The Current State of Things: A lot has been made of artificial intelligence (AI) recently. After decades of disappointment, increased computing power and masses of usable data that computers can make sense of are changing things. Google just announced that an AI beat a human grandmaster at the game of Go partially using a technique dubbed “deep learning”. However, in the real world outside of the laboratory, I still find computers frustrating when it comes to being smart and there are some near term opportunities that are being missed. One of the causes for this is the way data is handled in our current environment. AI cannot work without data to process and learn. Humans have senses, computers have data. Much of the data we generate today is lost or not accessible to developers because it lives in silos. In the course of using and carrying a smartphone, you generate lots of useful data, sometimes called “data exhaust”. There are privacy challenges around this and it’s likely the companies best positioned to solve it are platform companies like Apple and Google who have all the data. A frustrating example in the current state of things: I type Zenefits in an email to a friend. My Mac autocorrects it to “benefits”. I then change it back. I use Zenefits again later in the email (note the capital Z so it’s a proper name) and my Mac (using Mac’s default mail program) corrects it back again to benefits. There is no learning there and I’m actually having to fight with autocorrect. Some Meta level understanding of language and better use of context should change things in the coming years and computers will move from robotic to smart. One of the first areas I hope AI takes aim at is reducing cognitive load for people.
What is cognitive load? A good example of cognitive load is driving. Driving is particularly tiresome in stop and go traffic, rush hour, or an area you are unfamiliar with. It takes mental work but creates no value for you other than taking you somewhere. With the exception of fun driving, day to day driving especially in large cities like Los Angeles, New York, etc adds complexity to you life. Google Maps, Waze and other apps have made things exponentially easier especially for driving in unfamiliar places. I saw a tweet recently that said “Uber is my self driving car” and it’s a good point — Uber has made it much more cost effective to have a dedicated person drive you. At the same time, this is an example of something computers can and should be able to do for us. It is one of the more difficult but exciting examples of AI and a numerous companies are pouring resources into it.
As an investor, I spend a lot of time dealing with small stuff day to day that isn’t value creating but has to get done. These include managing and coordinating my schedule, dealing with companies if I book a rental car or hotel, booking flights, etc. These are essentially commodity tasks. The interface to companies that you buy goods and services from got a lot more exciting with the advent of the web. And subsequently apps. Apps have cleaned up and streamlined the web interface and reduced many clicks to a few taps. I’d love to see AI take these programatic interfaces plus our data exhaust and start doing useful things unsolicited or with simple instructions given in plain English. “Move my flight to New York from today to tomorrow” would include dealing with the airline, maybe getting permission to pay a $200 change fee, notifying the hotel I’ll check in a day later, adjusting the rental car reservation, etc. There are a lot of predictable consequences that come from making a flight change in the example and Google and Apple have all this data in email, calendar, etc. They could also build a repository in their OS for apps to put this data for developers to build AI on top of.
This is not a new vision. You see it in movies, examples include Jarvis from Ironman (Mark Zuckerburg is trying to code one for himself right now) and go all the way back to Apple’s Knowledge Navigator created in 1987:
In the video, the screen and graphics look clunky compared to an iPad today. We now really have all this data on a global world wide network (the web, which didn’t exist at the time of this concept video), we can do the collaboration in Google Hangout. Computers are also pretty good now at turning speech to text. But the intelligence is missing.
Weak, Medium and Strong AI: The AI you see in the video involves a mix of challenges, some requiring strong AI, where the conversation with the professor has with his agent is sophisticated and nuanced. And others are more mundane, booking an appointment into a calendar. So to classify the nature of AI into three groups is helpful — weak AI, medium AI and strong AI. I’d like to see AI move from the labs into our every day lives, and it feels like weak AI should be ready now. Siri and Google Now are good examples of weak AI that most of us have access to. The voice interface on the new Apple TV is really effective because the scope is limited to TV and movies — in that context it does much better. And Amazon Echo is presenting a voice interface to the smart home. So expect to see AI get better in specific domains (this form of AI is referred to as expert systems). It’s cool to see x.ai building a personal assistant to schedule meetings for you — we may see some exciting AI companies emerge on top of email. Apple recently acquired VocalIQ which is working to improve the way computers understand human language. AI will become the preferred computer interface replacing taps and clicks.
The best example of medium AI I can think of is self driving cars (or significant driver assist) and there will be a positive network effect as cars share data with each other. Teslas are on the road now providing sensor data that future models will learn from. Some of the corner cases will require strong AI. Except to see medium AI in five years or less.
And strong AI would involved complex conversations with an AI that include reasoning, nuance, and something resembling understanding. Something like Jarvis or the Knowledge Navigator. The arrival of strong AI is hard to predict. I’m not sure when it’s coming but is probably another ten years away. It’s a challenging research problem and will require innovation beyond just Moore’s Law to get us there.
For now, as an angel investor, I’m interested in finding companies that are building weak AI that solves specific problems that reduce cognitive load, improve user experience and overall make life easier.
Jan 6, 2015
As we begin the new year and the United States Congress comes back in session one of the stated top goals for Republicans is to approve the Keystone XL pipeline. The pipeline is intended to deliver oil from Canada’s tar sands in Alberta to refineries in the United Sates, but with oil trading at record low prices, it’s a very odd priority.
The pipeline itself reminds me of the recent NYT article in which the new Whitehouse technology advisor, Megan Smith, is working to wean the Whitehouse off of floppy disks. It’s a reminder how comically slow moving the government can be. To me, oil feels like the floppy disk, an antiquated technology that we should be planning to move beyond. Unlike floppies, our reliance on oil production and exploration is progressively more harmful and far more deeply ingrained in US government policy.
Below is an image of Tar Sands mining in Alberta. Incredibly, Canadians are planning to do this to an area the size of Florida:
Finding and extracting oil is getting progressively harder as the “low hanging fruit” has been taken. Tar sands involve cutting down all the trees in the area, clearing away the top soil and then begins the nasty, grubby process of open-pit mining. Open-pit mining is extremely destructive and the resulting tar sands require many sessions of refining and processing, thus reducing the net value of tar sands oil. In the old days, you could poke a hole in the ground and “light sweet crude” would gush out. Those days are over.
Petroleum has a long history with various uses in construction (asphalt and tar) as well as oil lamps by the Babylonians, Greeks, Chinese, etc. But what really made the petroleum industry a mainstay was the invention and widespread use of the internal combustion engine and the automobile. Aerosolized gas is sprayed into a chamber and ignited, the resulting explosion moves a mechanical piston. This is roughly a 150 year old technology. The combustion engine has evolved, they are less noisy and more efficient but the underlying premise is still the same.
Petrol has been useful for transportation because its energy density is high. With a tank of gas, a car, boat or plane can go a long way. Petrol doesn’t make sense for powering homes or office buildings or putting power onto the grid. For all these needs, we mostly burn coal (an even older technology). Solar, nuclear, hydro and wind are other good technologies for putting energy on the grid. And solar is unique in that it has no moving parts and can be local (on your roof or backyard) so it can be at the point of consumption (your house). Most importantly solar looks to be on a Moore’s Law Curve. And the sun is a source of huge amounts of energy, 470 exajoules in less than an hour and a half, as much energy as humanity consumes in a year.
Petroleum will be replaced just as the hard drive replaced the floppy disk. And flash memory is replacing hard disks. With flash memory, things like smartphones became possible. We are already seeing battery technology changing transportation with Tesla. Indeed, electric cars are fast becoming mainstream. The mass market for computer laptops and consumer electronics pushed battery technology to the point that electric cars possess significant range and power. And battery technology is getting better every day. Meanwhile, oil is getting harder to obtain, is more damaging to the environment, and is often empowering politically unsavory leaders. As Peter Thiel succinctly puts it in his book Zero to One, “In a world of scarce resources, globalization without new technology is unsustainable.”
For politicians who have run out of ideas, how about pushing for Energy 2.0? There are lots of things the government can do, like tax incentives, research funding, and improved mass transit. The Japanese and Chinese are building magnetic levitating trains that go over 300mph. And Japan’s bullet train, the Shinkansen has already carried over 10 billion passengers. Smart grids, smart homes (Nest, etc), efficiency (Opower), solar, battery research and production, and modern nuclear technology are great starting points. For entrepreneurs who are thinking about the next big thing, this is a great area to focus on.
I started programming on a computer when I was about 10 years old. The Apple //e had just come out and I wanted one very badly. They had Apple //+ ‘s in my school and I was lucky enough that my parents bought me an Apple //e for Christmas. I couldn’t believe they actually got me one and it changed my life forever. I started playing games, making my own games, buying books on programming and thought about what it would be like to work at Apple. I added RAM to it, ending up with 128k and had two disk drives for copying games from floppy disk to floppy disk.
When I was in high school they came out with the color Macs and I got the first one, which was something like a Mac IIx. I studied computer science and ended up working for Apple, first as an intern for two summers and then full time as an engineer.
When I came to Apple, Steve Jobs had just come back. His impact was immediate and powerful. He did many great things (including upgrading the sushi chef at the company cafeteria) but one of the things that stands out the most was his “Think Different” campaign. Read More…
A lot of investors have become gold bugs recently. This is understandable given the instability in global financial markets and with the risk of inflation as the US and other G20 countries take on massive debt. Smart friends of mine are buying gold as a hedge against inflation while the US prints money. The US national debt is about 13 trillion dollars, more info here.
After the housing meltdown and recent stock market volatility it’s natural for capital to go to safe places like gold and treasuries. However, there are some cogent counter points to this strategy that I feel aren’t being voiced enough by the media nor by people who manage money.
Some reasons not to buy gold: Gold will hedge against inflation but won’t pay a premium to it. It will track inflation, but won’t give you a premium over it. Gold is a shiny metal with limited industrial uses. Its value is based on history, civilizations have valued it for millennia based on a primitive attraction to the material itself. Gold is atomic, and you know what it is, but therein lies the rub: Read More…
Growing as an entrepreneur is a function of how you grow as a person. It’s a process that mostly consists of making decisions, living with the results, and trying to incorporate those results into future decision making. We’re all bound to this process, but some of us are much better at it than others. Being right is of course important, but being wrong is more important. Being wrong lets you adjust your aim and improve. Without it you are lost. So get good at it. Admit mistakes, don’t be afraid of them, and try to understand what happened. Step outside of yourself and see how open you are to seeing things clearly, not emotionally.
If you can’t admit you are wrong emotionally, if you are basically too immature to do that, then you won’t grow. This is a process that is more emotional and having to do with your character than it is intellectual. It’s pretty rare these days for people to admit they got something wrong. Politicians don’t do it. Companies don’t do it (lawsuits certainly don’t help). Doctors don’t do it (lawsuits again). But great leaders can and should acknowledge when they got something wrong and focus on how to get better. Read More…
When companies are good, or great, they find a way to get things right. And when they’re broken, they can’t seem to get anything right. This fundamentally comes from leadership at the company. So I guess it shouldn’t be surprising that when I tried to call B of A the other day to get information on my multi-million dollar mortgage, I couldn’t get a human being on the line. I waited 20 minutes, and because I didn’t have my mortgage number handy, I was stuck in an automated loop. I had to eventually give up. It was frustrating and pretty pathetic given that I pay six digits of interest ever year to them.
Shortly after, I was flying in to LAX. It was about 11:00 at night, and I was starving because the airplane food is inedible. So when I landed, I decided to call In-N-Out Burger to see if they were still open. There is an In-N-Out close to LAX. I Googled the number from my iPhone and dialed them up. A guy picked up on the second ring. I asked if the one near LAX was open this late. He knew the answer immediately (of course they were) and was very polite.
How is it that a company that I have a multi-million dollar business relationship with can’t get a human being to pick up the phone when I call? Yet a company like In-N-Out where the margin must be $1 / meal gets everything right? Read More…