying for the tool, then yeah, you can absolutely upload information up there and it will be private between you and the company. So private between you and say claud or Anthropic as an example, And that's great. You can do that, and certainly you can upload files and it's going to be private between the two of you. But here's the thing, do you really want your say, health records. If you're going to upload health records, do you want your health records to be sitting in a server of Anthropic in America. That's up to you. Maybe the answers yes, maybe the answers no. Do you want to have a conversation with AI and say, here's all the psychological problems I'm having at the moment, and for that to be then available in the cloud and maybe someone might be able to access your computer and even see your previous chats. That's up to you as well. What I personally like to do is I like to de identify that. So I've had blood tests before, and I've wondered, is you know a seven point three million miles per liter are good or bad for that particular thing? Just give it that, don't give your whole blood tests. Just say this is a number should I Is that interesting? Or is that above or below what it should be?
00:03:27
Speaker 1: I want to get into how it's different between the platforms. Let's assume that we've got a paid subscription, what then should we be thinking about in the context of work and what we should upload work.
00:03:44
Speaker 2: It's a bit different to home because if it's your stuff, then of course you've got a choice to upload it to these services or not. Work is really different because you're then beholden by whatever your company is saying you need to do. So, if you've got a company and they're saying they're going to go copilot, for example, that means you can't use your home copilot where you're logged in with hot mail to send your company information too. You can only use the work copilot when you're logged in with your work account. So that's really important, And a lot of companies are saying used to allowed to use this one tool, not many tools, because they don't want all of their client customer data just to be spread everywhere, or their company secrets and things like that. So the basic rule there is, if your company has given you one, two or three tools, only use those one, two or three tools and be logged in with the accounts that your company has let you be logged in with.
00:04:34
Speaker 1: Now, a lot of clients that we do work with have an AI use policy or an AI strategy, but there are still quite a lot of organizations that do not. So just say I'm working for an organization, maybe they do or they don't have an AI use policy, but I have no idea what it is. What should I be doing in terms of what should I feel comfortable giving to AI at work and we'll talk about personal stuff in a moment, and what should I not be uploading?
00:05:02
Speaker 2: No policy means you can do what you want, right, there's no policy, no, obviously not. The first thing I'd look at is your privacy policy, believe it or not, because how you deal with your client, customer data and even partners and suppliers and whatnot, that'll be covered in your privacy policy. And with AI you do not get to get out of jail free card with privacy, So you've got to secure your customer, client extra data, just as you would with any other piece of information because you're using this as any other tool. So look at that first. If there is no policy, if there's a policy, what we found in some companies, and we've have very large companies where people read the policy and they one department thinks you can do it whatever you want with it, and another department reads the policies. Oh no, I can't use much AI. So I encourage companies to actually give really good examples and guidance on what you can and can't do to adhere to that policy because it is hard to understand, like don't be an idiot is effectively what most of the policies say. But there's my version of don't be the idiot's very different to your version of don't be an idiot. So a couple of things i'd say here is if you're in a regulated industry, so if you're in healthcare, finance, all of those kind of things, then there are absolutely some very big, hard and fast rules there and you will definitely need to follow those. So some of them are data to sovereignty. In other words, all of your data needs to sit in an Australian server and things like that. Under certain information bits of data and whatnot, there will be other things there which is kind of like what your company thinks and expects. So, for example, tax file numbers in Australia, they're like the social security number, like this is very very important information and it's not something you want to just spread around. And so some companies are saying, look, there's some information that doesn't really belong on AI, and I'd say that tax file numbers is definitely one of those. I'd say that people's bank account details doesn't really need to go to AI. I don't understand why I need to have a conversation with AI about those kind of bits of information health records, client, those client things like that. Those things probably don't belong there. Even if you could put them on the policy, I'd probably say not, it's probably a good idea just to leave them alone. But really find it what the policy is. If there is a policy, then absolutely try to follow it.
00:07:21
Speaker 1: What about things like you know, if you've got a big customer database like names and email addresses and phone numbers, should I upload that to AI? Yeah?
00:07:32
Speaker 2: I had asked a question, why are you doing that? So if you're doing like analytics, then do you really need the name or the phone number for that? Maybe you just need that there is a record with these people, and maybe their postcode is the only thing you need to get to be able to get some analytics out of it. So what information can you redact, effectively remove or strike out? I use it say with contracts. So if there's a contract there, then AID does not need to know who the contract is between me and the other company. But I'm quite happy to get excerpts of the contract and get it to review that. No problems whatsoever. So it's not a big task for you to go through and just remove some things. It or be columns in a spreadsheet or rows on a word document or something.
00:08:17
Speaker 1: So on the work front, NEO, what do you think about with what you're uploading to inventis AI tools, Like what would surprise people about what you're say redacting versus what you're really comfortable uploading.
00:08:33
Speaker 2: I generally don't like to put any non work personally identifiable information there. So I'm comfortable, for example, it to look through my emails because I'm in copilot already touches my emails and it's already there, and so picking up that I sent an email to Amantha Ember on the fourteenth of April. Fine, happy days, no problems there. These work details are great. But if I had client information, and we don't, we're really lucky. We don't have people's home addresses. We don't need your home address to give you AI training. So but if I had home addresses, that wouldn't go there. Certainly with staff stuff, I know we don't do this, but we don't have HR related you know, tax and pay type things going into AI that just doesn't belong there. At all. So those kind of really personal they call it PI for Personal Identifying Information and SPI, which is sensitive personal information, whether it be like health records, union membership, those kind of things. As a general rule, ask yourself like, do I really need AI to have this to be able to then get the information out of it? And I'd say good, ninety percent of the time the answer is no. And for that last ten percent of time, it's like, can I get a similar kind of output by redacting it? Other things that I don't generally put in there are things that have the company's details. So if I'm working on things for some clients and I'm giving a lot of background to what the client's back problems are and the issues and all those kind of things, I just say I've got a client that has this. I don't say Acmeincorporated or road Runner Industries or whatever. Yeah, I'm showing my age here. Yeah, But I don't need to give it all those names to be able to get that information out. On a personal level, if you're going to use AI for coaching, counseling and things like that, or health records, it doesn't need to know that it's actually your health record or your coaching or counseling things for some of these sensitive things. So redacting is a really good way to go.
00:10:32
Speaker 1: So let's actually shift into personal So what things would you on your personal AI that you use? What do you deliberately not share or redact there?
00:10:44
Speaker 2: I've given it health things before, but I don't give it the entire piece. So I will say, here's a blood test result, not necessarily saying mine. I'm saying is it above or below the iron levels that I need to have for those kind of things. For bank statements, I don't give it bank statements. But what I have done in the past is I've done an analysis on things where I haven't got the dollar figures or the actual bank statements, and instead I've said, here's a bunch of transactions. I want you to categorize. Those transactions might be how much my spending on food versus how much am I spending on entertainment. So those kind of things are fine because if I'm spending twenty dollars on that or forty bucks on that, I'm fine with that. But it doesn't need to know my bank account details.
00:11:30
Speaker 1: And then how about with things relating to mental health and using the AI as a you know, a therapist like tool, because I know that you've built an agent that is like that for you. What do you choose to share and what do you not share? There?
00:11:48
Speaker 2: That's a wider discussion, and I love using it for this, and I want people to think of AI as a potential buddy or confident or someone you can dump a bunch of stuff on and it doesn't complain. But it's also not great if you're in a moment of crisis. So first off, if you're in a moment of crisis, then definitely humans are the absolutely best for those kind of things. But it's great for Hey, I had this situation, do you reckon I managed it well or I'm second guessing myself on a you know, help me see it from different angles. AI is amazing for those kind of things. And I would say that that's not highly highly personal, Like I'm quite happy to tell other people that I had a moment of questioning and whatnot, But it's not that I'm I'm having a horrendous time and I'm you know, I need serious, serious help. Those kind of things I would definitely get humans to help me with. But you will find that some people open up more and more to AI, and it's a slippery slope because you feel it's really it knows a lot about you, and it's quite helpful on those kind of things. It's not a problem from a privacy kind of perspective, but it's more about how can you use AI so that it gives you the right steer because a I does like to agree with you and go, oh, you're so right. I don't want someone say you're so right if I'm absolutely crazy and about what I've said, like it's a stupid idea, Neil, I want someone to tell me that was a stupid idea. Of course you should have given someone the seat or whatever it is. So having it as a challenger sometimes is really useful. So yeah, for counseling things, your miling is is going to be varying, but really watch out for those yourself.
00:13:21
Speaker 1: If someone is just going to do one thing differently based on just being more sensible with what they upload to AI, what would you recommend out of everything we've chatted about?
00:13:32
Speaker 2: Question? So before you upload something, before you give something to AI, ask what am I trying to get out of it? Out of the AI? What kind of benefits am I going to get out of it? And then question does it need to have all that information in order to give me some or most of those benefits? So often you can remove the heading, the title, the customer, the client, that row of everyone's phone numbers, and you can still get some amazing insights out of it, and it doesn't take that much extra work. So just question it doesn't need all of this stuff, and if it doesn't, just remove it and you'll probably find you're getting a great result.
00:14:09
Speaker 1: I love that. Neo, thank you so much for helping us all use AI a little bit more safely and securely. How AI was hosted by me Amantha Imber and Neo Applan. A big thank you to Martin Imber who does our sound editing, and Jim Rubio for production support, and thank you to John Kilby who composed the theme music.
