The A.I. Revolution

Mohammad Islam, Investor, DFJMeaningful By We are only at the beginning of understanding what may be the greatest technology ever created because of its unique potential to revolutionize every industry. Artificial intelligence (A.I.) is the field of research that investigates how machines can perform tasks like speech recognition, visual perception and natural language processing. Machine learning is a field of A.I. that has become particularly good at these tasks because it allows us to give machines the ability to perform tasks without being explicitly programmed; instead they learn. And the pace of this technological innovation is nothing but exponential. Within the last five years, a subset of machine learning, called deep learning, has revolutionized the way we can make sense of big data by discovering intricate structures in high-dimensional data to progressively learn. Although decades-old in existence, the deep learning models made up of artificial neural networks emerged from the A.I. winter of the 1980s to finally fulfill its promise, revitalized by extremely powerful computation and enormous amounts of data. The convergence of these trends has led deep learning to become the state-of-the-art in speech, vision, and natural language, and I believe it has the potential to transform nearly every industry. Forboth new entrants and old incumbents, it will be increasingly important to develop an A.I. strategy.

What is Deep Learning?

Deep learning uses artificial neural networks to learn representations of data with increasing orders of complexity as data traverses layers of the network. The neural network is inspired by the neurons and synapses (connections between neurons) within our own brain, but the size of neural networks we can create today still pale in comparison to the 100 trillion synapses present in an adult human being. The model of the neural network is shown below in Figure 1.

These models are incredibly data-hungry. After raw data enters the input layers, each neuron layer “learns” about representations in the data with increasing complexity. In the example shown in Figure 2, we can see deep learning being used for computer vision. The image (a matrix of pixels) is fed to the neural network, which learns concepts of edges, object parts, and then finally object models in order to learn how to identity these in the wild once trained with enough example images (training data).

Companies must embrace the A.I. revolution happening now, and understand how A.I. is going to change the nature of their businesses by enabling the creation of new A.I.-enabled products

A.I. Infrastructure

In order for these deep learning algorithms to process enormous amounts of data, the requisite hard ware was important to enable this. Luckily, it already existed in the gaming market. The parallelism inherent within graphical processing units (GPUs) made them extremely well-suited to run the large number of matrix multiplications within a deep neural network. As the A.I. revolution began to take place, demand for GPUs from NVIDIA, the market leader, increased rapidly and its stock price was rewarded handsomely (shown in Figure 3).

Moreover, recently there has been a proliferation of specialized deep learning chips entering the market. The semiconductor space became revitalized because of A.I. DFJ was an early investor in Nervana, which was building a cloud-based training platform for deep learning that was to be powered by application-specific integrated circuits (ASICs).These specialized chips would give them an order-of-magnitude improvement in performance over GPUs. Nervana become one of the largest deep learning platform exits in history after getting acquired by Intel for over $400M only 18 months after its Series A fundraise. We complemented our investment in Nervana with a recent investment in another deep learning chip company called Mythic, which is building ultra-low power deep learning chips for A.I. at the edge. As we move from a world of cloud computing to edge computing, billions of internet-connected devices will become intelligent because of these powerful chips, and the vision of internet of things will finally become a reality because of A.I. Robots, satellites, cars, rockets, security cameras, and airplanes are all examples of where we will continue to see innovation because of deep learning.

A.I. Market

Investment in A.I. is at an all-time high, and I don’t see it slowing down anytime soon. Last year over $2.8B (see Figure 4) was invested in A.I. companies, and I expect to see increasing numbers for 2017. Moreover, M&A has been enormous in this space because of companies making acquisitions due to fears of disruption, competition for deep learning talent, and entry into new markets enabled by A.I. innovation. As shown in Figure 5, there was an enormous spike last year in A.I. acquisitions, and this year we have seen some huge deals as well, with Delphi buying autonomous driving software company nuTonomy for $400M, and Intel buying Mobileye for $15.3 billion. The A.I. market is still in its infancy, but I expect for us to see increasing amounts of A.I. startups getting venture capital and equally active M&A transactions by both big tech companies like Alphabet, Facebook, Microsoft and Amazon, and non-tech companies as the global economy embraces the A.I. revolution.

Across many domains, from agriculture to autonomous cars, we are building complex A.I. systems that transcend human comprehension. Companies must embrace the A.I. revolution happening now, and understand how A.I. is going to change the nature of their businesses by enabling the creation of new A.I.- enabled products. The cognitive level of work will move upwards in their organizations as more menial tasks become automated, and business operational efficiencies will increase as companies embrace predictive analytics powered by A.I. Although there has been incredible progress across verticals like agriculture, automotive, and aerospace in leveraging A.I., we are still only at the beginning of this revolution.