Prompt: What is a non-obvious trend that will see massive growth in the next 5 years?
For 21 years, the Churchill Club hosted an annual Top 10 Tech Trends Panel where guest panelists would answer (bring their own) and debate (others’ views) the above prompt. Steve Jurvetson was a regular panelist — but what struck me most was the accuracy of his “predictions” — his trends were correct, were mostly non-obvious, and have all played out within the given timeframe.
I’ve shared videos of and transcribed his trends — although I would highly recommend watching the entire videos for Steve’s responses to others’ trends and for the small-talk/banter defending his own.
*Note: I can’t find the recordings of Steve’s appearances in 2006 and 2007 — if you have them, please email them to me at firstname.lastname@example.org. Thanks!
- Vinod Khosla (Founder, Khosla Ventures)
- Josh Kopelman – (Managing Partner, First Round Capital)
- Roger McNamee – (Co-Founder, Elevation Partners)
- Joe Schoendorf – (Partner, Accel Partners)
Trend 1 (19:53 – 25:10)
Moderator: Number one is Steve’s and it’s basically Demographics are Destiny: Creating Opportunity. Every 11 seconds a baby boomer turns 60. This internet savvy cohort represents an enormous market of time and money driving new opportunities in mental exercise, online education, and eventually an ebay for information that exceeds the market for physical goods. So all I have to say is we’ll have to exercise some mental exercise to understand what this trend is. So will you care to explain what you mean here?
Steve Jurvetson: Sure. But I don’t promise to speak slowly. I can’t help myself. This is a pretty exciting one. It’s obviously a market trend and later we’ll get to some geeky stuff, but this is more a US, Canadian, UK, developed world market trend, so obviously I’m only speaking about part of the world in talking about baby boomers. But what’s exciting about this group is that they’re qualitatively different. This generation is qualitatively different from their predecessors. There’ll be the first internet savvy cohort of seniors and it is an enormous market. It’s hard to fathom how enormous this could be. To put some numbers behind this, they’re about three times as Internet active as the prior generation and twice as likely to have a college degree. So imagine a smart, active group not wanting to retire entering those ages when AARP knocks on your door with a different sentiment about what the future holds.
And they are basically driving the economy today, there are about 75 million of them, over half the workforce in America. And if you look forward just about 18 years to 2025, you can think of America, the entire country, looking like Florida does today. So there a lot of statistics you hear about, demographics, that is a certainty. There’s nothing that’s going to change that, it’s not really a forecast, it’s not something to vote on, because demographics being destiny is not something that’s likely to change barring bird flu or something dramatic like that.
So given that though, what’s interesting about it? What makes it special? Some, like Mary Furlong in the back, have estimated that this market for healthy aging is a half-trillion dollars in size. And this includes all kinds of interesting opportunities that you might not expect, like the fact that according to our estimates over half of all businesses and franchises today are started by people in this cohort, entering a sort of a second phase of their life. So they’re a very active group, and getting to this bold prediction that one day they may usher us into an eBay for information kind of era.
Let me share some of the thinking behind that. Because that’s just one of many opportunities that the Boomers provide. If they’re at home, educated, internet-savvy, what might they do? Some of the greatest pearls of wisdom that have been accumulated across a variety of service industries are in their minds. They have a connection now like they’ve never had before, and they’re not looking to retire. There are a lot of technologies coming along that allow you to partition workflows to farm out work, to outsource if you will, freely and more granularly than ever before. Think of everything from simple stuff like web services and legal services and advice on this that or the other thing. The services economy is the majority of the developed world economy and it will increasingly become so. And services aren’t easily traded today and they should be. So I think one day the services economy online will exceed that of physical goods if you think a marketplace is like eBay.
And I think boomers might be a beneficiary. Clearly others will as well, and this will be a globalization trend, not just a US trend, but it’s an opportunity for those boomers to really tap into the economy. Because there’s a wonderful asymmetry between those who have money and those who have time, and those who are seeking answers and those who don’t have the answers that they’re looking for. And you can imagine boomers working flex time from home on a variety of work that’s partitioned out to them. So this, I think, creates investment opportunities in education and retraining.
Investment opportunities in telecommunicating technologies and, you know, doing all this over the internet and video. And, sort of a kicker to make this maybe feel different than our parents’ generation, we might remember the last generation of folks going through this with a mental acuity that we haven’t seen before. And this gets to the mental exercise point that I think is very important and it probably relates to everyone in the room. It’s sort of a take-home thought: that if you’re 35 or older, your rate of cognitive decline right now is the same as a healthy 80 year old. It’s the same slope. It’s just when you’re older, you’ve accumulated more and you forget most what you try to remember. It’s sort of an important threshold, but your pace of decline is the same. But that’s just the average.
You can dramatically affect that outcome. If no one exercised, we’d see a very unhealthy population physically. If no one does mental exercise you see a very unhealthy population mentally. As our medical systems get better and better and extend our lives, we’re not doing anything to exercise the brain. So I think we’ll look back five, ten years from now and be amazed that the concept of mental exercise wasn’t on our radar screen just 10, 20 years ago. And as one data point of what it can do, there’s a company called Posit Science up in San Francisco that’s done a bunch of tests with UCSF and Mayo Clinic, and they’ve shown that just with 40 hours of trust and trained exercises, you can take 10 years off that pace of decline.
So now imagine these boomers in their 60s with the mental acuity of someone – we don’t know how far this can go, right? The research is just beginning – you know, someone with the mental acuity of a 50, 40, 30 year old now, but with all the accumulated wisdom of a 60 year old. And now with an Internet globally connected, economy to tap into. It could be a brave new world unlike we’ve ever seen. So in short, I think lifelong learning is more than just enlightenment. It’s an economic imperative.
Vinod Khosla – red
Josh Kopelman – green
Roger McNamee – green
Joe Schoendorf – green
Trend 2 (1:11:25 – 1:16:20)
Moderator: Alright, Steve’s second one. And this one’s giving me flashbacks when I had dyslexia: Evolution Trumps Design: Many interesting, unsolved problems in computer science, nanotechnology, and synthetic biology require the construction of complex systems. Evolutionary algorithms are a powerful alternative to traditional design, blossoming first in neural networks, now in microbial re-engineering, and eventually in artificial intelligence. Now what the hell does that mean?
Steve Jurvetson: Ok, so I’m a geek. I’m sorry. It’s a thing that excites me a lot. It’s a weird pattern, it seems to stripe across a variety of disparate areas of technology. So this is obviously a very different kind of trend from the demographic market trend before. This is a pure geek tech trend and it’s also a combination of near-term obvious stuff like neural networks, that have already happened – microbial stuff I’ll focus on because that’s the current activity that’s pretty exciting in terms of what the artificial evolution could do. In the long term, the big hairy unsolved problems like: how do you make operating systems not crash? How can you make artificial intelligence that exceeds that of a human?
These kinds of problems that may seem impossible to tackle may be solvable using evolutionary algorithms. So I’ll explain what that means in a moment and that’s why I think the trend is important even though I’ll dive deep a little bit into microbial stuff because it’s the here and now. I think the trend transcends that. So what are we talking about? What in the world is this thing? So evolution, we all know biological evolution, we’re trained about that – variation, selection, differential selection – you get evolution. In fact, some like Daniel Dennett will tell you that you will always get evolution if you have the simple ingredients of variation, selection, the rinse and repeat. In every context. So you can instantiate this in a computer program that just trains computer programs and breeds them and cross pollinates them and selects the one that does some task well. Brute force over generations. That’s an approach to artificial evolution programs.
So the forecast, the trend, the near term trend that I guess we’ll ask for a vote on, because there’ll be a mishmash of stuff in here, is that in the near term, the next year or two, the important components of the most successful and the most robust microbial re-engineering projects will involve some form of evolutions in their making. They won’t just purely be designed. It won’t be biotech where you splice a gene because you suspect it works, stick it in an organism, have it do something useful, and have complete control over the process. It’ll be a hybrid, where you designed for evolution. You design something, then evolve it further to make it more successful. So I step back a bit: where has this been successful? It has been used, as I mentioned, in neural networks. So if you have speech recognition, almost all those are run by neural networks that are not programmed by people that figured out how to recognize speech. When a computer wants to recognize either vision systems or acoustic signals it uses a neural network much like our brain is trained through an iterative algorithm.
Currently in the microbial work – I’ll get to that more – people are doing directed evolution and adaptive evolution where they, what they literally do is cripple a microbe so that all of its redundant metabolic pathways are removed. And there’s only one way that it can live. One way it can process a food, a feedstock, to produce the sugars and energy it needs. And all other redundant pathways are removed and the reason you do this is not because you don’t need those other pathways. In fact, you’ve crippled the organism. It’s because the organism will then evolve to do the one thing it can do to survive better and better and better. And then that one thing you’re left with makes some chemical of interest as a by-product.
So this feeds into a whole raft of companies that Vinod and I invested in, and our firm DFJ, that are doing this as we speak, to make chemicals of interest, industrial chemicals, biofuels, everything from jet fuel to diesel substitutes and what have you. And a lot of things in the petrochemical industry that aren’t as well-known – in pure industrial chemicals. So, the future of this – oh, let me just say what it’s produced so far. So so far in terms of published work, people doing design using their brain have been able to improve the microbes’ ability to make chemicals about 4x in some easy cases – just put the genes in, it does what you want, about a 4 X improvement.
The company, same company, in this case Genomatica, that then evolved that organism with the method I described, found a 20x improvement in the chemical of interest. So the organism that reproduces the fastest, by definition makes the chemical you want the most. And all you do is reproduce – they reproduce like bunnies on steroids – every 20 minutes you’re pulling off a new generation that’s the fastest growing and so on. And by the end, you have a solution that’s better than any human’s design, and no one knows how it works. And that’s how evolution tends to go. This has been applied in analog circuit design – there’s a little company, not even venture backed – that’s got 23 patents all done by computers running the algorithms to do analog circuit design, antenna design. And they claim that they have machines routinely beating human intelligence. Better than any human on the planet in those two domains.
And that’s just the beginning – what I hope to excite you about is a trend that’s playing out today in microbes, in the past in pattern recognition, in the future in what we would think of as AI or human intelligence gray. And I think it’s important.
Vinod Khosla: Half
Josh Kopelman: Neither. Had no idea what Steve was talking about
Roger McNamee: Green. “This is a matter of trust. I’m going with Steve”
Joe Schoendorf: Neither. “I wish I was smart enough to know what the hell Steve was talking about”
- Vinod Khosla (Founder, Khosla Ventures)
- Ann Winblad (Partner, Hummer Winblad Venture Partners)
- Ram Shriram (Managing Partner, Sherpalo Ventures)
- Joe Schoendorf (Partner, Accel Partners)
Trend (1:17:10 – 1:24:00)
Steve Jurvetson: So perhaps like Vinod’s forecast, I’m trying to look a little farther out than just the next year. So I think there’ll be some important trends from this that play out in the next year but it’s more over the next five or so years that I see this having an impact. What is it referring to? The distributed web. The aggregate power of all of you – what you think, what you do, your daily activities – will have some pretty profound impact not only on large media properties of the past, but also telecom industries, and I’ll try to end with one forward-looking example: search as we know it going forward.
So what does the trend say? That innovation occurs at the edge. This is true in biological innovation, this is true in startups. It’s not the warm center of the thinking or centralized model, but people out on the fringe. Sort of as a metaphor. More precisely, what I’m referring to is the power of you. So, user generated content is already more important than the centralized media of the world. Most of what a teenager reads today is written by someone they know. Which is kind of astounding – it speaks perhaps to Joe’s point. We’re inevitably moving down a path where the aggregate creation of many authors is more important than a few centralized media outlets. If you think about where information is spread and how people communicate, you look within social networks, it’s again, among clusters of small social graphs – a group of peers communicating with each other within a social network, not all reading the same centralized star topology of information flooding out from one place. And recently there was a Nielsen report just a couple days ago showing that sure enough, as you suspected, more time is spent on social networks than email. So we’ve reached that tipping point already. Now that’s the past.
What I think is interesting about the trend is, how can you leverage this from a business sense? Where are the opportunities for new entrants to make some money off this, to aggregate and take advantage of this capability at the peers, out at the nodes, out at your clients’ desktop, as opposed to a web server. And of the interesting things, I think, is that although intelligence moves through the edge, whether that be web services, whether that be client software on your iphone and iapps, whether that be peer to peer networking and what you can do with that, that’s a trend that’s been going on for many years. But people haven’t really figured out a business model. How can you take advantage of all that aggregate output that’s out there? In fact, the concept of crowdsourcing is one way of thinking about that – how can you rise up one layer, and say how can I take advantage of the group? The fact that half of this year’s were crowdsourced, if you will, is one indication that things are tipping for the first time.
So how do you take advantage of this federated group? Let me give three quick examples.
In the area of content, let’s say you’re a media property, let’s say you want to reach women on the web. iVillage has been trying this for 10 years with a centralized model. And recently they’ve changed, but for 10 years, they tried the centralized model. And they were stuck at around 16 million uniques, meaning it just wasn’t that compelling to women. I don’t know how many of you go to iVillage regularly, but in aggregate, the statistics say they’re not the winner. They were number one; they’re no longer number one. A new entrant from 2005 came in called Glam, an Accel portfolio company as well as one of ours, to be fair. And they aggregated all the long-tail, all the little blogging sites, all the user generated content, rolled them up, and provided an advertising model, a business model, for all that content that was going unmonetized. And they rose, within two years, to be the number one women’s site. One year later, they became a top 10 property in the internet, according to ComScore. More than Facebook, more users per month than many other companies you’ve heard of. And certainly within their demographic, which is aiming just at women, they exceed the traffic of all traditional media properties combined: iVillage, Yahoo Women, AOL Women, Conde Nast, and just about every one you can think of that was targeting women. Add them all up – this one user generated content blogging aggregation play gets more views per month. 100 million a month. Pretty amazing. And they are just one example. There’s countless examples of this. If you’re Facebook, right? Very interesting story, right? How would you best advertise on Facebook? We’re just now learning that you don’t, let’s say, if you’re BMW, you don’t advertise to someone who says “I like BMW. I have one.” Much better to look at the social graph and advertise to the person whose friends say they like BMW, but they don’t. They’re more likely to be a buyer than an already-existing customer or enthusiast. And we’re just beginning to see the beginning of this. I think UGC will trump centralized content.
I’ll be more brief in the second example. Telecom. Trillion dollars a year services industry. Radically restructured. One day, you will not pay a phone bill as a separate bill. That is just a truism. And it’ll probably take 20 years before we look back and say, “Oh yeah, I once paid a phone bill” just like one day, I mean today, you look back and say “I once paid an email bill. I used to pay for email software and I used to pay an annual $5 subscription fee.” It’s just a data service. Telephony is just a data service. Video conferencing is just a data service. And when Skype spins out of eBay and goes public in six months, which they claim they’re going to do, you’ll finally see what they always dreamed of. Which is mobile communications. They were Europeans, for God’s sake – they never thought you’d speak with a computer. Their founding vision was wifi handsets – and it’s been the carriers’ perpetual attempt to crush Skype that has prevented that from happening. And because eBay is a multinational corporation they have to kowtow to the carriers and their overt pressure. A free standalone company – you’ll finally see the wifi phones. They’re always meant to be. Whenever you’re at home, work, or anywhere you get wifi, you just talk for free. And that’ll be, of course, the catalyst for changing the entire telecom industry to data services overlay.
Last example – looking forward. Something that hasn’t been announced yet. But there are a number of companies working on an entirely new, a better way, of doing search. You hear some rumbles about it around Twitter and other products that can give you the real-time web. What’s hot right now? What’s going on right now? The blog search engines don’t quite do it – they’re getting better. What you’d really love to know is where are people surfing right now? For certain types of search. Not information search. The Wolfram Alphas will still have a role. Google will still have a role. But there’s a kind of finding of information that’s missing out there. And if you think about it, a lot of that has to do with web crawling – the centralized model. Build a big data center, heat it, put tons of energy into it, and try to know everything from the structure of the web. The founding vision of PageRank. Much more effective would be: What if I knew what actually was going on in every browser? Where you were actually surfing. If I could preserve privacy, yet watch where people surf – where do they spend time? You will get recency (for free), you’ll avoid spammers and SEO (kinds of problems – the search engine optimization game of cat and mouse that Google has – so all the bad results around IRS searches and stuff that have been notorious would go away for free), you get the deep web for free, Craigslist, eBay, things you didn’t even think to find in the search engine. Because you need to be very fast and not put a load on all those sites that ban robots. You get all that for free too. So all the problems with search are solved if you instead distribute the entire search engine, have no servers, no crawlers, and federate the desktops of many individuals and what they do day-to-day. There’s companies like Wowed that will be soon releasing a product in the US – it’s right now only in Europe – and others that are pursuing that vision. And I think that’d be quite exciting. It may fail, but if it succeeds, I think it would harness the power of the collective wisdom of the crowds in a way that you can’t do with just algorithms and centralized servers. I think you all are smarter than any one computer.
Vinod Khosla – green
Ann Winblad – red
Ram Shriram – red
Joe Schoendorf – green
- Aneesh Chopra (CTO, United States of America)
- Ajay Royan (MD, Clarium Capital)
- Paul Saffo (MD, Foresight, Discern Analytics)
This year the Churchill Club tried another different format – none of the panelists brought any of their own trends. They simply commented on the ones proposed.
- Kevin Efrusy (GP, Accel Partners)
- Bing Gordon (Investment Partner, KPCB)
- Reid Hoffman (Partner, Greylock)
- Peter Thiel (President, Clarium Capital)
Trend 1 (33:35 – 36:10)
Moderator: All vehicles will go electric. Remember, these trends, the timeframe is five years, huge growth within five years. So eventually all motor vehicles will transition to – we already have a red card back there – to an electric drivetrain affording greater efficiency, convenience in a multitude of new design options. Within five years, this inevitability will become clear. This is Steve’s trend – this excludes rockets, I assume – and includes golf carts. So we’re already on the kind of the shallow part of the curve with golf carts and Segways. But, Steve, Explain.
Steve Jurvetson: Sure, and very important. Of course, if anyone was saying red already that’s because it’s not going to happen in five years. We’re not going to switch a billion vehicles in five years. The point of the trend is that hopefully by the end of five years, most people in the room wouldn’t disagree with the trend. So maybe it’s a bit self-referential in that way. So why is it important?
First of all, the US spends two billion dollars a day on oil and another four billion a day if you count full indirect cost, economic and military, to protect that supply chain. It’s big. If you ask anyone, inevitably we will be off oil. You just have to look far enough in the future. It’s kind of an obvious point. You can’t do it forever. It’s a question of when, how. And so why would it be an important trend over the next five years? What’s going to be catalytic to change all this?
Well the first is why are you going to go electric? What about hybrids or biofuels or this or that? So even if you wanted to burn oil for some reason, and you wanted to burn ethanol and biofuels, you’d be better off doing it in a centralized, large plant – the Cogen plant from Siemens is about seventy percent efficient converting those fuels to electricity – transmission line, battery. You’re still more efficient doing that than today’s cars, which average about 20% in waste. All that extra energy and heat. They’re basically too small an engine to burn oil. So even if you want to burn oil, don’t do it in these little engines.
Second point of course, is that it’s just much better if you don’t burn oil. That’s sort of obvious, right? An electric car can be up to 88% efficient – the ones that are driving around today. Okay, so it’s a great endpoint. How do we get there? Why now? If we can show a slide, I actually put some visuals up to give people a sense of why you might want to do this. And the short argument is that it’s going to be cheaper, more convenient, and give you all kinds of interesting new design options.
I don’t know if you can put the slide up – you have it up there? Okay.
So there’s a bunch of vehicles, all different types, all different cars that are going to come out either in the next few weeks or in the next couple years. Back to 29 different electric cars coming out in the next 18 months. Interesting thing about all these that you see up there, all these passenger cars, total cost of ownership is less than a Ford Taurus. They’re really nice cars. They accelerate like crazy. The one on the top right with that really funky doors – an SUV with more space than a minivan and more convenience but outperforms a Porsche 911 both acceleration and handling. These are cars people are going to want. They’re going to be cheaper to own, there’s almost no maintenance, and it’s just a more efficient way to go.
Now this is great on the high-end, but what about the rest of the world? Wouldn’t China and other places want to also do this? And how might they afford to go electric? And I’m here to tell you today there are now 200 million electric vehicles buzzing around China. Today. They’re mostly two-wheeled scooter Vespa-like things, but there’s 200 million of them. If you can go the next slide please.
It’s been the most popular vehicle in China since about 2007. It’s been growing like crazy – originally around SARS fears, and now they realize it’s cheaper to do one of these than to ride a bus. So in summary, well all I’m saying, is that people will realize, not that we’re shifting a billion cars in the next five years, but we’ll hopefully shift a billion minds. Just like we did around smoking, which we thought was inevitable. Just like our cars, they should not be smoking. And I think we can shift that zeitgeist and that people’s thinking around it because they’ll realize the alternative is not like going off the binge or the addiction of oil – it’s a car you’re going to want to drive, that’ll be cheaper to own, and frankly is a lot more fun .
Kevin Efrusy – Red
Bing Gordon – Red
Reid Hoffman – Red
Peter Thiel – Red (exchange is really funny)
Trend 2 (1:20:55 – 1:25:50)
Moderator: Moore’s Law accelerates beyond silicon. This is just like a really big idea in a fairly quick statement. When we consider Moore’s Law in the abstract, the dropping cost of computation is not tapering off. Rather, it will accelerate further still as we look beyond the silicon era. What’s beyond the silicon era?
Steve Jurvetson: Something big, new, and different. If we can put the slide up please. So obviously I’m not going to try to defend Moore’s Law – it just chugs along and nothing’s new. That would not be new. In fact, Moore’s law is in many ways the mother of all trends, sort of metaphorically in the sense that there’s not any other trend you can think that last 110 years, is important, and relates of technologies – at least I can’t think of one. And if you look at Kurzweil’s version of this, which you see a sort of a modern rendition up here, there’s nothing about Moore’s Law that’s specific to integrated circuits. It translates back through vacuum tubes to mechanical devices and on this – by the way, this is an exponential curve to look at the right axis – a straight line would be an exponential.
So the trend, the prediction, the argument is that if in fact we’re going to pick up in speed, that the experience of how much compute power you can buy for a fixed amount of money is actually going to accelerate beyond the pace we’ve been on. And one simple way would be if this curve in fact is curving up, which some data analysis indicates it might, that trend would be a self-fulfilling prophecy. But how? Beyond silicon, we’ll probably move to a moleculartronics, doing things with carbon nanotubes, as IBM and a bunch of companies are all predicting, it will have to keep scaling things smaller and smaller at the transistor level. But it’s also, I should come back to this to say, it’s also the mother of all trends in that just about everything we’ve talked about on this panel and just about everything disruptive in entrepreneurial innovation relies on Moore’s law continuing. So whether you’re in biotech, telecom, obviously computing, software, increasingly just about every industry, eventually becomes subject to simulation science instead of a trial and error experimental science. And it moves on to a much more rapid pace of innovation. So this is what we see in industry after industry and why entrepreneurs keep having opportunities to start new businesses in new and interesting areas. So it’s important. A lot of people within the industry predict it’s going to end when they look just at silicon.
So what could take us beyond this? Well, it’s not just getting them smaller, smaller, smaller, smaller, because eventually you’re going to reach some fundamental physics limits. It’s going to be architectural improvements. Even Intel’s CEO, who has a lot to vest in the old model, says we’re going to have much more computational and architectural improvements in the next 10 years than the prior 30. Not Moore’s Law itself, but the systems level improvements. Things like mimicking the brain, which is 100 million times more power efficient per calculation than our best computers.
But let me share something if I may on the next slide, which is completely different. Something to give you – I try to geek out at least once every one of these sessions. Something completely out of left field that may or may not change the world and is nothing like what you’ve seen before. This guy, Jordy Rose, started a company called D-wave. Uses little niobium circuits that are super cooled down just above absolute zero to engage literally parallel universes to compute. Sounds like science fiction – he’s here, sold one to Lockheed Martin.
So if you go to the next slide, and I’ll just try to end with this, if you do the build. He’s plotted a curve, I called it Rose’s Law, with just a few data points going over seven years, showing the number of qubits again growing exponentially. But here the power of the computer grows to the power of 2^n, meaning every time you add a single qubit, a single qubit, it doubles the power of the computer. So if you go forward one more click, soon, within a year we’ll have computers that are competitive with existing ones. In 2009 they already out-performed Google Goggles by Google – they used this computer to outperform their own data centers. One more click, it gets kind of weird. Suddenly it starts to outperform all computers on earth combined. One more click, it outperforms the universe. One right click, what am I talking about? This is if you use the entire matter of the universe to compute for all time, they can solve certain problems that are otherwise unsolvable, like the Traveling Salesman Problem, portfolio optimization, things that are used throughout machine learning that are used today through heuristics – we do our best. We don’t actually solve these problems. It relates to molecular simulation and everything. Now, it may or may not play out. It’s just one example. But it’s completely unlike anything we’ve ever seen with Moore’s law before, and if it continues for just three more years the way it has in the past seven, then within the next five years we’ll see computing accelerate way off of Moore’s law within the scientific community domain.
Just try to summarize: It’s that it’s not just silicon. There’s new things coming. And it could accelerate the pace of change when you measure it not as transistor counts, but it’s computational power that you can buy for a certain amount of money. There’s a 25 gigaflop computer being developed this year by Nvidia, the GPU company. Fastest computer on earth. It’s a [compiative (???)] different architectures, different ways of building computers that will continue Moore’s Law beyond what we’ve known it to be.
I’m arguing for a curve that leaves – exactly. On a on a linear scale it’d be massive, on an exponential scale, it’ll be like that [hand gesture curving up].
No, it’s actually much lower energy consumption per calculation of course. Otherwise it’d be nonviable.
You have to have a different programming model, more like machine learning where you just you program inputs and outputs like a neural network and it just learns.
By the way, I should say, I knew in advance that my trends were two of the things that Peter disagrees with more than anything else I know that he disagrees with.
Kevin Efrusy – red
Bing Gordon – green
Reid Hoffman – red
Peter Thiel – red
- David Cowan (Partner, Bessemer Venture Partners)
- Venky Ganesan (MD, Menlo Ventures)
- Alfred Lin (Partner, Sequoia Capital)
- George Zachary (GP, Charles River Ventures)
Trend 1 (33:50 – 37:15)
Moderator: Deus Ex Machina: Machine Learning Innervates the Tech Frontier. Machine learning is the technology under the covers that power many of the exciting new products that leverage big data to appear nearly magical. Imagine a Google Research approach to everything.
Steve Jurvetson: So, this is a bit abstract. But I think it’s really important, and hopefully interesting. Innervate means to envelop, take over, sort of like a nervous system. To add intelligence to things that formerly weren’t intelligent.
So what is machine learning? It’s a set of software algorithms that allow computers to learn patterns in data without explicit programming models. So as a programmer, or designer, you don’t tell the computer how to recognize faces, or human speech, or translate. You just give it thousands and millions and billions of examples, and give it a learning engine. So it’s a little out of control, you sort of get what you want, but you don’t have precise control of the process. You in a sense have to relax your presumption control. So companies like Google are all over it as a humble approach to the world. Later, physicians and bankers might eventually adopt it whole cloth. But not early on, because of that relaxation of control.
Now, why is it important? Well, the argument, the prediction is that if there is anything in the next five years that blows you away, like Siri might have, or Google autonomous cars would, or these Google glasses, or Google beer goggles thing I just described, it probably was made possible because of machine learning. And the thing to pay attention to, as a technologist, a programmer, or someone in a tech or nearly-tech business, is how could I use machine learning in my business?
If you look at big data, you hear a lot about that, it’s not interesting if not for machine learning. There’s nothing special of big data this time around that’s any different from data overload, there’s data everywhere. It’s the way we now process data. Like Google processes data. The same is true for the Internet of Things – sensors everywhere. All these phones being sensor networks – only interesting if you have machine learning to process that which is beyond human comprehension.
Increasingly, the artifacts we want to build exceed human comprehension. You couldn’t sit down and program the Google autonomous car to do what it does. You need a learning algorithm. It’s like growing a kid or growing a brain. You don’t control what you make, but you control the process of its creation. It’s a very powerful tool.
So the prediction, again: things that blow you away. What might that be? Humanoid robots entering the workplace, from Baxter from Rethink and others, that seem and act uncannily human and have common sense in the way they behave. That’s just now starting to ship. You may or may not have been exposed to some of that. Or the Google Glass applications. Or autonomous cars, for anyone who’s worried about a kid that’s eventually going to drive. Depending on how old they are – if they’re less than 10, they may never need to drive. I hope my kids never drive. And I actually honestly believe they won’t. And that’ll be great. Real-time translation services. Just about anything you’ve ever heard of at Google that sounds interesting and new, was based on machine learning. And it’s the one way to make coherence out of their business strategy after the fact. Is that they’re looking at every industry. Everywhere that technology is starting to percolate into an otherwise prosaic non-tech industry. Apply big data, apply machine learning, revolutionize it. Even places like personalized medicine and eventually places like – actually, it’s starting to happen in banking, to tell you the truth. They’re already starting to use it for programming trading. But it’s just starting. Because of that out of control element of it.
So again, the prediction is: machine learning is important. It will innervate businesses that weren’t smart before, and make them smart, and it’s important.
David Cowan – red
Venky Ganesan – green
Alfred Lin – red
George Zachary – green
Trend 2 (1:24:50 – 1:28:00)
Moderator: Erasing the Digital Divide Ironically Accelerates the Rich-Poor Gap. Technology democratizes upward mobility and raises the bottom of the pyramid bu stretches it into a conical spike – where an ever-shrinking percentage of people control the info economy embedded with winner-take-all network effects and power laws.
Jurvetson: This is an awkward one for me. I am a raging techno-enthusiast. I almost always talk about very geeky subjects. I’ve done this panel for 10 years and I’ve only talked about geeky subjects. I know nothing about politics. But this is such a large societal trend that seems to be the elephant in the room behind so many debates that are talking about other things and missing the core issue that I thought I should at least surface it and have us explore how technology might play an intimate role in the future.
So the rich-poor gap – pick your various statistics to the scale of countries, companies like Google or Apple, or people and power laws and income distribution – it’s been accelerating over the last 30 years. Most recent statistic I saw was in 2010: 93% of the income growth in America accrued to the top 1%. Now, whether or not you believe that, in the past or the present, doesn’t really matter. My only point is: what if that’s not self-rectifying? What if, in fact, it doesn’t just get better if we ignore it? What if, in fact, we’re at peak jobs and the future where every business becomes an information business – something that I proselytize, and that’s everything that we invest in, is the conversion of traditional, non-tech businesses into tech businesses – brings them onto the information economies and network effects of global information work.
If you think about why that might accelerate the rich-poor gap, the first obvious one we just talked about with the MOOCs, which is, if you’re competing globally all of a sudden, life might be a little bit more difficult. Historically, you’ve had regional isolation separation. So imagine you’re a mediocre barber. You could have lifelong employment if you move to a small town. True for most service industries. As soon as you’re competing globally, let’s say as a programmer, as an information age worker, you can’t be a mediocre one if you want the best jobs.
So the competitive landscapes with crowdsourcing, with MOOCs, democratizing – this is all good. Raising the bottom of the pyramid for all, democratizing access for the global population. Oh by the way, two billion people come online in the next five years for the first time. Is going to create an influx of new talent competing for jobs. And these network effects – you don’t 100 companies winning in search, 100 companies winning in social networking. You see a few. And more and more businesses will be like that. This could rend the fabric of society, if we don’t address it in some way, to think: how do we get to a post-scarcity world of abundance? Where you don’t worry about your job to provide healthcare, food, shelter, and clothing. That sort of, in some way, that I can’t even anticipate, is not your cause of existential angst. And you can focus on creation, and nirvana, of this world we might have in the future.
But in the near term, I see a bifurcation of people. Those who are on the tech bandwagon, and those who withdraw. In the past, you could recover. In the future, as the drum beat of destiny becomes decades instead of centuries, I don’t know if you can ever catch up if you opt out of the technology venture of the future.
David Cowan – red
Venky Ganesan – green
Alfred Lin – green
George Zachary – red
- Shervin Pishevar (MD, Sherpa Ventures)
- Rebecca Lynn (Co-founder & GP, Canvas Ventures)
- Jenny Lee (Managing Partner, GGV Capital)
- Bill Gurley (GP, Benchmark)
Trend 1 (37:55 – 40:50)
Moderator: The Skynet Economy: broadband access for the unconnected billions. From thousands of satellites orbiting around the poles, and new airborne transponders, the entire Earth will be bathed in broadband, bringing an unprecedented influx of human talent to the global economy.
Steve Jurvetson: So I’ll try to be different. As I thought about Top 10 trends, I tried to think of something that was profoundly important for a huge number of people. And this, I think, will be for most people on Earth. Those people who are not currently on the internet, who don’t have access to all the things we take for granted – whether it be online education, whether it be access to resources, starting a company, connecting with people, communicating. They’re fundamentally decoupled from the global economy, certainly decoupled from modernity. And 2/3 of those people who aren’t on the internet today are that way because they live in rural areas outside of the cities, and it’s just too expensive run fiber out to all those different places. So they don’t have broadband.
Some of you are familiar, if you want to show the slide, with Google’s approach to this. They are going at it with drones and with balloons to try and cover urban areas primarily, and that’s great. It’ll go directly to the handset, route around governments and telcos and direct to the consumer. That’s fantastic, but it doesn’t really solve this last mile problem out to the hinterlands.
So if you go to the next slide, what I’m describing is a new generation of low-altitude satellites, much lower than geosynchronous satellites, about 35x closer to Earth. So the latency is actually better than fiber for most terrestrial applications. So think of something that’s fast, about 16 gigabits per second, to each space station, and only possible now because of technology in phaser ray antennas and some new solid state amplifiers that make these things cost effective. They wouldn’t have been just as recently as five years ago.
So you have thousands of these satellites circling the Earth so that every part of the Earth is equally covered. Every remote farm, every boat, plane, car, anywhere you might to put a ground station. And if we go to the last slide, I can show you what some of those look like. This one that you see here is from a company called OneWeb backed by Virgin and Qualcomm. And it literally is everything that you need: the solar cells, the batteries, the transponder (which is in that little dome). It’s $250. And you get 16 gigabits per second anywhere. No power needed, no grid needed. You can put it in a village, on a school roof, wherever.
This is going to be transformative for the developing world. And I think it’s also going to be transformative for all the entrepreneurs in the room who are thinking about their target audience in the future, which could be, not just three billion people on the internet, but eight billion people on the internet by the year 2025. And by the year 2020, this constellation will be first operational. That’s the deadline set by the regulatory bodies that issues spectrum rights. And so SpaceX and OneWeb are racing towards that goal and fully intend to have two constellations operational by then. There’s possibly a third one in Europe as well.
This is going to profoundly change the lives of these people. They can have access to online education, they can entrepreneurs, they can do the sort of things we take for granted. And I think it allows the American Dream to reach everyone on this planet. And to me, I can’t think of anything more important over the next five years than giving people the basic access that makes all the other trends that we’re going to talk about today possible
Shervin Pishevar – green
Rebecca Lynn – half
Jenny Lee – green
Bill Gurley – red
Trend 2 (1:28:25 – 1:31:30)
Moderator: Rise of the Robocars: Driven by a Machine. By 2020 we will no longer debate the inevitability of autonomous electric vehicles when we first experience the convenience and efficiency of urban autonomous driving services.
Steve Jurvetson: So again, trying to think what’s really big, important, going to affect a lot of people. The $2T automotive industry comes to mind. The fact that we spend $2B/day on oil in the United States, just the United States. And it’s an incredibly inefficient system. To be clear here, I’m not saying you’re all going to have robocars – I wish I could have one by then, I probably won’t. But for those of us who have the chance to be in one, whether it’s in Vegas, wherever the pilot deployments occur, there’ll be one of those epiphanies. Kind of like she had driving her Tesla that’s like, you would never go back. Like once you’ve experienced what this can be, you’ll see what it can be.
I’ve been in these vehicles, you can show the one slide I have if you want, several times, several different types. And I can tell you: I believe they’re already safer than my parents. And I would trust my kids with them. No doubt about it. And they’re just going to get better. The sensors are getting cheaper, the Moore’s Law is improving. And the opportunity here is quite amazing. So initially they’ll be low-speed services, 25 mph or less in urban settings. Something like a robo-Uber or robo-Lyft. It’ll be feathered into those services of course, as well. Not just new entrants that are doing it.
And what it’ll offer is unprecedented efficiency – both fuel efficiency, timing efficiency, the safety concerns that people mentioned would of course go away. It will be the ultimate future. But it’s also going to extend out to all the other kinds of vehicles. Every automotive maker is working down this path. And it’ll be for both urban and highway driving, and it will offer you opportunities that are roughly 10x better. All the deaths that were mentioned, all those problems, at least a 10x improvement, 10x efficiency, and the number of cars you need per city to get the same job done.
With many fewer taxis than we have in New York today, you can get one of these within 30 seconds of wanting one. At a much lower cost per mile driven. It’s in fact more cost effective than mass transit itself in urban and suburban environments, which is kind of amazing. The thing that many of you probably felt today trying to drive here, is that hellish commute that we all experience. The average American spends 52 minutes a day, 4 billion hours a year, wasted in a car. These autonomous vehicles, because they know before they move, are they going to accelerate, brake, turn, they can compensate with active suspension to totally eliminate all the drifts. So you could read without getting nauseous. You could actually be productive. You could be drunk. You could do whatever you want in your car. Because you’re not driving. And, this to me, will be such a compelling future that it will drive urban redesign, the removal of parking lots. The fact that people spend 40% of their urban time looking for parking shows you how painful it is in an urban environment. So it’ll start there. Once you have a taste of it, you’ll never want to go back. And it’ll usher in a future where we’ll just have much safer roads, avoid global gridlock, and oh yeah, and ideally no teen driving, so hopefully your son will be like mine – just avoid those first years altogether. Here you can see crash risk by age. By the way, boys with ADD are worse than drunk drivers every day of their life. I want that to end.
Shervin Pishevar – half
Rebecca Lynn – green
Jenny Lee – half
Bill Gurley – half
- Mike Abbot (Partner, KPCB)
- Rebecca Lynn (Co-founder& GP, Canvas Ventures)
- Sarah Tavel (Partner Benchmark)
- Hans Tung (Managing Partner, GGV Capital)
Trend 1 (21:50 – 24:35)
Moderator: The Revival of Voice. The multi-touch screen UI finally cracked the code for smartphones. Major breakthroughs in voice will vastly broaden the compute fabric of the world.
Steve Jurvetson: This, I think is a big thing that could affect a lot of people, certainly the fabric of society and how we think of computers much like the smartphone did to its predecessors. We gained some perspective on voice from our investments in Skype and Twilio and TellMe, but what I’m talking about today is something entirely different. A renaissance, if you will, in the use of voice as a user interface in our everyday life. And if you have an Alexa or Google at home product, you might know roughly what I’m talking about. And for those who’ve only experienced Siri and the other false starts we’ve had in voice recognition, believe me – it’s getting a lot better.
And let me tell you why. The computes gotten a lot better, the data for training sets have gotten better, and the use of deep learning in the background to make this all work and make it human-like is making a huge improvement. There’s also some good chips by companies like Vesper and Mythic that basically make it a near zero power thing that you could layer onto anything – into a Fitbit, into a Roomba, into any appliance in your home for roughly 40 cents of cost. So you could have an always-on, ambient listening thing that could respond to your voice and only your voice and do what you want it to do.
Again, think of something the size of the button on your shirt that could do all this in the near future. Right now, speech recognition has an error rate of about 5%, the same as a human, interestingly. And just a few years ago, in 2013, that error rate was 23%. So that’s probably your experience – most of the time it gets something wrong in the sentence you’re trying to tell it. So we’re stuck to short commands and things like that today.
But it is already the case, according a recent study out of Stanford, that we can speak as a user interface input three times faster than we can type and five times faster than if you’re using a glass keyboard. It’s just a better way to speak, especially for rapid speakers like myself.
And if you look at Alexa, you get a peek into that future. Alexa’s in 10% of homes today. And today, literally today, compared to a year ago it’s grown 500%. They have partners throughout Whirlpool, GE, light bulbs, all over the place, to bring Alexa everywhere. And you have Apple and Facebook entering later this year with their comparable products. So what’s next?
From interactive voice response, you’re going to go to continuous, persistent communication where it knows you and knows what you want and the context. That requires memory within the deep learning networks. That’s what a lot of the research is focused on, and in the next two to five years all that will bear fruit. These things will be battery powered. You won’t even need an internet connectivity. And you’ll have this long march from the keyboard to the mouse to the touchscreen to voice as the ideal user interface. The thing that is most natural and feels like you don’t even notice that technology is there.
And if that wasn’t compelling enough, think about the rest of the world. Globally, the majority of people will not be using the screens that we’re still stuck in, because they’re either children, they’re illiterate, or they’re too old to use that user interface. So actually, for the majority of the world and the developing world, as they get online medical care and education, it’s going to be by voice.
Mike Abbot – green
Rebecca Lynn – green
Sarah Tavel – half
Hans Tung – green
Trend 2 (1:04:55 – 1:07:55)
Moderator: Steve Jurvetson with The Deep Edge: The Embedding of Inference Engines/Neural Nets/Tiny Brains in Everything. Couple local intelligence to each sensor and the Internet of Things becomes the sensory cortex of the planet.
Steve Jurvetson: So in 2013 when I was on this stage, I gave what I thought was my favorite trend that I’ve ever done in these things and that was this machine learning thing which at the time deep learning, machine learning was fairly nascent and I really think it’s become important. I’ve been continuing to invest in it and follow up on it ever since. And what I’m trying to share today is something new and different that I think is the next phase of this. And that is pushing intelligence out to the edge. And by the edge we mean all the devices that are out there. The edge is your car. It’s your appliances. It’s your Fitbit. It’s of course drones and robots and all those things out there that are connected to the Internet. And you hear about the Internet of Things by a lot of people who don’t really know what it is – like this Internet of Things – because they know what’s coming but they can’t really describe why.
Well, this is the thing that’s going to make it happen. That is, putting a little brain in each one of those sensors, adjacent to it. They have to be cheap, like 40 cents or less. But they’re going to be everywhere. So the perspective I’m sharing comes from the chip companies we’ve invested in since that 2013 prediction. Companies like Nervana and Movidius, both bought by Intel, and most recently my newest investment Mythic, which is doing an analog chip in a really interesting way to dramatically lower the cost and power consumption of these little brains that you can put anywhere. Basically little neural networks, little deep learning engines. There’s many terms for the same thing.
So this Internet of Things, everyone thinks it’s going to be big. 30 billion devices connected the internet just in the next three years. And somewhere between two and 14 trillion dollars of economic value just in that, making it much larger than the smartphone industry. So it’s a big thing. And this I think is the thing that’s going to enable that big thing. In fact, Nvidia’s CEO says that there’ll be trillions of these devices connected to the Internet and he’s not going to be in that business because these are chips that are much smaller than anything Nvidia makes.
Now, Cisco themselves estimate that of the data from these devices that comes back to the data center in any way, there’s 270 times as much data that’s going to stay local and be processed local. So think of a video camera doesn’t send all the video up. It has an intelligent algorithm looking for faces, looking for your face, looking for something that’s not your face before it sends that Dropcam via data up to the internet. It gives you privacy, it gives you better latency, it gives you better security, it gives you much lower cost, much better bandwidth utilization. In fact, if it weren’t for these smart brains out at the edge, the internet would collapse under the weight of all the IOT traffic.
So it’s an essential thing, but it wasn’t possible until now with these new cheap neural networks. The other thing that it also enables is the AI in any device. So imagine your fitness band or your device becomes a fitness coach. Your fridge becomes a health monitor. All these security cameras become a smart security system or built in building inspectors. Your self-driving car has to have this. You cannot rely on the cloud to make a split-second, last-minute decisions on what you’re doing. Those neural nets have to run locally.
And in a sense it’s like we are recapitulating our own biology. We have a sensory cortex and reflexes. We only share the subset of that information to our frontal lobe to make strategic decisions. That’s what the internet and computing [fabic] of the future will look like as well.
Mike Abbot – green
Rebecca Lynn – green
Sarah Tavel – green
Hans Tung – half