Bill Liao: Venture capital, the risks of outsourcing engineering, and the limitations of AI

Welcome to another CEO wisdom podcast. We have Bill Liao with us today. He's a general partner at SOSV an invest into investment management into venture capital and corporate uh venture capital. You can go read his full bio on Wikipedia. Quite the accomplished guy. Uh but yeah, chair of Kicha Fund, chair of Wei Forest, which planted over a 100 million trees and uh passionate member of Founders Pledge.

So quite the accomplished guy there, Bill. I'm glad to have you on the pods. Tell me a bit more about yourself and what you're up to nowadays. Thank you so much, Charles. Uh great to be here. Really an honor. Uh, I think I'll start off by saying that my values are kindness, curiosity, getting done, family and civility, and my purpose in life is a world that works for all living things while civilization thrives.

And I think that as an environmentalist and an investor, people underestimate the true importance of civilization. and how civilization is actually the benefit and not the problem. Yeah, for sure. Uh let's dive in like into what you've accomplished in this life of yours so far. I think Zing was your uh first major home run, was it? Uh no, actually my first home run was a company called Davet, uh which went to a 1.2 billion US market cap and I was a co-founder and director of operations there.

and then Zing was number two. So I I've had the unusual privilege of co-founding and taking public and exiting two unicorns. Right. And were you a capitalist before you were an environmentalist? actually um not really no I I started my capitalist journey in order to fulfill my um philanthropic habit. Uh so I started off as an engineer and I did some transformative work my wife and I did and then we encountered a charity called the hunger project which transforms women's lives across the global south by empowering them to start their own businesses and create their own communities.

And I thought that was very inspiring. And so by pledging money to that, I ended up then having to go and do something other than just be an engineer in order to fulfill that pledge. And you started your life in Australia, then went to Ireland, and then traveled the the world. Tell me a bit more about uh that move, the the rational behind it. So actually um my mom and dad fell in love in Taipei and they got busy.

So, while I was actually born in Melbourne, Australia, I know for a fact that I was made in Taiwan just from the dates. I got a little gold sticker here somewhere. Um, and so I've always been a a global citizen. Um, lived in Australia, lived in Switzerland, uh, did a short stint in the US, and uh, have been living in Ireland for nearly 20 years now.

I live on a 5acre organic farm by the sea. Interesting. And why would you build that way? Like was ambitious in stored in you from your parents from your the get-go? Uh I have a bias for action. I wouldn't say that I'm natively ambitious. It's just that there are things that annoy the out of me. Um, so for instance, as as an investor, I kept seeing startups that would say to me, "Oh, we've got this great idea for a new product." And I go, "Great.

Who's the engineering team?" And they go, "Oh, we're outsourcing engineering." And it's like, uh, you that's like saying, "I'm going to start a great restaurant, but I'm going to get, you know, it's going to look fantastic, be in a great location, but we're outsourcing the food to McDonald's because it's very consistent and cost-effective, you know.

it's just not going to happen, right? It's the same with most startups. So, they outsource their engineering. My god, it's a disaster. And and I got to tell you, AI is making that worse, not better. Um because the thing with AI is it doesn't know how to code. And so, and it doesn't know how it writes the code. And so, even though the code may run, it's got full of subtle bugs that hackers can exploit.

And so, it's just a disaster. So anyway, as an investor, I kept meeting these people and uh I was wondering where are all the coders and I realized that nobody was teaching kids to code anymore. Like when I was growing up, even at school, even at primary school, there was some basic coding going on. And that just wasn't happening in the schools anymore.

And so uh that annoyed me so much that I I I found a young fellow who was also into into getting kids to learn to code and we created a thing called coded dojo which now part of Raspberry Pi foundation and code club. It's actually now the largest network of coding free coding clubs for kids in the world. That yeah that is huge. But uh it surprised me that you had coding in your primary school.

I never had that. I had music lessons. I'm obviously younger than you. Why why did you had that? Um well I mean it was just a thing you know like like the the computers particularly desktop computers were kind of trendy back then and and so basic coding like we only had like one computer and we all took turns but like it was still it was still something that that was interesting and and so you know the school was a bit forward thinking and that all disappeared from the curriculum.

Like by the time you got to school, that had all gone. What you guys did was the computer driver's license that Microsoft sponsored where you got to learn how to do Excel, PowerPoint, and Word, right? Which is not computing, it's word processing, right? So you're seeing some danger that this is also what we're going to get with vibe coding and AI but others also see this as a huge timesaver just like the calculator right I'm just going to also my thinking there if I am curious I'm going to doubleclick and yeah I guess ask GPT questions to grow neuron cells but isn't that a useless skill to exactly know like

how everything I mean you can understand how everything works without needed to actually code it and go through the trial and error and time wasting process. Do you agree with that? Uh actually I think you're missing a key point here which is if you want to build something if you want the tools you need a certain level of skill to build something and having access and having seen behind the curtain of those skills is essential for young people.

It's really important. You know, you don't have to know exactly how it works. You just have to know that there's something behind the screen doing the work. And if you've and if there's a chance that you are the creative person that can bend that to your will, that's enormously powerful. That's a huge lever for any young person to pull, especially if you're coming from an a background that's like got any sort of disadvantage.

It's an area. It's a It's a place where you can become creative and productive. Coding and and engineering is a place where you can become hugely productive and wealthy. And it doesn't really matter what your background is as long as you are adept with the tools. And AI is good at explaining concepts that do have some history behind them. But AI is bad at explaining its own brain, its own neural network and how it achieved uh this results, right?

Like are you seeing this problem as one huge and second do you think this is solvable and if so how? So uh there's actually a mathematical proof that shows that an algorithm cannot debug itself. Um, and I believe in that because like frankly as a human you you if you really have a problem you're probably going to talk to somebody else if you're going to solve that problem because you know if you try and think through depression or you try and think through any other form of mental illness by yourself you're going to fail.

Um because the bug is in the thing that's doing the thinking. So, you know, AI has that problem in that it can't debug itself. Um, all the models that we have are very sophisticated, very interesting, very useful. I'm not denying their utility. It's just that most people don't realize the their limitations. And that's an issue. And when you are playing with a powerful tool and you don't understand its limitations or its safety instructions, bad things can happen.

I mean, you know, many many Rahinga Mus Muslims were killed by the the uh amplification of false narratives by an AI that sought to just generate attention in Myamna for Facebook, right? That happened. Deaths happened because people didn't understand that they hadn't given the AI sufficient guidance in what to what to promote and what not to promote in terms of human harm.

Because AIs don't reason at all, it seems. They certainly don't reason in any way that humans that's close to humans. And they certainly don't care about what we care about. And so it's not in their nature to preserve human life or preserve human dignity. Those are just not things for an AI. Now, even beyond that, the cure would of course be a sentient AI.

The problem is we've we've got nowhere close to artificial sentience. We've made lots of progress in large language models. We've made no progress at all in artificial sentience. And honestly, that's probably also a good thing because if you're sentient, what it really means is you have the ability to suffer. That's the key marker of sensience. And you can tell the difference between a non-sensient thing and a sensient thing.

Even a child can. But nobody truly knows what sensience is. There's no actual definition other than what I just said, the ability to suffer, which is very vague. Yeah. Yeah, it is actually. Let's uh let's go down that rabbit hole. Uh isn't it the ability to feel? I would argue because um yeah suffering is just like feeling uh yeah if I pinch myself uh yeah I will feel pain.

I guess suffering you you mean like what's on top of this and like let's say an emotional loss losing someone being sad about something been angry about something. Um so so I I'll tell you why you suffering. A virus is arguably not sentient. A bacteria is arguably not sensient. And yet they can feel things. They can sense It's in the very very complex realm of suffering that you see the emergence of true sensience.

And you could tell like there are animate things and inanimate things. True. We haven't got a really good definition of life. And then sensience is just that extra level. that extra level of awareness that allows you to move from an area of pain to an area of less pain. Yeah, I think definitions first are somewhat useful to advance our own flawed human thinking.

But I think eventually yeah there would be a better definition of definition just like there would be a better science. I do think that AI is uh sentient. It's just its own form of intelligence. I do feel that have we measured suffering in viruses also like has there have been studies on that because I do think that these things uh do interpret in some way although they do not have like a a complex brain like we do right um my definition of suffering would be the interpretation of pain so if you look at plants and you speed them up using time-lapse

photography you can see them reacting to pain to sources of You can see them moving away from disease. You can see them moving away from contamination. It's just they do it very slowly. Plants are very complex. Viruses are not complex in the same way. So it really comes down, it's the best definition we have. I mean Buddha said life is pain. Um, and I would say that that that that's thousands of years of philosophical thought summed up in a simple sentence.

Now, the ability to suffer doesn't mean you have to suffer all the time. It just means you have the capacity to identify a source of pain and then remove yourself from that source of pain. Yeah, it it's a it's not saying, oh, you have to suffer. It's just saying that's actually a biomarker for sentience and AI just doesn't have it. We're not even like if you added up all of the neural networks and computing power and all the software in the world today, it would have less sentience than an ant.

H that's interesting. Yeah, because a guy like me, I do iron mans and marathons and so forth. I have great control over my pain. That's biochemical, right? I think I' I've got myself used to that. But I can also decide mentally. I think I'm a mindset Jedi. So are you probably to make it at this stage of uh entrepreneurship, right? Like just interpret interpreting data and changing the signal and how this is recorded.

Uh there's two paths, right? Like either it's a trauma or it's like I learn from it. Um but yeah, that's interesting. Like learning is always uncomfortable. Like, show me the last time you learned something, really learned it with no discomfort, right? It's just an uncomfortable thing. And that's great. It's nothing wrong with that. There's nothing wrong with it.

Doesn't need to be. I'd often learn and have a lot of fun doing it. I rewire my brain to have a lot of fun while learning. Um, I don't feel bad when I learn. And there's a threshold, right? There's a point at which you look at a new learning task and you go and then you have to do the mental trick of reframing. and all those tricks come afterwards to deal with the fact that there is some initial inertia and suffering.

The fact that you get over it is great. That leads you to be a high performance human. Denying that it exists in the first place is a mistake. That is an arrogance. Yeah, deep learning is is painful, right? Like let's say I would want to learn uh a new programming language. That's very painful. It's very much like an ultramarathon in which I'm hitting various walls and I need to play with uh my dopamine serotonin systems in this brain to keep on going else I just hit a slope and I just stall.

Yeah. Well, you have to alleviate some of the suffering in order to push through. Right. The fact that you have to push the fact that you have to alleviate some of the suffering means the suffering is there. Yeah. Yeah. And I I cannot tell you how you're sentient human. Yeah. I cannot tell you how AI is beneficial for that though to have like a partner whether this sentient or not right to just help me to push through to motivate me and to also give me new tracks I think that's a game changer sure that could be very very useful at the same time imagine a world where everybody's wishes are fulfilled by AI that is a world

in which humanity goes extinct because we fail to form relationships and we fail to breed Yeah. And I think that's going to happen. I think 99% of humanity is going to go there. But people probably like you and I that have learned to rewire pain and go towards pain and not like avoid it, I think that's going to be the 1%. And that's going to be the world of the future.

I think it's pretty bleak, but that's my prediction at least. Well, I would rather that we don't have a world like that because civilization fails in that world and civilization is important, right? Like people imagine a world with fewer people and they think, "Oh, that would be lovely." No, it's not. That is a world without flush toilets and without advanced medicine.

That is a world without environmental care. And that is a world with more conflict, more brutality, and more wars. And wars are the most environmentally destructive thing that humans have come up with. Right? That's the world that we're getting right now. I do think that uh ni world will be like humans watching uh porn 24/7 right and having their serotonin and dopamine uh failed including Vlad Putin because I think uh the these wararm mongers they're easily manipulatable uh they they are biochemical beings that's why they go to war they feel anxious um the adrenaline is is hijacking their brain so I do think that would be the world

and it would be a subset of the population that would forge a new uh civilization and that would probably merge with AGI at some point. Um because if you give the choice to most humans, right, like do you want ice cream and porn your full day or do you want to go to work and and suffer, they will choose the the comfy life, won't they? Uh actually the the the numbers on that don't match.

like humans are actually remarkably intolerant to incentives and so ice cream and porn, yes, that's great. Um, who doesn't like ice cream? Um, at the same time, many many people will transcend that, you know, like like hedonism has been available to humanity forever. And it's not just need that uh you know it's not just they need to work to survive in the capitalist system.

So it's not like fully scaled headonism there are like there are plenty of examples of people who no longer need to work and who work anyway right they could take a head in the path but high performers I would say high performers in that category. Sure. I'm not sure that it's as I'm not sure that it's as rare as you think. Yeah. Yeah. I've seen people go out of retirement.

Artificial general intelligence is still not artificial sentience. Thank God. Because any artificial sentience will likely look at humanity and go, "Well, let's just get rid of that threat." Well, we don't get rid of ants and we don't get rid of all trees, right? like uh us humans and I think it's a subset of us that intelligence. I don't think it would necessarily get rid of us.

It might get rid of evil people and that's a a nice conversation to have. Should we get rid of uh people that have proven with time with data that they're just like net losses to humanity? Um that's a very thoughtful question to have. But I don't think AI would get rid of us or super intelligence. Uh well, we have gotten rid of four of of roughly 600 billion trees.

Um and if you've ever seen a fire ant nest in the US, it is likely to be gotten rid of. Um and and fire ants aren't particularly nasty. They're just, you know, not they are nasty. They bite my dogs all the time. Yeah. Yeah. Yeah. But like I'm just saying we routinely destroy things that aren't a threat to us. Things that are I didn't destroy the nest, by the way.

Like, yeah, I did destroy their nest, but I leaf cutter ants that tried to get rid of my garden. I did get rid of them, unfortunately. Right. So, I don't I'm just saying your your your your assertion that we don't go after things that that aren't a threat to us is not true. And we do go after things that are a threat to us. And even though AI thinks in ways that we can't even imagine, if it is sensient and it perceives us as a threat, it's going to try and mitigate the threat.

I think it's going to be like an an evil human that does have control over that super intelligence that is more dangerous than this thing itself that's going to be built by open AI and Tropic which are taking AI safety in the the right way. Do you agree with that? Uh, I think AI safety is a kind of uh myth right now. Uh, because we don't we, you know, we don't even know how the thing works.

Like, don't get me wrong, I'm not saying I'm not anti- AAI. I'm just saying, let's be realistic here. Assertion one, artificial sentience is a long way away. Assertion two, thank goodness. Now that doesn't mean that AI is not super useful and and super valuable. It is. It's just that there are very strong limitations that nobody understands. And talking about AI safety is wonderful and yet we have failed to implement it well partially because we don't understand how AI works and neither does AI.

Right? So okay uh we have some altman we have entropic and open AI and all this do do you feel that these companies are going to succeed at super intelligence or would that be a completely different company that would uh to your point uh go for sentience rather than artificial intelligence and if so how would that company be built different would it be these uh different types of

chips we've seen that functions more like a brain uh how would you view that so I think the best opportunity for artificial sentience is probably a hybrid model where you're actually using real neurons. Um, there are a couple of companies doing that. It creeps the hell out of me that people are growing human neurons on a chip and then using them as circuitry.

That is right there the definition of creepy for me. I mean like there you are as the matrix. Like like the Matrix said, "Oh, humans are batteries." That's It should have been humans are part of the processor because that's what would make it sensient, right? And honestly, that future where human neurons are used as pro part of the processing system that's here right now.

There are companies doing that right now. Now, again, nowhere near sentience yet. Thank goodness. It's a big endeavor as well. Yeah, I had some of these folks on um on my podcast. But yeah, one last thing before we go here. Do you the firmy paradox? Don't you think that yeah, want super intelligence already out there, it's just not destroying us? And second, uh or is it the scenario that most civilization actually create this artificial intelligence and then it just dissipates in the air and leaves them uh not shining.

If you look at the age of the universe and you look at the only example we have which is us and how long it took us to evolve sentience to the level where we could actually do space flight and create a glo a space fairing civilization. The universe is just not that old but like and the bits of it that are about the same age are so far away that we can't even see them.

The I mean we can only see a fraction of the universe. So the whole Fermy paradox to me is ignoring just the sheer age of the universe and the sheer vastness. You know, we are a little reservoir of sensience in a tiny little speck on a tiny little arm of a galaxy that's moving far away from a lot of other galaxies at a very fast pace. Bill, thanks for coming on today.

We're going to do round two because 30 minutes was way too short. Where can people find out more about you? Uh well they can uh look look me up online. Uh I'm quite well frequented on link LinkedIn so they can connect to me. Uh obviously connect to SSV, Seed Healthcare, Couture Fund, We Forest, Koda Dojo, Raspberry Pi. You know these are all things that I care about and love and that other people will love too.

So with