By: Ben Freeberg
Machine Intelligence and the Second Web
Whether it is as simple as a temperature sensor, or as complex as Hadoop, the open source, distributed big data analytics platform, we will see the distribution of IOT devices, smartphones and tablets continue to boom over the next 10 years. We are at an interesting time in this space, as machines are developing human responses, copying and emulating human learning properties using unsupervised deep learning algorithm and human supervised machine learning. As evident through the development of self-driving cars, machines now have to be able to react to a number of different scenarios.
The human mind is essentially a low cycle computer, but with trillions of neural network connections and with machines exhibiting more and more human
behavior, we are beginning to be able to combine our experience-related intelligence as humans with the large data-processing capability of computers.
This capability has come at an opportune time, as we are doubling the amount of data created every 18 months, and most of the data is i
ncredibly unstructured and difficult to analyze. CPUs are not designed to handle all of this data, so we have seen GPUs, which render graphics on a screen, dominate the machine learning space.
GPUs are able to run deep learning algorithms, which means that instead of GPUs’ traditional rendering of pixels on a screen to produce graphics, GPUs are now looking to extract features from a large data matrix. Amazon’s P2 Instances and Google’s Cloud Engine have surfaced as the top GPU-powered virtual machines in the cloud.
The End of Moores Law and the Entrance of New Computing Technologies
Morado Ventures, and others, believe that Moore’s Law, which states that the number of transistors that can be put on a microchip doubles every year or so, and therefore doubles its performance, has a few more more die shrinks before getting to the smallest possible chip. Once transistors get below 4 nanometers (roughly the size of 400 hydrogen atoms), quantum tunneling begins to take place (hint: this does not bode well for engineers).
Morado believes we are just several years away from edge of where computing can go, meaning data processing on a single CPU won’t get faster. If we continue on the same trajectory with Silicon today, there will likely be more innovation around computing systems, i.e. data centers, neural networks, multicore processors, but we’ve hit the glass ceiling in terms of transistor size and clock speeds. We have to rethink how computers are going to look if we are going to continue accelerate amongst Moore’s Law.
One example is light-based computers. Instead of using electrons on a transistor, these computers use special composite materials and lenses, producing logic gates similar to silicon transistors, to allow us to compute orders of magnitude faster than traditional means. This way of computing will also be cheaper, undercutting the market by 50% in terms of price and reducing the cost to run these light-based computers (in comparison to traditional GPU-run computers) to 1% of the cost. The cost savings is not only in the computing power, but the overall power savings. Unlike electrons, photons minimally react with their environment, and are therefore frictionless, whereas the friction of electrons in silicon can create immense amounts of waste heat.
Another example of a new paradigm in computing is Quantum Computing. Quantum Computers use qubits, otherwise known as quantum bits, as a way to encode data. Like the bit, a qubit can be represented as 0 and 1, however unlike the bit, a qubit can also be in a superposition of both 0 and 1 at the same time, visa vie quantum entanglement. This allows quantum computers to simulate properties of physics, at a subatomic level (i.e. how electron valence shells interact with each other in complex molecular structures).
Net net, Quantum Computers will allow us to understand how to produce and store energy as efficiently as nature does, through means like photosynthesis. No matter how much computing power we have today with the most powerful classical supercomputers, we could never simulate complex molecular behaviors at scale to even come close to understanding a mechanism like photosynthesis.
This industry shift will allow computer systems to only be limited by the data they are aware of, not by their algorithms or hardware.
Let’s Talk Applications
Over the past few years, we have been limited to running more “primitive” operations, but now that hardware is catching up, algorithms are catching up and we can take massive amounts of unstructured data and turn it into actionable insights.
For example, the beverage industry in the U.S. is incredibly fragmented. The largest loss for distributors is spoiled goods, which could be limited if they were given visibility on what specific rates beverages were purchased over what time frame. Companies with a large supply of perishables, such as Coconut water, are losing 30-50% of their goods because they don’t have enough visibility into their own supply chain.
Starting from the distributor and going all the way back to the supplier, companies are creating more efficient supply chains to save money. What used to be done by manual scrubbing of data in the Philippines is now being done with Bayesian regression algorithms at a much faster pace in an effort to squeeze out all inefficiencies throughout the entire supply chain.
As this level of efficiency rises, the demand for, and shift towards, real-time analytics has increased in parallel. For instance, just two years ago Google Analytics didn’t have a real-time option available for users to view their current traffic and user detail.
Now, we can see how changing a single pixel on a site by moving it from left to right affects the integrity and ease of your website.
How Does Morado’s Portfolio Fit In?
Zipline, a drone delivery service for medical devices and goods to remote areas across the world, such as Rwanda, uses machine learning for multiple aspects of its
product, including creating the most efficient geometry for their drone aircraft.
What started off as a toy robot company a few years back, turned into a visionary, humanitarian company which Morado backed in the Seed Round. Zipline has since received investments from Sequoia, Andreessen and Subtraction Capital. From air traffic control to the design of aircraft to the ability to drop a package within 0.5 meters of where you place your pin, Zipline is making full use of the latest achievements in data processing and development.
Savioke has 
made a robot for the hospitality industry that delivers food and supplies to hotel guest’s room. It will even operate the elevator and ring the doorbell (all with a smile). Essentially they have built a self-driving car from scratch, giving its robots the ability to avoid obstacles. Savioke has written the algorithms for a self-driving car in a smaller, safer setting. The applications of a robot being able to provide a human touch without an actual human touch are quite vast, as Savioke plans to enter the Hospital industry next.
Rigetti is building the world’s fastest computer. The company aims to produce a prototype chip by the end of 2017 that is significantly more complex than
those built by other groups like Google, Microsoft and IBM, who are also working on fully programmable quantum computers. The following generation o f chips should be able to accelerate some kinds of machine learning and run highly accurate chemistry simulations that might unlock new kinds of industrial processes. Morado seeded the company in April 2014.
Who is Morado Ventures?
Morado sees the world through a data lens. The companies they invest in all have some aspect of a strategic data angle. Further, the companies either need a first mover advantage, have the ability to acquire data that is not publicly available, or have founders who are veterans in an industry that has not yet been disrupted by technology. The team seeks out investments where the Company has written highly defensible algorithms that are analyzing the data they process in a new way.
Morado sees data disrupting our economy within the next 10 years just as the Oil industry did during the turn of the century. Less than 10% of all data today is structured, and there is a large opportunity to make the world more efficient, especially for industries that have not yet been disrupted by technology.
Thank you to Dominik Andrzejczuk and the Morado Ventures team for assisting us with this post.