This is a very rough transcript of a discussion between that I did for personal use, sharing it here in case anyone finds it useful
You can watch the amazing episode here :
Who do you think is going to make money when AI comes along? Is it the AI providers, AI services..?
There are going to be a few obvious service providers.
So that’s the general answer and the deeper answer is there is an underlying question, which I think is an even bigger question is about AI that reflects correctly on this which is ” Is AI a feature or an architecture?”
Is AI a feature? – we see this with pitches we get now – which is like – we get the pitch and it is like – here are the 5 things my product does and oh yeah, the number 6 is AI. It is always number 6 right because it was added after they created the rest of the deck. If AI is a feature then it is actually correct. Which is like every basically, everything is going to have AI sprinkled on it, there would be AI features in kind of every product. That is possible.
We are more believers in the other scenario that AI is more of a platform and is an architecture. In the same sense, as that mainframe was architecture, the mini-computer architecture, the PC, the internet, the cloud have been architectures. We think AI is the next one of those.
If that is the case, when there is an architecture shift, in our business it means, everything else above the architecture gets rebuilt from scratch because of the fundamental assumption of what you are building change. You are no longer building a website, you are no longer building a mobile app, you are no longer building any of those things, you are instead building an AI engine, in the ideal case is giving you the answer to whatever the question is. If that is the case, then all applications will change. Basically the whole industry will turn over again. The same way it did with the internet, mobile, and cloud.
If that is the case, it is going to be an absolutely explosive period of growth for the entire industry.
All the products we see today will not be relevant anymore.
I will give you an example – there are lots and lots of business applications, where basically you type data into a form and then they store the data and later you run reports against it, charts etc, that has been the model of business applications for 50 years and in different versions, what if that is just not needed anymore?
What if in the future you give access to your AI to all your emails, all phone calls, all everything, all business records, all financials in the company and it let AI give you the answer to whatever the question was and you do not go through any of the other steps. Google is a very good example of this they are pushing hard on this, the consumer version of this is search.
Search has been the 10 blue links, you know for 25 years now, google they talk about this publicly, that what they are pushing for is just give the answer. Which is what they are trying to do with the Voice UI. That concept might get generalize out – and then everything gets rebuilt.
So one of the new interfaces to AI is voice. What do we likely to get wrong about voice.
If we are going to get something wrong about voice it is going to the thinking that it would be a wonderful replacement for existing user interaction models. So it would be like a replacement for a keyboard, mouse or touch. Probably not.
On the other hand, there has been this turning point reached, it feels like in AI applied to language and then to voice and text to speech. For 50 years people have been working on these methods.. And what we are seeing is that now has started to work, in the same way as machine vision has started to work about 7 years ago. And so, if that is the case, then all of a sudden the conversational UIs are going to get much better.
Then you couple it with what you are trying to achieve when you “talk” to a computer are you trying to write a document or are you trying to read an email or are you trying to do everything you do today or are you fundamentally be doing something different because the machines are going to be so much smarter. I think that is a very interesting open question.
When I think about the AR mirror world, I find it hard to imagine it without it having a voice component where we can understand where you are saying besides what you are looking at is that an essential part of the AR world?
I go so far as to say that it may be the case of voice is the key to the AR world, voice may be the thing, the foundation of the whole thing. Apple airpods were kind of a fundamental breakthrough and it is getting better… the visual overlay of AR will obviously be important and valuable. The visual overlay might be supportive of on top of the voice experience.
We could very quickly have universal language translation speaking in our ears. I think people underestimate the change that it would bring about in the world, you would have millions of people who are highly skilled in everything except the skill of English now being able to participate in the global economy.
We were talking about unexpected things, biology which is a million times as complicated as digital, we are now talking about biotech revolution are we misunderstanding what biotechnology actually is?
This is the big bet we made with our bio effort that we started a few years back. We think biological sciences is at a turning point at the scientific level. It is at a turning point from basically being a process of discovery of how biology works to be able to engineer biology. Up to and including being able to program biology. Being able to use engineering field skills to biology. If we are right about that then the whole concept of how bio or biotech work is the verge of really changing.
The obvious application would be in pharmaceuticals. You know there is this concept of drug discovery, right, it is always the word discovery, discovery sounds great, it is optimistic, it is like discovering things is fantastic. The problem is discovering is like, they have to run all these tests and reverse engineer back from nature and the problem is that sometimes they discover, sometimes they don’t. So the example we always give, is we talked about, with computers we have been in this kind of 50-year track of what we call as Moore’s law where chips were getting faster and cheaper every year for a long time.
In drug discovery, they have what the call Eroom’s law which is Moore’s spelled backward and that is the cost of discovering a new drug and it is exactly in the wrong direction. Billion of dollars now. If you could actually engineer biology, then all of a sudden you can apply the decades of skills we have built up on how to engineer things and we would be able to engineer new pharmaceuticals from scratch.
And all this runs on basically Moore’s law which has been foundational to “this here” it is hard to imagine anything we have in the modern world today without Moore’s law. Do you think Moor’s law has another 30 years run, is it limited, is it finite, will go on forever? Morre’s law is defined in a broader sense of computer getting cheaper by half every couple of years. So what is your take on Moore’s law?
The traditional definition is the computer in the form of the chip, specifically a chip. The Moore’s law has always been expressed as unit one of chip and that could be right, there could be a CPU, or that could be a graphics card, could be a graphic chip or a memory chip. Specifically what you were doing was that you were able to put more transistors on that chip, right for the same cost and actually for a long time, you were able to reduce the power requirement per transistor, which is this added benefit. So chips got faster, cheaper and more power-efficient.
And that sort of to quote what you said most of what you see today in the computer industry. The bad news is that, that in that form is coming to some form of an end, we kind of hit the limits of fundamental physics we are now sort of at the deep atomic level. It is becoming much harder to make progress at the chip level.
The good news is that the industry, starting 10 or 15 years ago, the computer industry broadly, refocused off of what you do with a chip to what you do with a large number of chips, the old model was you make a chimp more powerful because you are trying to scale up. The new model is you use 1000s of chips in parallel and you have this kind of approach to scaling out and the full form of that is now what is known as the cloud.
So we have 15 years of steam going as we are getting good at using a lot of chips to do things and that’s why you see the continued ability to accelerate many many things that you deal with are getting much faster and the experience we get is getting faster.
- The rise in the scale of architecture is a really big deal, a modern developer does not care about what the power of any particular chip is you just light up some more of them and they do not cost that much.
- The other thing is now the chips are now specializing, particularly you have the raise of these dedicated chips for things like neural networks, where there is another level of opportunity to optimize. And then the other kicker is the programmers, software people like me get to step up, in the old days, when computers were expensive, programmers were good at optimizing every single step of a software program. Programmers got out of that habit, probably starting 30 years ago, where it did not matter as much anymore because Moore’s law was working so well so software today is massively inefficient. There is something called worth’s law or something which is somebody did benchmarks of taking Microsoft office from 2000 on a PC from 2000 and 2007 on a pc from 2007 every function you could do you could do now in twice the time.
It has become primetime again for software developers to be really good at optimization. Which is what is happening in the AI and cryptocurrency world. So we seem to have decades of advancements ahead that is not purely dependent on Moore’s law.
If we think long term, thinking of a 100-year span, to have prosperity like we have seen will kind of require that power gets cheaper every year, is your confidence that we could kind of keep this going is based on human ingenuity or do you think there is some basic principle of science that suggests that we are just at the
beginning of what we can discover?
Gorden Moore who invented the law said it was not a prophecy, it was a goal. It was a goal of what you could do if you focused intensely on a set of optimizations over a long period of time. He used to say there was nothing inevitable about it.
I think that the answer to that is we have many many areas of improvement – we do not have the one we had which is the transistor doubling kind of effect – we have a lot of mind power going into this – the transistor was not obvious – then they invented that – the integrated microchip was not so obvious but they invented that – so you do not quite know – there are lots of technical proposals right on how to get to the next level in Moore’s law.
There are all these sorts of theories on optical computing and in the long run, Quantum computing, biological computing, so over the course of the next 20 years, let’s put it this way, this is one of the world’s largest prizes. If you are the engineer that figures out how to accelerate Moore’s law or how to shift computing into a new substrate like biology, that is the thing to do and so that is the prize and that historically has been pretty motivating.
We have 4G, we are talking about 5G, so far 5G seems like faster 4G with a lot of hype added to it, there is a technical specification for 5G which is really awesome, you know 100 gigabytes, 2 ms latency, almost impossible, are you counting on that for the next decade?
5G has become a national Geo-Political battle. Actually it is a very interesting twist. It has become a primary like you know if the cold war between the US and USSR was defined by the space race, at least the nascent cold war with China is actually a lot of it is around 5G interestingly enough, it could have been around a lot of things but it happens that is happening around 5G and so you have nation-state that very very badly need to win, two big nation-states in particular, so I think there is going to be a lot of, let’s talk about the payoff from the space race, is like all the products that go spun off from that, satellites and GPS and everything.
The other thing about 5G, people sometimes say it will lead to applications that we have never thought of yet, I think it is sort of true, but I look at it a little bit differently, which is a little bit kind of the Moore’s law conversation we were having which is a little bit math kind of question which is, there is 3 classic rules on how networks scale, and turn into value or usefulness and there is, historically it is called Sarnoff’s law.
( Sarnoff’s Law is simple – it states that the value of a broadcast network increases in direct proportion with the number of users. … An understanding of why and how broadcast networks are valued helps us understand why and how digital networks are valued)
Which was based on broadcast tv which is that the value of the network is equivalent to the number of nodes right so it scales with N, tv network with 10 million views is twice as valuable as tv network with 5 million viewers, that is kind of the obvious one?
Then there is Metcalfe’s law, which is the value of a network is equal to the number of connections between two points, that is how email works, emails from person to person. And that correlates to N2. ( n squared) the value of the network raises exponentially.
Then there is something called Reed’s law, which is called the group forming law, the value of a network is proportional to the number of groups and subgroups that can form inside the network which turns out to be 2 to the nth. If you want to have fun chart N, N Squared and 2 to the nth, 2 to the nth goes like straight vertical.
2 to the nth is what is happening with social networks – WhatsApp groups, games, and these other things.
So there are sort of 3 ways in which network growth pays off, all 3 are working now based on broadband, mobile..
The introduction of 5G is the way I think about it is it will turbocharge those 3 networks, in particular, that last one, you know those last two, so it is going to add a lot more N, a lot more devices on the network, there is going to be a lot more thing those devices can do, there is going to be a lot more point to point connections that make sense to have, there is going to be a lot more groups that are going to form, lots more economic activity that happens.
Something that we are expecting to happen but did not happen was in the world of what is sometimes called the sharing economy. After AirBnB and Uber there was a stampede of companies that were going to be uber for x and then X was everything in the world, very few of them have succeeded, again there was an expectation that we will see more of them but we haven’t, so is that whole idea is kind of at a dead-end or is it we are in a slow disruption, this is going to take a while, like the generational requirements we were talking about technology earlier, or is there something else, what do you think will happen there and what are we looking at?
Once again I object to that question – so I look at a bit differently.
Great ideas look like very specific theories not generic theories like uber of x. They tend to be very specific to the details of the market involved.
One of the things we have learned about ride-sharing why ride-sharing works so well, as long as the driver is good, it doesn’t matter who the driver is. One of the classic examples was uber for cleaning your apartment – you do not want a different person coming every week to your apartment to clear it. So a lot of simple applications of that idea won’t necessarily work.
One of my companies is Honor – loosely you might think it is uber for senior care, in-home care for seniors, actually, it turns out it is really a loose model for a couple of reasons. Honor actually has full-time employment relationships with the workers, which of course is very different from the uber and lyft model
It actually turns out the matching problem is much more complicated. When you matching someone to take care of him at his home, there are like 20 variables you need to match on so that everybody is comfortable with the experience. To give you an example in some cases you need people who have the physical strength to be able to lift the people when you are caring for them. You do have to be able to do these kinds of multidimensional mapping. That model is really working.
I think there is a big kind of vista of exploration that can happen from here, and there could be 100s of new models that people will figure out.
Speaking of new models, do you think about new models for the VC industry itself and how you would apply the principles of innovation and disruption to what you do in general? So as you look out 30 years what kind of innovation would you expect in the basic business that you are in?
The term “carry” the profit share concept comes from the percentage of the whale that was carried by a ship that was financed by the angel syndicate.
There is a timelessness part of the way these kinds of expeditions are financed, that is likely to endure.
The big question for me is how the shape of the companies and let’s say the ventures themselves change. And so, today there is a well known, understood template for prototypical silicon valley venture investment. A company in a certain place, it is a C corporation, it is financed a certain way, 30 years from now are we going to finance companies here or you know anywhere, 2 places, 50 places, 500 places are the companies still going to have physical place, are they going to be virtual, are they all going to be companies are they all going to be blockchains, or are they going to have actually employment relationships? We do not even know what is the shape of companies looks like in 30 years.
Long term thinking is often ignored – how can long term thinking be applied to silicon valley?
Long term thinking is core and central. It feels it’s all going so fast, any of the important product takes a decade to build and a lot of things actually feels as slow.
It is really easy if you know if the thing is going to work. The problem is long term thinking cross with uncertainty and fatality.
The secret weapon is venture capital – we have the concept of the portfolio wired into the model. So if you have to adapt your mentality from will this thing work to will this portfolio of things work at that point you can make risk a somewhat tractable thing to contemplate. It is still hard to divorce yourself emotionally from it.
The best-run companies tend to run multiple experiments against their goals, in a lot of cases they run experiments in parallel as they are trying to try different things. So the mentality of portfolio risk also applies to how you run a company. You should have a great deal of conviction on where you want to get, but you need to have a great deal of flexibility on how you are going to get there. You can diversify the risk on any one theory by doing that.
Can you have prosperity with low growth or fixed growth – at a civilization scale?
I am a very strong proponent that growth is absolutely necessary. You can have fundamentally two different mindsets of how the world works, one is positive-sum, a rising tide lifts all boats, we can all do better together, and the other is zero-sum ie for me to win someone else has to lose and vice versa.
The reason why economic growth is so important is that if there is fast economic growth, then we have positive-sum politics. And we have all these discussions about all these things we can do as a society. If we have a flat growth or negative growth, all of a sudden the politics become sharply zero-sum.
You can see this if you track the political climate. In the wake of every economic recession, the politics just go seriously negative in term of thinking the world is zero-sum and when you get a zero-sum outlook in politics, you get anti-immigration, anti-trade, anti-tech
So economic growth is necessary for prosperity and good politics.
Globally resource use and carbon emissions are raising whereas in the US it is actually falling. So in the US, we have figured out a way to grow while reducing the use of natural resources, which is a completely unexpected twist. When economies reach a certain point they become really really good at doing more with less.
Really good at efficiency, energy efficiency, use of environmental resources, very good at recycling, and they really get good at what is called “dematerialization” which is what is happening with digital technology. Which is basically taking things that required atoms and turning them into bits.
Growth also pays for the social safety net. Growth is the biggest form of magic that we have to actually make progress and hold the whole thing together.
I think the idea of leapfrogging in technology is a myth, it actually does not work, if you need the high tech infrastructure, you need to have the roads, clean water, you cannot skip over that. They need to be built out in order for them to have prosperity at the end.
One thing I worry about is cyber conflict – cyberwar, partly because we have no consensus about what is allowable. Does this worry you at all?
I will make an optimistic argument which is going to sound a little strange.
If you project fwd with what is happening with generally with information, but also with drones, UAVs, unmanned fighter jets, etc if you project this fwd, you start to get this very interesting world in which all conflicts between countries so far has been throwing people at each other, you do have the prospect of basically a new world of both offense and defense, completely software and technology-driven. The optimistic view would be – is it not good that there are no human beings involved?
If the machines are shooting at each other, is it not good, it is not better than they are not shooting at us?
I do know if I am in favor of machines making the kill decisions, decisions on whom to shoot, one thing I know is human do that very very very badly. I am the opposite of pro-war, I do not want any of this stuff actually played out, but if that has to play out, maybe having it between software and machines could actually be a better outcome.
We want to take 18-year-olds out of their homes and we want to put a gun in their hands and send them someplace and tell them to decide whom to shoot, that is going to go down in history and have never been a good idea.
What is your optimism based on?
The science fiction authors often talk about the singularity.
My reading of history is that we are in the singularity already, if you look at any human welfare in history, 300 years ago, all the indicators for human welfare in societies that were making progress, rise of democracy, the scientific method, free speech and it has been going around the world in concentric circles.
We just have to apply those systems to have prosperity. Double down on economic growth, capitalism, scientific method, technology, and tech tools.
We know how to make the planet much better we just have to keep doing that.
All the indicators are up into the right when I see things beyond the news.
KK – so let’s go out and make the world better…