203: How to be Human in the Age of AI

Joshua Burkhow: Kai, welcome. director of data and AI strategy at ThoughtWorks. ⁓ ⁓ transformation work inside enterprises ⁓ and you're written extensively human centered approaches to data and AI. ⁓ Welcome to the podcast, my friend.

Welcome to alter everything a podcast about AI analytics and the future of work. I'm Joshua Burka. Now I want you to do a thought experiment with me and imagine it's late 2026 and AI isn't just giving advice. It's shaping decisions and workflows and outcomes.

And the real question isn't how smart AI is. It's whether we've built the human systems to handle it. Tiankai Feng: Thank you very much. It's great to be here.

Joshua Burkhow: Cool. ⁓ You've often that AI problems ⁓ aren't necessarily There's more of a human element to them. They're human. What exactly is being missed in most today?

Just to kick it off. So ask yourself, who's making the call? Who's accountable? Who's affected?

Today's conversation is about humanizing AI, not as some buzzword, but as a practical way to design AI that actually works in this real world we live in. When we talk about AI, it's easy to get stuck at the system level, models, tools, roadmaps, but... Tiankai Feng: Yeah, absolutely. And I think, well, I think it's really human because we act like AI is very autonomous and has its own agency.

But in reality, it's really like we have built it the way we wanted it to be, right? And I feel like if anything goes wrong with AI, it's usually that we as human beings have not taken our role more seriously and we were not diligent enough to actually avoid what was happening. And we kind of just got lost on the way of being excited about technology to forget that serious things can happen about it. Joshua Burkhow: right?

AI doesn't live in isolation. It lives inside human decisions. Most AI strategies start with tools. But my guest today, Tiankai Feng, argues that they actually should start with people.

Because AI rarely fails because of the weak models. It fails because human systems aren't ready. So today, we're gonna structure this conversation around Tiankai Feng: So in the end, it's really about human beings not making the right decisions and being sometimes a little bit reluctant to actually taking AI as a technology seriously. And that's why bad things happen, right?

And that's, think, what's missing. Joshua Burkhow: Yeah. Yeah, absolutely. ⁓ kick off this sort introduction, can you tell folks, you know, what you would say your role is today at ThoughtWorks?

three simple layers. First one is me, how AI changes individual judgment. And then two is we, how it changes teams and collaborations. And then third is they, how it affects people who aren't in the room.

And underneath that framing is a deeper question. What human capabilities must organizations build? Tiankai Feng: Absolutely. So I work as a director for data and AI strategy.

That is a global role, which means I actually am the global lead for setting up the approaches and frameworks of how we solve strategic problems in data and AI. So that's all about even articulating a strategy, but it's all about change management and operating models. It's also about data governance, AI governance, and all these topics, and basically how to solve the people and process side of things before we talk about technology. Joshua Burkhow: if they want AI to succeed responsibly.

Let's get into it. Tiankai Feng: and how does it all integrate into one thing. And I think nowadays it's really exciting that data and AI is hyped more than ever, right? But luckily this year and even last year, it drove more towards data because everyone realized that good data is needed to actually have good AI.

And I think that overall has turned into a really big demand from the industry side for us to help, basically. ⁓ Joshua Burkhow: a bit. That's So one of the things we'll talk about through this podcast, I'm just like over the moon about this because it's probably the thing that I would want to talk about and understand and learn more than anything in the AI space is where the humans come together. Like what happens with the humans?

Because we're the ones that created it, but we're hopefully, we're the ones, I think we're both positive guys. I think we're the ones that are gonna be with it, you know, for a distinct amount of time. sort of at a high level, where does the humans and the AI come together? And know that's a fuzzy question, but it's really like, what's your perspective how AI and humans interact today?

Tiankai Feng: Yes, absolutely. an individual level, would that AI really is a productivity booster right now. That's, I think, where it's established. I think nowadays where AI is being most used in our day-to-day is for ⁓ individual productivity reasons, ⁓ no if it's summarizing or coming up with the text again, ⁓ or creating a picture out the blue and with a few prompts, ⁓ or reviewing stuff.

⁓ All those things have very manual in the past and has Joshua Burkhow: Mm-hmm. Tiankai Feng: cost a lot of time and efforts for us that we can now delegate actually to AI to help us with. this, think, though, is the right way of thinking the working mode of humans and AI, where we have a target state and an outcome in mind, and we use AI to help us to get there faster and more effectively, which is what should happen. ⁓ And even we scale it up to...

like LLMs and agents and what everyone, multi-agent environments, that is still the truth, right? In the end, it's about us setting objectives and goals and not only the why, but also the what and the how for AI to help us to actually reach that. But I think in a way that is where a lot of people are misinterpreting it and it's not happening that way, Which is, for example, again, ⁓ said, looking at hallucinations of AI because there's missing context. and taking that for real and not fact-checking it, for example, or vibe coding, which very hyped, to just take the code, not review it anymore, and realize that it's really very inefficient afterwards because it's not really scalable that way.

⁓ And these things are then rooted in that AI cannot compensate for our missing skills and expertise. We still need that to actually build on it to accelerate and amplify. But if we don't do that, then AI is actually gonna hurt us more than it actually helps us in that. Joshua Burkhow: The, vibe coding piece is interesting because, personally I've created, whole documents of 10,000, you ⁓ know, a hundred thousand lines code that ⁓ Frankly, no one could check, right?

So we use tools, we use tools to do the testing. I'm sort of in this conundrum where I'm like, okay, we can build super fast. Like we can build applications in a matter of minutes now. We can build websites in a matter of minutes, but.

Tiankai Feng: Two. Mm-hmm. Joshua Burkhow: it's almost just as important or more important to have big test frameworks, have the structure, have the sort of because I have a really hard time believing that people are gonna slow down, right? They're not gonna, you know, they're not gonna come and be like, ⁓ yeah, this could be dangerous.

Let's just tone it down a little bit, right? They're gonna be like, hey, I need to do this fast. How do I make sure that whatever I'm building as fast as possible ⁓ Tiankai Feng: Right? Joshua Burkhow: well-tested, validated, secure, all those things.

⁓ priorities compete and that we need to have those. ⁓ I'd say platforms are probably bringing that into play. Tiankai Feng: Absolutely, I fully agree with you. Actually, that reminds me to also think about the human nature of things, right?

Because with all of these preventative or kind of mitigating issues we usually don't prioritize it until something actually breaks. And then suddenly it's top of mind for us, right? But we never want to upfront invest in it. It's like, I don't think that's that matters, right?

Nothing will go wrong. It will probably go right. And then something really big happens and suddenly we go back and like, ⁓ we have to rethink everything, right? This cannot happen again.

Yeah. Joshua Burkhow: Yeah, or if it does fail, are like, well, even it does break, it's not going to be that bad because it's, you know, just go to whatever. but ⁓ we accept AI as it is ⁓ the capabilities that it's presenting itself and humans have to change in relation to new worldview of intelligence and, Tiankai Feng: Mm-hmm. Joshua Burkhow: So there's like these two competing factions in the world AI today, the two big ones that I notice is AI is going forward no matter what, just accept it and humans need to change.

⁓ Or there's this worldview that AI is a thing, but as us humans, ⁓ we should... Tiankai Feng: Mm-hmm. Yeah. Joshua Burkhow: be putting the box around the AI and saying, hey, you can only go this far outside the box or inside the box and no more, know, the regulations and security and all that.

Do you have a perspective on that? Tiankai Feng: Yeah. Yeah, that's a really good question. as you pointed out correctly, we cannot judge AI being good or bad, right?

Because in the end, it is a tool and it's still acting depending on what we as people give us as basically directions. But nonetheless, we do have to consider that there is bad outcomes, either intentionally or unintentionally, right? Intentionally by malicious actors who use it for making bad things with it. Or accidentally, we wanted to do something well, but we accidentally did something bad with it, which is also there.

So as the people who are innovating AI, we need to, unfortunately, be able to predict the malicious acting and the unintentional bad acting the best way possible and make it very hard for them to succeed. And that is, think, where it gets tricky because it is inherently against Joshua Burkhow: Yeah. Tiankai Feng: why we innovated in the first place. Because we innovated usually with a good objective in mind, and we wanted to save the world, we wanted to cure cancer, whatever we had in mind to do it.

And now people are creating autonomous weapons and they're using it for scamming people. And all these bad things that happens, how could we have predicted it that far in advance? And how much could we have anticipated it? But that is where I think the regulation part and guard rating and governance part all comes in.

Joshua Burkhow: Yeah. Yeah. Tiankai Feng: And by nature, not against innovation, right? It's about making it very hard to do bad things with it.

And finding that balance though, in reality, is just very hard between like not over-governing it and still allowing and enabling the innovation part, but to then just cause guardrails for bad people to not succeed. I think the answer to is it depends on how much you involve AI into your decision making and taking actions, right? Because, I mean, we could look at it the more traditional way, let's say like predictive modeling in the AI space. And that would give me some kind of insights, right?

I would see a little bit how the future could look like, but I still as a human being am deciding, right? How much do I believe in this future? And how much do I need to change my plans now to actually make this future happen or not happen, right? That is kind of...

Joshua Burkhow: Hmm. Tiankai Feng: how it would work. But with AI now being more human-like in their communication and also being able to be agentic, which means they are not only informing us more autonomously, but also acting autonomously if we want them to, then that means they can even decide and act on our behalf if we want them to. And that is very tricky because inherently that means that we would trust an agent to be as smart and intuitive as we as human beings are when we look at things and make decisions and act on it, which in some cases could be true.

I think like with a repetitive tedious task that are always the same, let's move from table A to table B and just change the structure. That probably is okay for an agent to do. But let's say making like complicated transactions from a bank account to other bank accounts and it's like thousands or millions of dollars then I don't know if an AI agent should do it on my behalf or if I should be participating in some kind of degree to make sure that it's all happening in the right way.

So that sense, I think it's about how comfortable we are in one hand side with trusting AI to take our decisions while we are still responsible for whatever the AI does versus how much do we want to just keep it to a certain degree of involvement within us. And we are just making peace with that. And this is the mode we're doing it. It all comes down for me into the, we define our relationship with AI, right?

In different use cases, is AI right now just an informant for me, or is it a colleague for me, or is it an intern for me? What is the relationship? And I think that then basically derives ⁓ how we want to do things. Joshua Burkhow: I really like that perspective because it's, know, for those of us who are building AI, don't, I don't know.

think about that. mean, I've got friends ⁓ and family that use it as their ⁓ They use it as their ⁓ guidance how do I handle this or how do I handle this relationship Because ⁓ you and I can use AI at work, but we also use AI at home, right? Or in our personal lives, as you will. And I think this is, as an individual, think there's something powerful about just taking a momentary step back, be a human for a minute ⁓ and just say, Tiankai Feng: Absolutely correct.

Mm-hmm. Joshua Burkhow: How does this relationship and I'm ⁓ bona fide, I use technology for everything. I use AI a lot, but just the evaluation of how ⁓ I approaching AI probably a good thing to have in the back of your pocket. You know?

⁓ Tiankai Feng: Yes, absolutely. Yeah. And I think also with the relationship, it can be different depending on the situation you're in, right? I think we just need to be aware of that and be aware of the risks that we're taking or the risk that we're decreasing because of it and the way we should act then more diligently or less diligently depending on what we're asking it to do, right?

So I think that is the only implication there because in the end, we are still as the AI user responsible for whatever the AI does, right? In the end, we have directed it to do things. Joshua Burkhow: Yeah. Tiankai Feng: And when things go wrong, we as the human being need to take accountability.

Joshua Burkhow: Yeah, this is a hard one to swallow for most people because, you know, the first answer most people say is, don't even know how it works. It just works, right? Like I just put something in there and it comes back with this response that seems pretty intelligent to me. It's like, ⁓ it thought about things that I didn't think about.

And it's teaching me things that I didn't know that are, you know, somehow valuable, but. You you said this earlier in the podcast that you're checking your answers and validating that it's not giving you a bunch of BS to There's a percentage of the population who are using AI. tells them that, or it's saying, you know, ⁓ something that may or may not be true. Tiankai Feng: Yeah.

Joshua Burkhow: it comes down humans keeping the power of decisions. sort of talked to other people about this concept is have... Some would call it decision support systems. We have these sort of tools in place in the work environment when we're working with data and analytics to help us make decisions now decision making as a Field as ⁓ ⁓ topic of study has been pretty well researched how humans Make decisions how they're influenced by decisions how decisions a very human thing that we're talking about here ⁓ Tiankai Feng: Mm-hmm.

Mm-hmm. Joshua Burkhow: For example, they did the study that's pretty famous. but the decisions made by judges in a court, they found that during the morning when ⁓ the judge was, well rested, fresh mind made certain types of decisions in a certain versus you saw the decisions that he made in the afternoon. And so, you this sort of Tiankai Feng: Mm-hmm.

Joshua Burkhow: The question is, does AI cause us to have a sort of decision fatigue because we're now in the process or in the sort of realm of having to make way more decisions than we've ever had to make in our life. Tiankai Feng: Interesting question. I'm thinking about if actually, and that is not meant as disagreeing with you, but I'm wondering if we're actually making more decisions. Yeah, I'm wondering if we're making more decisions in a day today or if we're just more explicit about it because we're saying it out loud, right?

Because I think we are still making as many decisions every day as before. Joshua Burkhow: You know? Please do disagree. Yeah.

Yeah. Please do. Tiankai Feng: It just forces out with AI to be more explicit about it and prompt it out, so to say, right? And to let AI help us with it.

So in that way, I think it's not decision fatigue. What I would say it maybe is, is a prompting fatigue, right? I mean, for those who are actually using AI very heavily, like just explicitly having to write everything explicitly down can feel exhausting, right? Even if you use voice mode and you do it and speak it in, it can still feel exhausting having to...

Joshua Burkhow: Mm-hmm. Tiankai Feng: basically pronounce everything and say it out loud and put it into words, when often it just feels like, no, I'm not going to ask AI. I think I'm just going to decide, no, it's fine. Like, I do it now within one second, and I don't have to spend one minute interacting with an AI to actually do it, which I think could be a new trend too, right?

Whereas some things we don't even want to give to AI anymore because it's more effort than just coming up with a solution on ourselves. But nonetheless, though, on the decision making, what is interesting for me is that AI is imitating human behavior, Joshua Burkhow: Yeah. Tiankai Feng: By the way it communicates by default, it's very friendly and affirming of what you're saying, but at the same time also very confidently wrong, Because it can give you a wrong thing, but it acts like it's the hundred percent truth of it.

And until you tell it, no, that's wrong. And then it says, ⁓ yeah, I'm so sorry. ⁓ course I'm wrong. ⁓ it's the thing where acting like a human being actually is very misleading for us because we then treat the imitated confidence ⁓ ⁓ in facts and reality.

Joshua Burkhow: That's right. Yeah, exactly. Exactly. Yeah.

It's Tiankai Feng: whereas it is not because it's still just a statistical machine, let's say, as I would call it, ⁓ I think we need to just be clear ⁓ that it's an imitation and it's not really human. And that should guide us in the way we make decisions. Joshua Burkhow: Yeah, it's sort of funny. I had a conversation internally at work around how to change the response of voice ⁓ in some of the LLMs and the LLM to not respond in that same way.

Like, hey, I want you to be blunt with me. I want you to be straight with me. Don't be so affirmative, all that. And it's sort of interesting because...

Tiankai Feng: Mm-hmm. Joshua Burkhow: the language that comes out is one thing, but what it's actually doing behind the scenes as far as the reasoning and the processing of the model is the same per se. It's really interesting. ⁓ shift gears into, we sort of talked about a little bit the eye perspective.

Tiankai Feng: Absolutely. Yeah, that's true. Joshua Burkhow: I'd like to get into talk about the Wii perspective and things like, you know, working with teams and the idea that have trust amongst those teams and collaboration in this space as a human. ⁓ see ⁓ AI changing, ⁓ good or bad, how we work ⁓ Tiankai Feng: Mm-hmm.

Yes, think in principle, it would help us ⁓ in the collaboration. ⁓ And I even redefine collaboration of it, where it's not only human to human collaboration anymore, but human with AI, AI with human, and human to human, and AI to AI. So all of that, I think, is the new interaction mode of collaboration, especially when we think about agents and human beings all autonomously acting and collaborating together. And in that way, I think it has a lot of benefits already, right?

It mean just look at, I don't know, transcribers in meetings that are now helping to not only summarize what we talked about, but already giving us maybe like the next steps prompted. And we just have to say yes. So it takes the next steps, for example, but even in other things, right? Where it basically just helps in collaboration and product management tools where it just basically assigns smartly based on the context of emails into the right ways of doing that.

But nonetheless, there are bad things to it too. And those are always the unexpected consequences, There was this ⁓ Business Review study that I saw recently where ⁓ talked about how a work slop, ⁓ AI slop at work, would an impact on human beings. And they mentioned that if they would receive work slop from their colleagues, I think it was more than half would think less of them. afterwards, right?

They would feel like they're not taking their job seriously. They're not respecting them enough. They're not creative. They're not reliable.

All of those things suddenly dropped increasingly. And those are attributes that interestingly are very easy to lose and very hard to gain again, So once you are seen as unreliable, it's very hard to prove yourselves to be a reliable gain because you need to multiple times actually to show that you are in a reliable way. Joshua Burkhow: That's right. Yeah.

Yeah. Yeah. Tiankai Feng: And so I'm wondering how much that is happening at scale too, right? Because we're using it by focusing on the I and the me that we talked about before.

And we don't consider the consequences on the we and the us, Meaning suddenly we as human beings look worse because we're using AI in the bad way and it an unintentional friction in the way we are seen between how think we are looking like by being more productive. But now we're seeing as someone that actually is not as ⁓ great. ⁓ Joshua Burkhow: Yeah. Mm-hmm.

Yeah. It's, such an interesting point because I we all know that moment when send you an email or they put together a or they do something they send it to you and you just know just through experience that that was written with an LLM. And I would say, you know, if you and I were talking last year, Tiankai Feng: Mm-hmm. Joshua Burkhow: I would have been like, That's a problem.

you just cheated, you know, and, cheated being, being relative, but like cheated in the sense that you had the superpower duel do all the work for you when you should have gone through this and, and help you think thoughts, but then you write it. ⁓ Tiankai Feng: Yeah. Joshua Burkhow: And so all this time I'm at emails and I'm saying, yep, they wrote that with chat. ⁓ that one written.

⁓ wait, this one looks like it actually was written by a human. That's pretty cool. Hard to see those nowadays, Tiankai Feng: Yeah. Yes.

Joshua Burkhow: Because it changes the whole dynamic of how we interact with our teammates, how we interact with our leaders. I've gotten emails from leaders that I'm 99 % sure they were through Chat2PT or another LLM. I've actually had to stop myself and think through this. I'm like, Tiankai Feng: Yep.

Joshua Burkhow: If I was at this and I. ⁓ just get rid of the productivity quotient here and the idea that, ⁓ they wrote it to be faster so they could do more. because we're all trying to find a way to be more productive. But I can take my thoughts, however quirky or good or bad or misaligned they are, Tiankai Feng: Yeah.

Joshua Burkhow: put them into AI and AI can sort of check me and validate like, this is, I would say it this way. Hey, this tone might come off of being more aggressive than you, you're hoping for. This doesn't make sense. Rewrite this, all that.

That idea that we could then take that, know, work through AI and it out to our teammates is actually really cherished now. Like I actually, Tiankai Feng: Yeah. Joshua Burkhow: more appreciate that now being my human self most days that see that they've actually used tool to refine their thinking, refine their language, refine their tone, their all this stuff. Right?

Does that make sense? Tiankai Feng: Right. Yeah, absolutely. I do think you're absolutely right because from the outputs of our colleagues and collaborators, we infer what their intention was from before, right?

As you mentioned, they just want to save them time. But there's two ways of saving time, right? One is actually saving time on doing the right thing faster, right? Whereas as you said, it's actually collecting your thoughts and be more succinct and actually be much better in the combination than before.

or let it be like a generic kind of blah blah, where then you realize this is more about, I'm just not important to them. They don't want to write a specific email to me. They just like hit on, phrase this with one prompt and then basically send it to me, which is not really nice. And I think this is interesting.

⁓ also realized one thing where it's, ⁓ nowadays, you don't talk to all colleagues in person or in virtual meetings anymore, right? You just only talk via chat and emails ⁓ you don't... I mean, they could use AI for all of that communication now because it's like embedded into all of these applications, right? and then ⁓ find myself actually being cautious when I meet them then for the first time in person ⁓ feeling, let's see if they actually like that and actually say it in person like they do it in the emails.

Or do I have proof that actually they used AI for everything and they are a whole different person in person, right? Which not good, but it's like something that worries me and I ⁓ feel of. Joshua Burkhow: That's Yeah, yeah. this is...

⁓ Tiankai Feng: acting that way the whole time and like suspecting people. But I can't help it, right? This is, think, the world we're living in, right? We want authenticity.

Yeah. Joshua Burkhow: ⁓ it's so true. It well, don't I feel like you might've been reading my emails buddy. ⁓ this is stuff that I literally talked about this week on social media, do the blogging, obviously podcasts, other and One of the things that's become increasingly important for me, at least from a personal point of view, is I don't want the sort of corporate jargon.

And I call it marketing psychobabble. want the authentic human. So when I write in a blog post or I write an ⁓ that's gonna go out to thousands of people, I want it to have the language that I would actually say in person. ⁓ say things like, dude, cool.

⁓ Tiankai Feng: Absolutely, Joshua Burkhow: what's happening, I have all these sort of language constructs that have a to them, have a way I communicate. ⁓ exactly why, what said is, I someday come to Germany, we go and have a coffee, I don't want you to be like, wait, that's nothing like I thought you were. ⁓ And that authenticity ⁓ is huge. And I there's...

Tiankai Feng: Right. Yeah. Exactly, exactly. Joshua Burkhow: There's an underlying sort of message here in the, the we part of it is that maybe we can do better at making sure that AI doesn't change our voice, ⁓ doesn't change the way I would say things.

I'm purposefully trying to ⁓ laid back. I don't want to be ⁓ you know, tightwad or, or, ⁓ overly, corporate which is sort of a fluffy term again, but like, you know, I want to have fun. I'm a silly goofy dude ⁓ I like to laugh a lot and have jokes every email I have as a as a you know, a smiley face in it or something You know, mean, I don't want to sort of let AI Take that over right and and so really really interesting to to think about in How we go forward we make decisions use it.

⁓ like to shift gears now to Tiankai Feng: Yeah. Mm-hmm. Joshua Burkhow: to what you sort of, we're leaning towards is this idea of they now and about this idea that who's being affected even they're not part of the decision. Because the reality ⁓ that ⁓ talking all these things as if they're very close to home and it's me typing in, getting the prompt out and then I'm just gonna send this to an email to Tiankai Feng: Mm-hmm.

Joshua Burkhow: to the team and gonna be the self-contained. But we all know this, like social media. Look at social media. We know, I don't know, half or more of it is AI.

And I would say a good chunk of that is the AI slop that you refer to. Just people putting out stuff that literally makes no sense in any form and sort of human value to... ⁓ to this idea, it's just permeating problematic concepts and ideas and stuff that, ⁓ so I'd love to sort of get your take on this. Like how do you think about AI once it gets beyond the grasp of the AI, we, and is now sort of out in the masses and ⁓ they?

Tiankai Feng: Absolutely. think, I mean, I would break it down into what goes into AI and what comes out of AI roughly about how the day is impacted. And it all starts with what goes into the AI in the first place. And I mean, that is very publicly addressed, I think, everywhere, but it's still, I feel like, under discussed, which is, for example, intellectual property overall.

We know that all the big models, they all have scraped the internet and are everywhere and have gained a lot of content and we're trained on it. ⁓ But what does it mean for people that are original creators, writers and storytellers and musicians and all those people that are now all of their work is being used to train it. And then now people can very easily imitate it by just asking a model to do it. And every time we ask for, for example, write me like a novella in the form of Stephen King, in the style of Stephen King.

then we're at the same time probably hurting Stephen King a little bit, right? That is how it feels because who is, why would we then just use the technology that without his consensus potentially actually ⁓ imitate him? And is that really correct? The same is with bias, I think, which is a really tricky one, right?

Knowing that AI has inherently a bias based on the sample and that it can be discriminating towards certain groups of people and minorities, et cetera, Joshua Burkhow: very much. Tiankai Feng: that the irony of it all is that for AI to work and for, need to train it with a lot of data. And that means often we go back in history a lot, right? But the further we go back in history, the worse human behavior gets because we, the further we go back, is the worse we behaved generally in what we did, right?

Like just slavery, torture, whatever you want to call it. All of that happened in the past. And we now are saying, and we all agree that this has been wrong, but the further we go back, Joshua Burkhow: That's right. Yeah.

Yeah. Yep. Tiankai Feng: This is what we're actually giving AI to, right? So it's not only amplifying what we now see as good, it amplifies what we did as bad too, unless we curate it in a certain way.

So finally, that paradox of saying it needs to amplify only the good things and not the bad things needs us to make a trade-off, right? That, okay, let's not go back in history that much, and only recently, but that might be a little more volatile. We need to compensate it with more original samples that are now AI-generated, so we actually shape it towards in the future, working the best way. Joshua Burkhow: Yeah, yeah, yeah.

almost like building the humanity back into the AI systems because you and I didn't get here with our thoughts and our beliefs and our understandings of the world without all of that stuff happening before. You know, I got ⁓ raised the sort of patriotic idea, ideal of ⁓ US with Lincoln and Thomas Jefferson and these sort of prominent figures that weren't innocent. You know, they had their flaws and all, but they Tiankai Feng: Exactly. Joshua Burkhow: they were going through a time and place where they had to try to move humanity towards a new place.

Wouldn't it be great if our AI sort of took this into consideration? And I don't actually know the answer to that. I don't know that's sort of how we're building models at that level to where, know, you and I go into a prompt ⁓ to about, you know, thoughts on war, thoughts on ⁓ these around the world. Tiankai Feng: Mm-hmm.

Joshua Burkhow: What it's going to give is influenced both by life ⁓ and financial statements, you know, ⁓ business ideas. So, ⁓ super, super interesting to. Tiankai Feng: Exactly. Joshua Burkhow: to think about that, right?

Tiankai Feng: Absolutely. Yeah, I mean, and on that note, I think ⁓ leads me to the what comes out of AI part, right? That is also ⁓ an impact that we might not predict in that extent, ⁓ is disinformation, right? And I think I mentioned that before with the intentional and unintentional part where ⁓ can create intentional deep fakes and make things look really real, but be actually not real.

Joshua Burkhow: Mm-hmm. ⁓ Yeah. Right. Yeah.

Tiankai Feng: And we've seen that, I think, in the media when it came to war stories, that suddenly people generated videos if war happened in certain cities when it didn't actually happen. And it can have a really critical impact that makes people panic and makes people make really impulsive decisions. This is really bad, right? when we think about that at work too, That people spread misinformation and make it look like a real email that came from the CEO when it actually wasn't, for example.

And then our video of the CEO recording it, but it wasn't him. Joshua Burkhow: ⁓ yeah? ⁓ yeah. Yeah.

Yeah. Tiankai Feng: actually doing it, then it can cause a lot of panic too at scale. But also the unintentional part, right? As I mentioned with hallucinations and not fact-checking and just taking wrong things.

I think this is like a big problem that we have in the world where ⁓ we as human beings still are word ⁓ mouth people, right? So that causes virality in things. And no matter if it's within an organization or on social media or in a much bigger scale in the whole country or region. ⁓ If those unreal and fake things make it and it leads to ⁓ that are not rooted in reality, then that can be a real problem, And I think that is something we need to consider too.

Joshua Burkhow: I think what you're sort of hinting at is that critical thinking as a construct, human construct, is more precious than ever. Our ability to critically look at something Tiankai Feng: Yes. Joshua Burkhow: and not accepted the idea at the face value. mean, I don't even want to get into this because it's a whole podcast, but world politics, world politics is being shaped the viewpoints that people have are being accelerated and exaggerated ⁓ media and influence ⁓ of these of AI perspectives, if you will.

and ⁓ Tiankai Feng: Exactly. Yeah. Joshua Burkhow: ⁓ an idea gets put into someone's head, they read on Instagram or social or their friends sent them something, and taking it face value. and saying, ⁓ that has to be true because my friend sent it to me.

That has to be true because it's on this website that I trust. So we have to be very sort of like adamant about like, hey, this this isn't right or this something doesn't look right or let me validate it and make sure that it's a correct statement. But we're finding this conundrum in AI just in our lives in general is that we're not being as critical about everything that crosses our eyes. Tiankai Feng: Yeah, that is true.

it's a really good point, but one of the reasons also is that ⁓ in the past, it much easier to differentiate between true and fake, right? I think it was just very obvious and that made us very confident doing that. But because AI generation is getting so close to looking real, it's getting harder and harder, right? I mean...

If we just look around on social media nowadays, all of the comments are like, is this real or is this AI? Like people are just, everyone is doubting themselves, right? What is happening? It's everywhere.

It's like, is this really real or not? Like what is reality even? Joshua Burkhow: Right. More comments on that.

⁓ my gosh. More comments on that than anything else. Yeah. Well, ⁓ a good thing actually, right?

Like people are questioning, they're being They're, they're looking at it and being like, this is pretty sensationalist. This is pretty, you know, pretty strong statement. Let's validate. Like, is this real or did someone just make this up to get views, right?

Tiankai Feng: ⁓ one last point, maybe on the way only I wanted to add ⁓ ⁓ even bigger scale. And I want to group that under ecosystem sustainability, right? Because ⁓ AI to work requires lot of energy, right? So we have all these ⁓ data and computational centers that are only increasing and GPUs and whatnot, all the hardware that we need basically to run it.

And because it's now so mainstream that everyone is using it, we're burning through it a lot, right? Joshua Burkhow: Yeah. Tiankai Feng: But that has real consequences. feel like a lot of areas where people were living in peace are now being occupied by a huge data center that's creating a lot of changes for them for the worse.

And also generally when we think about global warming and whatnot, all these things, sustainability wise, are actually changing too, And I think that is like the much bigger scale of they. When we think about only not the societal dynamics that is impacted, the actual ⁓ environment impacted too. that at some point we need to also make hard choices, Do we actually run ⁓ we run now and scale in the way we scale now to ⁓ so much impact on the planet regarding energy consumption or are we gonna find a nicer balance in some?

Joshua Burkhow: I to be one of folks that... ⁓ I don't like a lot of rules that are unnecessary, but I've changed in my thinking a lot in the idea that the conversation is what's important. Tiankai Feng: Mm-hmm. Joshua Burkhow: let's get leaders together, but also AI experts.

Let's get people together to talk about these things and hash them out and figure out like, Hey, is this true? Provide your case, make a case for why it is a problem, why it isn't a problem. Let's come together. We're all.

And I just like this concept and at least in my head, because it's again, another check mark on the wall for being human. You know? Tiankai Feng: That is very true. I fully agree with you.

I mean, I would say, of course, that working in data and AI governance now, that governance is still a function ⁓ people see as trying to avoid because they see it as something that's slowing them down. So let's not tell them and do it just behind their backs. And then we can still ask for forgiveness afterwards. But let's do it first.

⁓ especially AI, it's very hard to stop things, too. Right? Like, I mean, once you actually do it and you run it and people get used to it, then stopping it again is very tricky. So we need to find that balance.

⁓ And governance itself needs to be have a new way of operating through it. It cannot be that manual auditing and a blocker kind of mechanism anymore. It needs to be more intuitive and more guard rail-y, I would say. But I think they are ways to do it.

Joshua Burkhow: Yeah, we're sort jumping into this area that I talk a lot more than I thought I would about is the governance landscape. And what I mean that is governance three years ago from a enterprise way, like, okay, how do we not put the company at risk. How do we make sure that our data is clean? Like sort of a structural governance.

Now, if you look at AI governance, I'm empathetic with the that are now in between users who want to use AI and ⁓ the end goal. And give you an example, like our legal our IT governance team. ⁓ These are folks that are Tiankai Feng: Mm-hmm. Yeah.

Joshua Burkhow: in their roles for a very good reason to protect, whether it's interests of the employee, interests of the ⁓ company, but they're right the middle of it and they're getting pressure from people, leaders, companies Tiankai Feng: Yeah. Joshua Burkhow: to tell us what this will actually hurt us or how it could actually hurt us and then we'll work from a shared understanding. Do you it as new thing or is it just sort of an evolution of where we're coming and not really a new problem?

Tiankai Feng: Yeah. ⁓ that's a great question. I do think it's an evolution, but more of an evolution of that ⁓ has changed and evolved so much that it doesn't even feel like the old governance anymore, right? ⁓ Because how was before, it was really indeed like a very authoritative and control function where it's now needs to be in the new world to be more of a...

guard railing and enabling function. And ⁓ as a mindset shift has a lot of implications on how it's being Meaning, for example, that it's not ⁓ there jump in to stop things anymore, but it needs to be early at the table to then set the right conditions for things to move forward, but in the right way and not come afterwards to say, stop, don't do ⁓ it. also with not define too many rules upfront. but rather define rules as we go to basically be dynamic and not having to consider all of the bad scenarios, but more actually focusing on the real ones that are more likely to happen, for example.

And in that way, it does feel like a new way of governance and a new way of doing it, simply because it wasn't done that way, right? It was easy to sit back and write policies of like 600 pages and like, now everyone just needs to follow the policies and we're done. Joshua Burkhow: Sure. Absolutely.

Tiankai Feng: But now it's like being part of it, sitting at the table, being part of the action, seeing how things progress and making decisions as we go that makes it much more integrated and embedded. And I think that is new, but for the better. Joshua Burkhow: Yeah. All right.

So I'm going to put you into a role. I'm going to make you now the CEO of this invented company. ⁓ And I want you to imagine for a second that you being pressured by the board ⁓ directors to enable across your organization. ⁓ How you address or do you think of it from Tiankai Feng: Okay.

Mm-hmm. Joshua Burkhow: this sort of leader perspective to ⁓ on how to use AI at work responsibly. do you ⁓ mandate a bunch of rules like we talked about, which I think would probably agree is the old governance way? ⁓ Do you provide at least guidance?

Do you lock down the tools? if you're coming from the viewpoint of being the leader of a company potentially thousands of employees that you know are all going out using AI in a hundred different ways. Tiankai Feng: Yes, I think that I would my shareholders and the board of directors about how we need a dual track approach, I would call it. ⁓ what I mean by that is that there's one way we say, OK, let's identify the most impactful AI cases ⁓ and let's on this because we cannot do everything at once.

Let's pick, make business cases and then focus on the ones that actually are going to move the needle. ⁓ let's say a core business process and we're gonna put AI agents on it and see how much it will make us more effective but also efficient and we're gonna gain some stuff. But that requires a lot of governance and rules and everything that needs to be done and a lot of change management to do it and so on. But it cannot be the bottleneck because everyone wants to do his other stuff with AI, So the other track is the sandbox track, Which means let's give our people an environment of a sandbox where they can try out everything they want.

but it's isolated from the rest and it's not going to impact our business as a whole. But we give them the opportunity to just play around and do what they need to do to try out things. And if within that sandbox, at some point, a great idea comes up that helps us with our priorities, let's take it out and put it into the process of the other track. But other than that, it's just a playground for our employees.

this way, I think we balance it. We have the whole sandbox and not stopping anyone to do things part here. But the other side is a more rigid as govern process where we actually put things from pilot into production and actually track value and govern it, et cetera, et cetera. And that should make the shareholders happy because it becomes ideally a flywheel of innovation to impact and then back to innovation again.

Joshua Burkhow: Right. It's pretty good. think you could, you got this game figured out in a little bit. I like the sandbox idea and ⁓ I try to mirror that with the reality that at least ⁓ I live in is a lot people are testing things out in production ⁓ and production meaning Tiankai Feng: Yes.

Joshua Burkhow: real life, We absolutely don't know what it's gonna come back with. And so we're in a development phase. We're looking at, what's gonna come back? Okay, that looks good.

I gotta change this, gotta change that. But we're not doing that in a sandbox like you proposed. We're doing that day-to-day business. Tiankai Feng: Exactly.

Joshua Burkhow: And I don't mean that as a slight anybody. It's just, that's a data systems concept that doesn't necessarily apply to, ⁓ the broad, broad scale of, of, you know, hundreds of millions, billions of AI users. yeah. Tiankai Feng: It's such a...

I mean, I think at least within organizations, we can make the effort as part of our general AI literacy efforts, right, to just guide our employees, right, to do certain things here and to do certain things here. But it's about helping you to do it in the right places. It's not about blocking you. It's just do it where it fits, right?

And that is kind of where It's the same with like giving them tools and to say use these tools for this but not for the other thing. If you want to do the other thing, use the other tool, but don't try to use the weakness of both tools in that way, right? For example, that's kind of what I was about. Joshua Burkhow: That's right.

right, man. We've been responsible adults here, as I often say. We're going to come in to the point, my favorite part of the podcast, I've really enjoyed this, it's called the lightning round. ⁓ I hope you're ready, man, because this is, I'm going to ⁓ fire, give you some questions.

Your job is ⁓ one only. You have to reply, but you can't. Tiankai Feng: Ha ha ha. Awesome.

Yes. Joshua Burkhow: It can't be a maybe it can't be a it depends. has to be sort of your, your first off cuff answer. couple, I five quick questions for you.

Are you ready? All right, first one. One capability that becomes ⁓ non-negotiable 2026. Tiankai Feng: Yes, I'm ready.

Learning and adapting. Joshua Burkhow: That's a good one. ⁓ AI capability ⁓ we are overestimating. think it's way better than maybe it actually is, or I'll even add to it, maybe it's changing.

Tiankai Feng: Yeah, critical thinking, I would say. Joshua Burkhow: Critical thinking, good, yep, we covered that, that's a good one. a governance myth? know, a myth that you where either people think is ⁓ part governance or is governance or is being implemented that really probably doesn't apply.

Tiankai Feng: Yeah, I would say the biggest myth I like to break is that it's the police, It's either data governance or data police, air governance or air police. I don't want it to be the police. I would rather see it as the judges, if that is, right? If you think about legislative versus executive and little bit more consultants rather than the police.

Joshua Burkhow: Yeah. Yeah. Yep, that's a good one. All right, one decision that must stay human.

Tiankai Feng: think the decisions over our own lives need to be still decisions. Whatever the steps we take and how we move on in our lives need to be human. Joshua Burkhow: And then I just came up with another one that I'm going to try to stump you on. But one thing leaders consistently about Tiankai Feng: Okay.

I think the... The effort needed to actually tailor it to our needs is consistently underestimated. It sounds like most of them just say, let's just plug in ChatDBT and then we can do everything. But it needs so much more than that and so much more integration and tailoring to our organization that it's often underestimated how much that costs.

And then suddenly, oh, that is a lot of work. But that is a reality. Joshua Burkhow: be there. Yeah.

Perfect. All right. Last one and then we'll wrap it up, my friend. What's the last thing you did using AI, whether prompting or work or whatever?

Tiankai Feng: Yeah, I mean, to be honest, this morning I looked up what a good swim training for my son would look like when we go swimming next ⁓ And what I like some of the ideas, I could playfully teach him how to swim. That was the last thing I was a girl. ⁓ Joshua Burkhow: There you go. There you go.

Good. Perfect, good. Tianka, thank you. I really enjoyed this conversation.

Time goes by so fast when you into these good, meaty conversations. Tiankai Feng: Same here. Joshua Burkhow: Before we go, can tell listeners where to ⁓ find you? If they wanted to reach to you, if they want to follow you, talk to you for more.

Tiankai Feng: Absolutely. Yeah, people are all welcome to connect with me on LinkedIn. That's where I'm most active. I also have a YouTube channel for making some music about data and AI as well.

I make parodies and original songs about it. And I've written two books that we briefly mentioned called Humanizing Data Strategy and Humanizing AI Strategy. A lot of the topics we talked about today are in there too, but they can be bought on Amazon and everywhere where they get books, basically they can look it up. So yeah, I would be looking forward for people to read it.

Joshua Burkhow: Mm-hmm. Yeah. awesome. ⁓ so cool.

⁓ we'll make sure to put this information in the show notes for people to find, but again my friend, thank you so much for joining and ⁓ hope we get to talk soon. Tiankai Feng: Absolutely. Thanks for having me. Joshua Burkhow: Thank you so much.