Artificial Intelligence. I can’t think of a reference to the intellect that would feel more unauthentic and fake. It’s no wonder people turn to The Terminator or The Matrix to fathom what it’s all about. Fortunately, after 60 years of AI rumors fueled by academia and movies, we’re finally starting to see signs that it means more than just robots taking over.

Working in the tech industry, it’s ironic that the AI lightbulb clicked not through an understanding of machine learning or engineering — but through the human challenges we face in the technology world.

See also: Look at all the amazing things AI can do for lawyers

While at Yahoo! and Apple, it was amazing to be part of technologies that not only helped enable the modern cloud today but continue to support its more than doubling in performance every year. But on the people side, opening a new data center meant hiring a team with over 100 man-years of various expertise and experience. As these skills became harder and harder to find, we realized that since we couldn’t find more people that had them, we needed to figure out how to get more out of people that didn’t.

We’re living in a world where the evolution of technology is exponentially outpacing that of people. AI is now empowering machines with the intelligence to form its own insights and thoughts. What were sensors are now evolving into senses, where machines can acutely comprehend things like sight, sound, and touch.

The Internet of Things (IoT) is enabling “self-driving” actions to be performed, based on those thoughts and senses. Right in front of our eyes, technology is making the leap from being the inanimate tools of yesterday to serving as our collaborative and conscious AI-powered co-workers of tomorrow. As people, our responsibility is to serve as the manager and mentor of this new form of co-worker.

Using comprehensive human-to-machine learning interfaces, all people will have the ability to effectively teach and mentor our new co-workers using one’s own natural human language. Yes, human knowledge will be the foundation of what makes artificial intelligence real. This is why AI will prove to be more useful as an extension of human intelligence, than a substitution for it. It is then that it will no longer be viewed as Artificial Intelligence, but understood as Augmented Intelligence.

Enter augmented intelligence

AI is going to augment natural human intelligence and enable people to gain the world’s collective expertise while requiring less time and study than what has been required to become an expert in any one thing today. Traditionally in humans, an expert’s mind possesses fewer possibilities for slower growth, while a beginners mind offers many possibilities for rapid growth.

Augmented Intelligence will empower us with the best of both. In doing so, there is no reason to think that our own personal capabilities for intellectual advancement cannot equal or surpass the doubling per year that we see in computers. Yes, we will see the day when the augmented human expert will be able to get 2x smarter, every year.

Being a leader of people, I learned long ago that my own success is more defined by the augmented intelligence and capabilities that my team provides than anything I could solely do as an individual. The same will apply to the new performance enhancing co-workers that will augment our intelligence and capabilities. Through today’s eye, this picture of our future may be seen as heresy or a superhuman threat. That said, the same fear occurred in the past when it was suggested that all people could be given the power of reading.

Evolution is life. Embrace the future.

We’ve all been inundated with the hype surrounding Internet of Things “smart stuff” and the impending arrival of our robot overlords, so we tend to minimize the mind-blowing wonder of the responsive and intelligent computing metamorphosis that is upon us.

For years, the IoT community has been saying that if we really want “things” to be of value, they cannot be dumb. The first wave was getting everything connected, and we have made headway there. The next step is to actually make “things” smarter. 

There are a variety of commercial solutions that do not really deliver on the promise of automating our way to more productive lives. And the concerns over properly securing our connected things still weigh heavily. But there really have been transformative leaps in computing capability and achievable functionality. The killer use case for IoT is on the horizon, but before defining what that is and describing how it is going to manifest, I think it’s important to broadly identify how we got here.

The “Trinity”

The impact of the open source movement in driving exponential leaps in technological advancement cannot be minimized. The algorithms and computing infrastructure that drive “smart” things — IoT, Artificial Intelligence, and machine learning capabilities — have been around for decades. Anyone at the NSA can tell you as much.

The difference now is in accessibility to the masses. These technologies were once jealously guarded, closed off from the wider world, and only available within formidable institutions possessing vast resources in both personnel and compute power. Open source changed all that. New things no longer have to be constructed from ground zero, thus supercharging the innovation cycle. The widespread access to knowledge bases and software allows anyone so inclined to build upon the shoulders of giants and leverage the wisdom of crowds.

The creative explosion fueled by open source helped give rise to the cloud, which is the second movement responsible for ushering in our new era of computing. Freed from the physical limitations and expense of individual server stacks and on-premise storage, the “app for everything” age dawned and the capacity for on-demand collection and consumption of big data was unleashed. Once we could scale compute power unconstrained by geography, our technology became mobile and the dream of smaller and increasingly powerful devices trafficking in colossal quantities of information became a reality.

Big data gives lifeblood to modern computing. But data does not do anything and, in itself, has no value. This brings us to the third movement in the “smart” revolution: analytics. The types of augmented computing that people encounter in everyday life now — voice recognition, image recognition, self-driving and driver-assisting cars — are founded in concepts that rose out of analytics and the pursuit of predictive analytics models, which was all the rage just a few short years ago.

The disheartening realization with predictive analytics was that, to train effective models, you need both massive amounts of data and scores of data scientists to continually build and maintain and improve data models. We were once again running up against the roadblocks of access and resource constraint.

And so we arrive at the present, where things are shifting in a new direction. The difference now is that we do not need to recruit an army of data scientists to build models; we have taught our programs to remove some of those roadblocks for themselves.

Inherent intelligence

Our AI-driven systems, especially Deep Learning systems, can now be fed millions upon millions of training sets, train in days/hours, and continuously re-train as more data becomes available. Open source tools and cloud computing are still important and evolving, and we still traffic in loads of data to perform lightning-fast analysis, but our programs now incorporate AI as the engine to make themselves “smarter.”

Expertise from vastly different computing realms has congealed to imbue programs with previously unimagined capabilities. The paradox is that as the cloud becomes ever more powerful and less expensive, the smart IoT strategy is to move much of the first line of entry processing away from the cloud and to the edge. This serves two purposes: to enable on-device decisions without needing cloud intervention and to deliver edge patterns and analytics to the cloud for fast second-stage analytics. Tiny AI engines can now perform analysis in near real time on edge devices and “things” no larger than a matchbook. And as these points of computational power grow increasingly commonplace in ordinary objects — intelligent routers and gateways, autonomous vehicles, real-time medical monitoring devices — their potential functionality expands exponentially.

John Crupi, Vice President and Engineering System Architect, Greenwave Systems

Artificial intelligence at the edge

In the early days of IoT (aka M2M), the focus was on getting data up to the cloud when possible. FTPing log files every night was the rage. When General Electric came on the scene with the “industrial internet,” everyone began talking about real-time live data connectivity. That was a big jump from FTP, but people treated edge devices as simply “things” that transferred data to the cloud for analytics. We are now in the midst of an exponential reverse fan out of that thinking. Real-time requirements are redefining the paradigm. The cloud is now shifting into the role of IoT support and second-tier analytics, and the processing is getting pushed out to the edge.

For example, we have been working with a company developing a next-generation medical monitoring device. Initially, we assumed with such a small device, we would send raw data from the device to the cloud for analysis. But that is not what was desired, nor is it what transpired. The company wanted the analytics on the monitor. They wanted the analytics and pattern detection to occur directly on the device, to take actions on the device, and for only “intelligent” (as opposed to raw) data be sent up to the cloud. The model differed dramatically from standard industrial M2M operations — where everything would be connected, and batches of data coming in from all sources would be collected and processed on some set timeline at some central repository.

The whole purpose of connecting now is to obtain instantaneous precision results at the point of entry for immediate answers. Even the low latency involved in “traditional” cloud-processing with hundreds of thousands if not millions and billions of devices is not as efficient for real-time edge analytics as using this new architecture. In some cases, you can achieve a data reduction of 1,000x by just sending the analytics and patterns vs. raw data to the cloud. 

We no longer deal in dumb collection devices; we need them to do more than just curate. They must be artificially (and naturally) intelligent — capable of doing pattern recognition and analytics in their tiny engines. They push those results up to the cloud for other uses. As this ideal proliferates, so, too, do the possible applications.

As is perfectly embodied in the example of an autonomous car, this dual edge/cloud analytics model produces precision, real-time results that can be continually and automatically refined against ever-growing troves of more data, thus producing valuable, useable information and powering productive action. Even a year ago, I would have called B.S. on this notion for widespread IoT and AI integration — but edge computing and AI have really broken out of the lab and into our world. It will yield outcomes we have never seen before.

The killer use cases for IoT are manifesting through truly intelligent edge devices — in solutions that are purpose-built for specific problems or tasks, then interconnected and subjected to patterns that move beyond their initial application. As more and more smarter, AI-enabled “things” are incorporated into our everyday lives and operate at the edges of our inter-communicating networks, we will see things moving beyond merely being connected and into actively embodying intelligence. Smart stuff indeed.

This article is produced in partnership with Greenwave Systems.

The author is Vice President and Engineering System Architect at Greenwave Systems, where he guides development on the edge-based visual analytics and real-time pattern discovery environment AXON Predict. He has over 25 years of experience executing enterprise systems and advanced visual analytics solutions.

From the stage of F8, Joaquin Quinonero, Facebook’s Director of Applied Machine Learning, described a new technique the company is using to improve the watching experience for 360 videos. The format is challenging to deliver because of its size, but Facebook is using machine learning to reduce the number of pixels that have to be rendered at any one time. By predicting where a viewer will look next, rendering priority can be given to that location  — particularly helpful for users with lower quality internet access.

The status quo for 360 videos is reactive rather than proactive rendering. Mike Coward, engineering director for Facebook’s VR video team echoed the frustration of users to me when he described the unpleasantness of turning your head in VR only to see a blurry scene.

One partial fix is to optimize compression. But teams at the company are already using machine learning to select across the thousand-plus compression techniques for individual snippets of video. The other way to reduce the streaming load is to just cut down on what you’re rendering. And rather than reduce quality across the board, Facebook’s approach improves resolution for exactly what you’re most likely to look at next.

Mike Coward, engineering director for Facebook’s VR video teamStep one was to use the resources of the company to monitor where people actually do look when watching 360 videos. Facebook’s VR video team created a heat-map that highlighted the most popular spots that users looked at within videos. From there, Facebook built a generative saliency map using a deep neural network. This model makes it possible to perform predictions on new videos that haven’t previously been watched or studied.

If a human were to be given the task of predicting where someone might look, they might study their natural environment and look for anomalies that could catch one’s interest — think birds or a car driving by.

Abstracting away to the neural net, the physical cars and birds cease to matter. Facebook’s model was trained on a massive corpus of videos to identify interesting subsets of a video frame. Coward told me that the model, when faced with a surfer in the ocean, is capable of picking selecting the surfer as most interesting, despite the fact that both are fast moving entities.

After implementing the prediction model, Facebook was able to increase resolution by 39 percent on VR devices. Aside from improving resolution and making 360 videos accessible to people without great network connections, the technology could some day make it possible to offer preemptive suggestions to creators on how to make videos more engaging.

Baidu is opening its self-driving vehicle platform in a bid to help drive the development of autonomous cars.

The Chinese internet giant today announced its Apollo project that will see its platform, including vehicle platform, hardware platform, software platform and cloud data services, opened to help others in the industry, particularly car manufacturers, to develop autonomous vehicles.

The initial target is to open the technologies up for vehicles in restricted environments this July. Baidu said it then plans to share technology for simple urban road conditions before the end of the year, with the ultimate goal of opening its full tech stack — covering fully autonomous driving capabilities on highways and open city roads — by 2020.

“AI has great potential to drive social development, and one of AI’s biggest opportunities is intelligent vehicles,” Qi Lu, the former Microsoft exec who recently became Baidu group president and COO, said in a statement.

Beyond offering up its platform and technologies, which the company has invested significant sums into, Baidu said it is also looking to add partners to the program to strengthen it, particularly around compatible vehicles, sensors, and other components. Baidu has partnerships with Chinese companies such as BAIC Motor, BYD and Chery, while a two-year relationship with BMW petered out last year over apparent differences in strategy.

This move to open source much of its self-driving tech seems like a move to gain a leg-up on more developed competitors such as Google and Tesla.

Baidu was one of the first major tech companies to embrace artificial intelligence and machine learning, and its autonomous vehicle push began with road testing in Beijing in 2015. Last November, it offered test rides to attendees of the World Internet Conference in Wuzhen, and the Chinese company also has a permit to test in California, which is where its research labs — including its AI division — is based. Baidu recently lost the head of that project, renowned AI expert Andrew Ng, after he announced the end of his three year stay at the company, but its AI group nevertheless employs around 1,300 people, with 300 of those in the Baidu Research division. That makes it one of the largest units of its kind in tech.

Today Facebook open sourced Caffe2. The deep learning framework follows in the steps of the original Caffe, a project started at the University of California, Berkeley. Caffe2 offers developers greater flexibility for building high-performance products that deploy efficiently.

This isn’t the first time that Facebook has engaged with the Caffe community. Back in October, Facebook announced Caffe2Go, what effectively was a mobile CPU and GPU optimized version of Caffe2 (they even both have Caffe2 in their names if you parse it right). Caffe2Go received attention at that time because its release coincided with Style Transfer.

Notably, the company also released extensions to the original Caffe. The majority of these changes make Caffe more attractive to developers building services for large audiences. For projects where resources are of no consequence, Facebook has historically turned to Torch — a library it finds optimal for research use cases.

Every tech company wants to tout the scalability of its machine learning framework of choice. I asked Yangqing Jia, the lead author on Caffe2, what he thought of MXNet and the noise Amazon has been making about its ability to scale. Reasonably, he was cautious about dropping benchmarking numbers for comparison. These numbers can have meaning, but they are heavily influenced by the actual implementation of a machine learning model and subject to a fair amount of “DIY” volatility.

Yangqing Jia, the lead author on Caffe2 and Alex Yu, leader of business development

“All frameworks are more or less at a similar scalability factor,” explained Jia. “We’re pretty confident that Caffe2 is probably a little bit better than the rest.”

Facebook is pouring a lot of resources into both Caffe2 and PyTorch. Today’s release accompanies partnerships at the hardware, device and cloud levels. Alex Yu, leader of business development for Caffe2, explained to me that Facebook aimed to include the market leaders in each category. This meant Nvidia and Intel on the hardware side, Qualcomm on the device side and Amazon and Microsoft on the cloud side. And while Google wasn’t targeted, a GCP partnership wouldn’t be out of the question going forward.

Prior to release, Caffe2 was deployed at scale across Facebook. The team also took considerations for the developer communities familiar with the original Caffe. Caffe models can be easily converted to Caffe2 models with a utility script. Facebook is releasing documentation and tutorials and has put Caffe2’s source code on GitHub.