Singularity Summit 2012

Olah, Deming & Other Thiel Fellows


For more transcripts, videos and audio of Singularity Summit talks visit

Speaker: Olah, Deming & Other Thiel Fellows

Transcriber(s): María Teresa Chávez, Ethan Dickinson, and Jeremy Miller

Moderator: Next stop we have four Thiel Fellows. You may know but if you don’t, the Thiel Fellowship, formerly known as the 20 Under 20, offers exceptional students under the age of 20 a hundred thousand dollars over two years, plus guidance, mentorship and other resources to leave college to pursue other work which could involve scientific research, founding a startup or working on a social movement. Today we are pleased to share the stage with 4 exceptionally talented Thiel Fellows who will share their ongoing projects. Please welcome the Thiel Fellows.

[Applause and music on the background while the fellows take a seat]

Moderator: Great. So this is the lightning round, each of the four fellows will give a short talk. We are going to begin with Noor Siddiqui.

Noor is from Virginia, she served on the founding team of a Virginia non-profit that grew to have international reach and her experience exposed her to the tenacity and resilience of the world’s poor and also motivated her to work to build a more inclusive world by tapping the currently untapped productive energy at the base of the economic pyramid. Please welcome Noor Siddiqui.


Noor Siddiqui: We need to move closer to the Singularity. How are we going to do that? We need to increase our pace with innovation and we need to remove any obstacles that might be preventing us from reaching that destination. So, circling back to that first point. How are we going to accelerate our pace of progress?

Today, our ability to innovate is restricted by the number of people that develop the world. Only in a developed world do we have the resources to allow our citizenry to spend time expanding the scope of human knowledge and pushing technological frontiers. But, the talent available in the developed world only represents a small fraction of total human ingenuity.

We are underutilizing four billion people at the base of the economic pyramid. The intellectual drive and energy of four billion people is being squandered. Their time is being spent scrounging for the essentials. For food, for water, for shelter. What if four billion more people were engaged in the highest of human endeavor?

People at the base of the economic pyramid earn incomes below three thousand dollars annually. But together they contribute five trillion dollars to the global consumer market. We now have enough information about these markets and enough experiences with viable business strategies to justify closer attention to this global opportunity.

Today, technology enables unprecedented opportunities for collaborations that allow the poor to be participating agents in the global economy. Engaging the base of the pyramid through market-based approaches is essential to raising welfare, productivity, and income. Engagement at the base of the pyramid is a critical part of wealth generation and inclusive growth strategy.

Access to the internet has advanced education, the sciences and global productivity immensely yet knowledge and skills are unevenly distributed. There is a growing disparity in access to information that is compounded by a language barrier. To overcome stagnation, we need to devise methods of learning and teaching much faster.

Most of the world’s population is unable to realize the transformative benefits of the information age. We can narrow this gap through the co-creation of businesses with local communities framing the benefits of information technology. And circling back to that second point: what obstacles prevent us from reaching our destination?

The greatest threat to continue to progress is human conflict. And at the root of all human conflict, is inequality. Severe economic disparity threatens stability. We have witnessed the tragedies that result once social unrest occurs in a society where mass unemployment has left youth frustrated and without hope. We can avoid conflict and stimulate growth broadly by connecting excluded populations to the digital economy.

Humanity’s greatest obstacle can concurrently be humanity’s greatest opportunity. The billions holding us back can be the ones to push us forward. We need to increase the pool of people contributing to the highest regions of human endeavor. Together we can alternate an area of global prosperity. Thank you.

[Applause. Noor takes a seat and the moderator stands up]

Moderator: Thank you Noor. Next stop is Laura Deming.

Laura matriculated to MIT at the age of fourteen where she first worked on artificial organic genetics and bone aging. She is now in San Francisco working to find and fund therapies to extend the human lifespan. Please welcome Laura Deming.


Laura Deming: There’s one fact that never fails to infuriate me. That fact is that every day 150,000 people die of a disease that we ignore.

I remember when I was eight, saying that I wanted to work on curing aging. And it was watching my grandma suffer the aches and pain of chronological age. Watching her try to play with my brother and I when wobbly knees and arthritic joints made playing painful.

I remember very clearly the deaths of three grandparents, three amazing grandparents, from this awful inexorable progression of aging damage that we’ve somehow come to view as something normal, natural and beautiful to be celebrated. At least, outside this room that seems to be the consensus. And growing up as a little eight year old kid in New Zealand that's the only answer I could get when I asked “Why? Why do things have to be this way?”

I can’t understand now what I couldn’t understand then, why anyone wouldn’t want to work on this problem. And so it took four years to find a way to get into a lab, but I’ve spent the past six years working through labs and literature to try to attack this.

A hundred years ago a little 12 year old kid in New Zealand wouldn’t have a shot, with no starting scientific knowledge or resources, at working on this. But the incredible fact about us, the amazing thing about now, is that we’ve got the toolkit to fight aging, to really cure age-related disease. I mean, just think how far we’ve come in a century.

At the start Schrodinger was trying to find ways to estimate the unknowable, mysterious area enclosing our genetic information carriers. Now we're creating card-carrying DNA calculators, computers out of genetic information. My God, we're watching the pathways that we worked on in worms turn into drug-discovering programs in humans, and we're swapping biological circuitry like it's Lego blocks.

The problem in aging research, is while we've got all these preclinical models that show therapeutic efficacy across age-related diseases, nobody's taking them toward the clinic. They're stuck on the shelf in labs. That's the part of the process that I've chosen to work on. Finding, and funding, therapeutics that will change medicine by getting at the root cause of age-related disease.

The good news is, people want to invest in a longer lifespan. It's not difficult to find people who want to put money into the kind of high-risk, extraordinarily high-return projects that we're working on, it's just a matter of linking them and the projects up. So I spend a lot of my time finding and talking to those people and then going to find companies with uniquely awesome technology that VCs ignore en masse in favor of late stage clinical companies.

To give you a quick overview of what I find most exciting – I mean, the science is the most exciting part, so let's talk a little bit about that. If you look inside the cell, there are two major pain points in the aging process.

One is the lysosome. It's a part of the cell that's supposed to degrade waste, but as you age, it gets filled up with this undegradable substance called lipofuscin. If we can find ways to get rid of the lipofuscin, we can reverse a lot of age-related disorders, Alzheimer's, Parkinson's, Huntington's. All these diseases, if you look in the neurons, you can see lysosomes full of dystrophic, undegradable garbage. We can create a therapeutic now, that has therapeutic efficacy now, that gets past the FDA now, by going after lipofuscin.

Another interesting target is the mitochondrion. This is a huge producer of oxide damage. If we can find a way to prevent that, to secure the mitochondrial DNA from mutation, or prevent components of the mitochondrion from getting even more screwed up with age, not only can we think about extending life spans, we can also get a baseline fitness.

There's other cool stuff too. If you take a worm, a mouse gonad from a young mouse and put it in an old mouse, it makes the old mouse live longer. This is an interesting pathway we might think about working on in humans. We can even mutate in star worms and use lasers to – OK, there's way too much to go into right now, which is why I'm going to end with two asks of you, the audience.

The first ask is, if you like talking about aging and science, I love talking about aging and science, so please, come talk to me later today about aging and science. The second ask is, if you know somebody thinking about getting into this space, be it a budding biologist, or somebody with the capital and commitment to invest, make them take that step. We're in the midst of an extraordinary fight, and we need every ounce of manpower we can muster. But the good news is, should we succeed, we'll have turned the most awful paradigm that we know on its head, that is to say, the inevitability of death.

Thank you.


Moderator: Thank you Laura. Chris Olah is next. He is a member of Hacklab Toronto and he's passionate about mathematics and technology. He presently serves Hacklab as one of its directors, and is primarily working on software for 3D printing, that he hopes will help people improve the quality of their lives. Please welcome Chris Olah.

Chris Olah: Reasoning about exponential trends is actually kind of tricky. While they're fundamentally pretty simple, they're unintuitive, and they require quite a bit of thought to really understand.

Unfortunately, many of the things we care about are sort of exponential, they're vaguely exponential, but they're more complicated. For example, technological trends have been observed to be generally exponential – that's why we're here today – but they're more complicated in that they surge and sag as we hit the limitations of a particular technology and transition to other ones.

These quasi-exponential trends are kind of intimidating, because exponential trends are already hard to think about, and these are clearly much more complicated. According to Alan Kay, point of view is worth 80 IQ points, so perhaps if we look at them from the right perspective, we can go and understand them and approach them more easily. I can't promise that it's worth 80 IQ points, but as an aspiring mathematician, there's a perspective I would like to share with you. This is the perspective of multiplicative calculus.

We're going to torture some imaginary university students. I promise you it's for a good cause.


Chris: Building intuition is always a good cause.

Alice and Bob are studying population dynamics, Lotka-Volterra, and so forth. They're going to go into a lab where they observe what happens to a population of bacteria when there's no constraints on the population. They go and collect data hourly, that's what they have to do, for several days, and they start plotting it. They plot all the 10 AM data, and it creates a perfect exponential curve, and they're very happy. This is what we expect. The bacteria divide every few hours, so the population doubles every few hours, so it's an exponential function.

Then they go and they plot all of their data, and certain hours are consistently below the curve, and certain hours are consistently above it, and this is very confusing. They puzzle over this, and after a little while, they realize that it's because there's fluctuations in temperature due to a nearby radiator, and this is causing the bacteria to reproduce at different rates.

This is no longer an exponential function, this is something much more complicated, because you're multiplying by different amounts at different points. They have to go and they have to write a report on this. So they find it awkward to write because not only do they have to explain why the bacteria were near a radiator, they need to go and explain and describe precisely what happened. I mean, this is something awkward to describe, you have to talk to precisely describe and then talk about the ratio between the derivative of the population and the population and then by a logarithm that corresponds to the ratio in which the worms were reproducing.

That’s very tedious to discuss, and is even worse to draw pictures. They go and they meddle through that but, what was wrong? What was so complicated?

Well, they used a derivative. And that’s natural, they were talking about a rate. But a derivative is fundamentally a local linear approximation. It is how much we have to add to move forward in a function. And that’s not what they are concerned with. That’s not what we are concerned with when we talk about exponential trends. We are concerned with how much we have to multiply by locally. And this is an exponential approximation. It is the multiplicative derivative.

This is actually somewhat of a useful notion. It's not useful because it gives us some sort of great predictive power all of a sudden, it unfortunately doesn't do that. What it does give us – anything we can accomplish with multiplicative calculus, we can go and accomplish with regular calculus if we go and we throw in some exponentials and logarithms and a bit of extra elbow grease.

What multiplicative calculus can give us is a way to precisely and easily discuss and reason about local behavior of exponential functions. I think that's really valuable. I hope this is a notion you'll find useful, and I think that it's only a single example of a large class of underutilized abstractions. It would be worth going and investigating whether there are abstractions that would be worthwhile for you to think about that are not the cookie-cutter abstractions that you've been taught in school.

Multiplicative calculus, again, if it's a constant multiplicative derivative, we get a nice exponential function. If it's slowly varying, slowly swaying, we get the sort of behavior we've been talking about with surges and sags, and vaguely exponential functions. If anybody's interested in talking about abstraction and notation, I'd be thrilled to speak with them, and thank you very much for your time.


Moderator: Thank you Chris. Jimmy Koppel is up next, he's the last of our four Thiel fellows. He is an internationally distinguished coder and expert in programming language theory. He worked on the typestate oriented programming at the Institute of Software Research, and developed the precursor to the feature extraction language that is now used by machine learning systems at Facebook. After graduating with dual degrees from Carnegie Melon University at the age of 20, he accepted the Thiel fellowship and founded Tarski Technologies, working to radically lower the cost of software  maintenance and evolution. Please welcome Jimmy Koppel.


Jimmy Koppel: I've heard the comment a few times that while everything else is growing exponentially, our rate of software engineering is growing linearly. So that’d be really bad for singularity, but I'd like to share some research that's made that false, but first, what’s holding software back?

Software is fundamentally about system design. [xx] Imagine that we want to change all the American outlets into European-style ones. It would make travel easier, I could plug things in upside down. But in order to do this, we would have to rip apart every building, and every manufacturer would have to upgrade their plugs, so this isn't going to happen. So we're stuck with this hundred-year-old design.

These are some of the problems with software. That is, at first it might be easy to change, but over time we’ve built more and more based on it leaves on it and have interlocking assumptions, it hardens, like clay, it would change into a brick. Because fundamentally software is still being made, it is no easier for me to change live code than to take a hammer to a wall.

What could happen if we can't change software? Well, that. [slide shows a poster which says "Y2K: A world in crisis?"]


Jimmy: Let's look at our analogy again. To change the outlets, I would both need to replace every existing instance of an outlet, and upgrade all the plugs. The former is very easy in the digital realm. If I upgrade my app, it's very easy for you to download the new version. But if I upgrade my computer, and all the apps on it break, then the developers will have to spend a lot of time fixing it.

We're starting to overcome the latter in the digital realm. That is, there is now a toolset out of the University of Illinois that does the digital equivalent of detecting when I change my outlets and upgrading all the plugs for me.

Let's look at the labor side. There's a case study from about a dozen years ago. Mozilla did the open source version of Netscape, they had been getting all these bug reports in the form, "If you click five times in the middle of this page, it'll crash." In order to get this into a form where they can actually help find the bug, they had to actually shrink this down, and have it say, "If I click once, will it still crash? What if I move that banner?" There's so many of these they just couldn't do it. So they started asking these people on the Internet to help, and sending them free T-shirts.

It wasn't until a year later that a guy came on and said, well, this whole thing was pretty mechanical, and he just wrote a script to automate that. Then he turned around and looked at the programmers, and saw that they were often load version that works, then a new version with the bug, and undoing and redoing all the changes to try to find where the bug was. [crosstalk] So the same algorithm that tried to .

Then a few years ago, someone took this a lot further, and showed that you can use evolutionary algorithms and have the computer do the whole process of finding and fixing the bug. He took a whole bunch of open-source projects, all the bugs for a month, ran his program on them for a few hundred dollars of cluster time, and fixed more than half of them.

Programs are inherently logical structures, so if we actually reason about the code, we go a lot further, we can synthesize new ones. Basically any device with a signal will process it using something called a fast Fourier transform [sp], so people have put a lot of effort into making these really fast. But they're all beaten by [xx], generated by a computer program called Spiral. You can write programs at a high level and have the computer fill in the holes. Someone came up with a great idea for how to write a really fast matrix transpose. Instead of spending half a day fiddling with bits and working out the details, a computer can synthesize it in under two minutes.

They've taken this a lot further. When programmers draw up code, they often draw these diagrams on white boards, so they can now turn these into code. They've used this code to synthesize a piece of itself.

As these tools begin escaping from the labs, I think we're going to see a lot of problematic software going away, and these tools building on each other in this massive positive feedback loop.

So while the past software engineering may seem linear, no. It's exponential, like everything else, and we need to embrace this. You can read more about this stuff here. [slide not shown] Thank you.

[laughter and applause]

Moderator: Thank you to Jimmy Koppel, and to all four, Noor Siddiqui, Laura Deming, Chris Olah, and Jimmy Koppel. Let's have another round of applause for our Thiel Fellows.


Moderator: And amazing, they were all under 20 when they accepted the fellowship, so congratulations to you guys. very bright futures ahead, and really great talks today.