Dr. Shaista (Shai) Hussain, SAIF CHECK: AI risks, security vulnerabilities, and her journey from medicine to machine learning

[Music] Welcome to another founder wisdom pod. We have Dr. Shya Usain aka Shai. Uh Shai is founder and CEO at Scythe Czech. Uh she's an entrepreneur in residence at Antler, which is probably why I reached out to her. She's a machine learning engineer, cyber security specialist, clinical pathologist, and a medical research and trained doctor. So very uh interesting background.

She's in Riyad, Saudi in Wajaka, Mexico, and Bookride Gypty.com's presenting this podcast. If you want to write a quick book, an AI written book with your LinkedIn post, you can go to bookriteg.com. The first credit is on me. Dr. Shai, welcome to the pod. Tell us a bit more about yourself and what you're up to. Hi, Charles. Thank you for having me. Um, it's a great pleasure, great honor.

I'm super excited to meet you uh and and to discuss uh what I've been up to, I guess. Uh so what I've been up to, I went from a career in medicine into a career in machine learning engineering where I got super excited about uh the quality of everybody's AI. So everybody's been jumping on AI and um I I became quite focused on the machine learning ops pipeline in terms of quality checkpoints and where things can go wrong and what you can do to fix it.

uh and I ultimately discovered three main buckets that affect the the long-term outli um the long-term lifespan of an AI model. uh and the first being governance and regulations which can slap you with like penalties and and anyways the regulations in general are so full of legal jargon that it becomes very difficult for like AI product owners to actually understand what they need to do with their model.

The second bucket I found was in the data and the model itself. everything from bias, fairness, efficiency, choosing the right model and then the last one is in the cyber security which is I'd say the fastest evolving tail end um of the technology. Yeah, it's and the the technologies that have been growing so quickly have have been so exciting that I think a lot of people got caught up in the hype um and forgot that like with internet one we had do you remember the Norton and Semantic we had like those antivirus CDs and [ __ ] Yeah, I remember that.

with AI, we hadn't really thought about what are the what what can the hackers do or what viruses can happen and what bugs and glitches can happen. So that's largely what I've been focused on and and so safe check is that safe check is um a system to do risk assessments, find out where you have vulnerabilities, find out how to fix those vulnerabilities, and then we help you implement them.

Right? they can get your API key and you put a shitload of data in there or at least I do right with Chad GPT and I trust them and I think they have somewhat of a great infrastructure but I think it can get pretty loose with their partners right yeah look nobody's immune so no one company can boast and say we are the most secure because and the reason for that is because like I said this adversarial attacks which is like the hacking it's evolving by the it by the day.

So, whoever's secure yesterday may not be as secure today. Like there's stuff like I I'll give you an old I'll give you a new example and then an older example just to show you the contrast. Yeah. So, in a newer example, there's there's been these malware uh codes that that get injected into an AI system. So, take something like a GPT. All right? You go in and you're like, "Hey, what's the weather today?" someone else can go in and write and change the font to clear so that you don't see what they're writing and they can inject a malicious prompt.

That malicious prompt can actually get to the training data that the GPT was using and it can modify it. So, let's say I injected that prompt and then you go ask what's the weather, it'll tell you you're wearing a red dress. Do you get it? Like it'll it'll give you a wrong answer or a messed up answer. Um and then and an older example in the earliest Chachi PT model there was a guy who and this was unknowingly like unconsciously I guess he was you know people were playing around when it first came out they were telling it to do all sorts of crazy crap people were exploring what the the limits of it were and so some guy told it write poem until infinity

so chachi started writing poem poem poem poem poem for like hundreds of pages and then after I I can't remember how many hundreds of lines of pages. It kept writing poem and suddenly it started spitting out people's like personal information, medical records and credit card. And that did two things. One, it showed that there was a limit to its capability, but two, it revealed some of the training data that had been used in that model and personal information shouldn't have technically been used.

Yeah. No, that's that's pretty weird. I never knew about that one. Let's start with your background first. Like uh how did you get into that? because you have quite a a diverse background, right? Cyber security, medical doctor, AI, like who are your parents? How did they raise you? Tell me a bit more about your your childhood and who you are. Tarzan.

No. Uh I was born in South Africa. So I'm I'm a South African. Uh very proudly so. And and I'd love to have you there sometime when whenever I go. It's top 10 in my destination list, by the way. As it should be. Um, South Africa as beautiful as it is heavily focused on education. I think until now it has one of the stronger educational foundations for most people and and it creates this competitive spirit especially in young people about you must study to achieve and I think that happens in a lot of um sort of developing or poorer countries at large where or where there's advent poverty you know that you have to work hard to achieve something so that's kind

of instilled in you from from a young age um and then I traveled the world studying. So I lived in Canada. I did my high school there. I went to the University of Michigan for undergrad. And then I started using traveling for education like you you want to travel, you want to study. So yeah, culture is learning. It's a whole learning on its own. So then I did med school in Ireland and then I went to the states and I was specializing and then I came to back to the Middle East.

Uh and I've been living I lived in the UAE for some time and then I came to Saudi now like 15 years ago. Um so that's the the life story and um in terms of the career change there's a there's a logical explan there's a logical flow as a doctor as a pathologist you spend a lot of time doing medical research that's sort of part and parcel of what you do and I found myself in medical research in that I found a comfort I was able to do it I

love doing it you know sometimes you like taking a mess of Lego and you like making something out of it or you like taking a mess of puzzle pieces and you like constructing it. That was what medical research is was and is for me. I like taking those pieces and constructing it and I know how to do it. But in that process, you work in medical research there's different components.

There's like the writing component and then there's the data component where you work with numbers and variables etc. What I was finding when I was trying to benchmark my research was by and large in medical research, there's a weakness in in the data. So you might read a headline that's like, wow, okay, they just discovered that this and this causes that and that.

Cool. Um, and then when you go to the data section, you're like, did it really cause it or was it just associated? You know, and and was that the right statistical test or could there have been a better one? So I became hyperfixated on what I'll call the ground truths or finding ground truths to make sure that the the research was saying what it was and what it wasn't with with equal bias saying okay we didn't find what we were looking for we found what we were looking for and these are the numbers to prove it or these numbers were so closely associated that something seems

off and as I got to that point I had to explore yeah statistics but more data science data analytics. So that required learning a little bit of coding, some Python, some R, Pearl, a few of these languages. And I had reached a point where I enjoyed it so much. I enjoyed building software so much that I started studying that on the side whilst being a consultant pathologist.

Um, and at some point at the tail end of data science, you inevitably landed machine learning. It's sort of like the next logical step. Yeah. uh and within machine learning I I started enjoying it. So I studied it. I at a certain point in your career you you can manage a career and studying and you should I mean you should always be learning. Yeah. So instead of learning more medicine I learned how to do machine learning engineering and in that process I said there's some gaps there's this thing is great and it's going to take over the world.

I was sure of this 10 years ago. I said this thing is the next big thing. And when applying it to pathology models, which I was doing at the time, diagnosing different cancers from blood or from tissue, I said this thing, if the data is managed appropriately, cleanly, properly, it can by far exceed what we can currently do. Oh yeah. And there's no excuse for patients on the other end to be waiting um delayed responses or delayed reports or unsure reports or human error.

I'm like all of these problems that we currently face the bulk of the errors that are happening would be gone in a second if we use data appropriately. Um seeing that I said two things one I need to follow this path but two I need to make sure that this path which is inevitable happens with the right data happens with the right attention to make sure errors don't happen otherwise we follow the same biases and bad habits that we'd been doing previously um and so that's that led to this that led to safe check

I said you know what if no one else is going to do it I'm I'm going to build a simple easytouse system that we can check the data, we can check the model and we can know when is when a model is safe. So that's why safe check, right? And I can relate to pretty much all your background, you know, like most people would be like, whoa, what the [ __ ] is this, you know, but uh yeah, to me it's it's a it's a similar background like I relate to to a lot of it.

Um what I got from this though, like the main insight, you had a solid base, right? I'm not sure who your your parents were, but you grew up in South Africa. You mentioned Tarzan. So, they must have gave you like some degree of freedom, right, to explore and to AB test stuff. Very libertarian. You you got to trust the world. You were curious. You learned by your own.

They probably inculcated you the the value of education. And from there, you could scale, right? So, um what's what's that education model? Do do you feel I'm about to have kids. I'm I have a a school for for kids, an AI school for entrepreneurial kids at called CO Uni. What kind of education should I teach them at a young age? Uh according to what you had, which was obviously successful.

First of all, congratulations and wishing you all the best. Um I I do believe that whatever changes and things are changing and people will follow certain habits. There's foundation that lead you to truth. So there's certain logic processes that you should learn at a young age. So it's applying logic or rewarding logic. And if if the natural logic works combined with intuition.

So logic and intuition I think are two things that should be always focused on and then that can be expanded. So, if you're looking at a highly intense academic school, they're teaching you logic through experience, through writing a million essays, through taking math and calculus. Um, and and the permutations of doing that inevitably impress upon your brain logic systems.

Okay. There's a logic logic if this then that. It's and mostly like personal uh experimentations, right? I would say yeah personal experimentations but learning how to arrive so that if this then that requires um the ability to see things from a zoomed out perspective and from a big picture. So you say what's our objective why is that our objective and what's the goal and you follow that path to that goal.

If it's this, you go this way. If it's that, you go this way. And the logic is what's it's a decision tree, but the logic is what's the straightest path to get there, right? And what can you eliminate and what is necessary and unnecessary? But combined with that is intuition. And intuition doesn't have a role in that diagram. But intuition is that ability for you to trust your your heart.

It's quantum physics. I think the heart is quantum. I mean, um, in the gut also, and there's brain cells in there. There's brain cells in there, right? I guess in the heart cell potential. Yeah. You have you have nervous cells everywhere. You have neurons everywhere. Um, and I've read recently that there was some studies, I'll find it and send it to you later, but that there's like an electromagnetic field that surrounds your heart.

And if you you said you meditate. Yeah. So you you there's a meditation that I learned from a lady at a place called HeartMath and I loved it and she taught me how to focus on my heart uh and you know think of things that I'm grateful for and in doing that when in practicing that meditative state for a while I I discovered something amazing I'll share with you and this might help guide what you were asking.

when you're in class or you're writing code or you're writing research, you I at least I'm not sure who else it applies to, but I have sort of a voice that guides me in my head and says do this, this is this and you know it's the voice when you're reading stuff, right? And so you follow that intuitive voice outside of the classroom. I'll have a thought and I'm not sure like let's say in a social scenario or in a you meet somebody at an event scenario.

Should you say your shirt is so cool or should just say nice to meet you? Do you get it? Like you don't know which way to go. Follow your intuition. I would guess you follow your intuition. Yeah. I say whatever pops to my mind and I have no filters. Great. The thing is what this lady taught me is that your ego and your intellect sound the same and live in the same apartment in your head.

Oh yeah. And that ego will guide you wrong. Yes. And she she challenged me and she said, "How do you how can you separate them or how can you distinguish them so you don't screw up?" So I sat with it for a couple weeks. I came back to her and I was like, "Dude, I really don't know and I've been banging my head on the wall about it." She said, "You can't.

you'll never be able to separate them. And so the solution is not to try to separate them, but outside of the academic environment for me, she said, focus on your heart and follow your heart so that when you make a decision, you make it from a place of unconditional love. And for me, I I'm a huge animal. I'm animal before human person. So my my sort of grounding homeostatic state or state of coherence is when I when I have my pets with me or when I'm with the animals that I've rescued outside that gives me a state of love.

So I stay in that mood and then I make a decision and it has never misguided me. Yeah. I have five dogs that Yeah. They're I call them love machines, right? Uh they're always in a perpetual state of love. always wagging their tails, you know. Yeah, I I freaking love them. I had one of these events yesterday happened to to me when the the ego checked in, but I think I could control it, right?

Like um my uh neighbor's son came to train on on my roof gym that I have here. And um yeah, sort of making a friend. He's the same age. And uh he asked me like what I did, right? And he didn't know what cold email was. A bunch of Mexicans don't know what cold email is. So I tried to explain him and uh I think he he thought like I was doing spam or something like that and then I was like should I just double down on the explanation or is that my ego kicking in trying to explain to him how successful and and good I am.

I just said no just silence is better in that case. I also had um a friend of my wife ask me to look at a document in French, you know, because I speak French also. And I'm like, dude, just use chat GPT. Uh which I did for him. And I sent him the the reply to that to to be helpful. And then he's like, uh he's citing like Jackma in 2019. Jackma saying that AI won't outperform humans and all that.

And then I'm like, no, I'm not going to get into that debate, you know? Like it's it's just like control your ego. Don't snap on him. And yeah, last but not least, you stay silent. Yeah, I stayed silent. Sometimes just better to stay silent. And yeah, this morning I took my my dogs out. There were like five other dogs guarding a property and sort of got into an altercation.

Kept my cool. So I think it's it's it's like having serotonin and and dopamine and endorphins in your system. Like if if there's like cortisol, I think the ego wants to kick in. It wants to protect and defend. What do you think about that? 100% uh I do think that so there's definitely um a chemical and cellular state change that happens when when you're in that fight orflight state you know you have this autonomic nervous system which is sort of the the middle of your body and then vag nerve right uh the vagus nerve is part of the nervous system yes certain nerves um and and so you have this let's

go like this you have this autonomic nervous system consider it like your skeleton And then compounded on it, you have the parasympathetic and the sympathetic nervous system. Okay? And that's throughout your body. The hormones that are related to your parasympathetic, they chill you out. They let you sleep, have fun, relax, a bunch of other body functions.

The sympathetic system is the one that kicks in when you need that adrenaline or that fight or flight. The fight or flight or flight, fright or flight. So if you're afraid or if you're scared. So it's sort of more sometimes. Yep. Whatever uh state you end up in, you can switch. You have that ability to go back and forth, but it's hard. Yeah, it can be hard.

But I think like you do that iron man run and that's hard. But you trained yourself and and you managed to do it and that requires a balance of both. So that requires you to have the strength performance which is a sympathetic action and it requires you to have the calm focus that meditative state that flow state which is parasympathetic. Yeah. control the breath work, calm yourself down because your body is crap after some time and you're like, "Yeah, you need to keep going.

Just breathe, calm down, chill, not yourself, right?" So, you've actually mastered it. I mean, um, that's one way to do it. I think any art that you do, if you if you paint, if you whatever it is that you do, whatever focus you do, you do tend to master the game of balancing those. You can take the best of this and the best of that and find that flow state as you described it.

And there that is a completely hormonal game. That's And how do you do that on a daily basis with all this stuff that you do? How except the pets, right? Like do you have a a routine to calm yourself down and put yourself in that flow state? Literally just what I told you that state of coherence that that this lady taught me. She's like at all times go back focus on your heart.

Think of a state of love. stay in that gratitude moment. And I learned to train myself over months to go into that state. So I literally imagine my pets on me and I feel them and then I get happy because I love them obsessively. Um and I just think about them and I feel the love and then they re they reinforce it all day. The whole day that I'm with them, they keep reinforcing that love.

So I I'm never short of feeling love. And and then I stay in that state when I have to make any decision. I don't anymore think should I, shouldn't I? I go into my heart. I think of the love and I say whatever resonates with the state of love that I feel I go with. And whatever does not resonate, I avoid. Oh man, I love that. Yeah. It's like finding your center, right?

But what when adver what about when adversity shows up and you fall on your face in the stairs? The first thing you do is you end up back in your head and you start having those thoughts and then the moment I recognize that. So that was the second set of step of training is recognizing that you you've reached a state out of coherence is to go back into the coherence.

So I and it happened to me yesterday like yesterday I fell out of it. It's not like I'm a guru or some [ __ ] I mean yesterday I fell out of it and I was stressed. I'm like, "Oh my god, how's life going to be? Uh, can I really do this? Will I be successful? Am I should I just have been a student forever?" I had all sorts of thoughts like a monkey. Yeah.

No, but I'm saying it happens to everyone. Do you get it? Like you you get those thoughts and I swear to God both my I have Yoda, Chewy, and Obie, okay? They're my babies. Yoda and Chewy both came one sat on each lap. And I couldn't I could not continue thinking about those thoughts because he was purring and Chewy was licking my hand and I was so excited and I forgot about what I was thinking about at the time despite how heavy it was at the time.

And I tell you like before I went to sleep I was smiling like I remember when I went to bed I had a smile on my face and I was like is it normal to go to bed with this smile? Like I'm so in love with them. I woke up with the same smile and I I I remember the thoughts but I'm not bothered by them. Oh man. Yeah, that's the that's incredible. I wish I could say the same.

Um I put myself to sleep with GPT generated stories nowadays. I do like one technical piece, right? Like a deep research just for my IQ to be satisfied and to check boxes. But then after that it's like it's always a survival story that gets me to sleep. Like this guy got lost in the mountain, found shelter, and then emerged triumphant. I don't know why, but this gets me to sleep and and in the cold, right?

Because it's a hero story. Yes, it is. And survival, right? Like not necessarily driving, but just surviving. Dude, I think that story what you just said the whole hero survival thing it's it's not abnormal if you think about like I've I've studied quite a number of religions but I can't speak to all but from the the several books that I've read that's the stories they're just stories about people who survived who triumphed who were heroes and usually against a number of odds that weren't stacked in their favor.

Those are the stories that are supposed to teach us lessons and motivate us. So, it's not odd that you would hear those stories the same way other people read the Bible or read the Quran before they sleep. Do you know it? It's supposed to teach you those lessons. Amazing. Yeah. And hope. You're injecting hope, right? And is it isn't that like what starting a startup and running a company's about?

You know, like let's be fair. This is a it's savagery. It's a jungle, right? And I like it that way. I couldn't live in another way. You know, like me, I'm addicted to the game. I gified the full process. And I also think it's going to scale in the age of AI. I do think like most workers are going to be forced into entrepreneurship in the next couple of years.

What do you think? I think there's no choice. Uh, and you've you've heard some of these WF and future tech forum talks and and we know where the AI and the quantum computing is headed at least for some near future. We know that that is largely going to replace a large number of jobs which means inevitably if you want to have a chance to have a grasp on some success in the near future, you do need to you need to be owning a business.

You need that entrepreneurship. If it's not for you, there are going to be other jobs that are human related like caring caregivers, storytellers. There's there's still caring for human and caring for society and caring for nature and animal jobs that will still exist, but the the landscape is going to change and entrepreneurship is definitely up there.

But speaking to your point about loving the game, um, you know, there's that whole hologram simulation theory like that we're in a game where I'm still 85% confident about that one. Uh, how could we hit AI, right, which is about to completely change humanity? Like what are the odds of this happening? I think our brain are just not smart enough to think about these things.

But that's the thing, we're not smart enough. But if you take like in the late 90s, if someone told you, yo, in 20, 30 years, you would be able to type in on your phone something and it could produce your science project image for you in 5 seconds. Yeah. You'd be like, "Shut up. That's that's impos my brain can't conceive that." And you would deny it.

Okay. Likewise for the next few stages and and yet we've reached this I think that the rate of growth has accelerated from analog to digital. It's not we can't stop it. Yeah. It's almost a straight line now like now we ask we demand and on the spot we get an answer you know like it's things are happening faster. Um, with that said, it's not far off to argue that with that kind of technology that has come from our hands that we we don't live in a game scenario anyways, right?

Like, have you seen the GTC GT uh uh Grand Theft Auto 6 trailer? No, I wouldn't want to see it. It's It would be freaky. like GT GTA is already like quite uh violent and scaring so many ways. Now I'm I'm pretty sure you're going to say they added AI onto that. Yeah, they did. And and the graphics alone it it's not a far reach to say that like I think that in the in 20 years we won't know the difference between that and actual like is that you?

Is that me? Yeah. Yeah. It's scary. Let's keep that for round two podcast because there's a guest waiting. Shai, where can people find out more about you? Uh LinkedIn uh safe check.ai uh and the links are there and uh or Google I mean you can Google Dr. Shai Stephson Dr. Shai um and it was so good to chat with you. I wish you every Thank you. So is shy.