GUEST: To borrow a punch line from Duke professor Dan Ariely, artificial intelligence is like teenage sex: “Everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it.” Even though AI systems can now learn a game and beat champions within hours, they are hard to apply t…Read More
Google released its Android Security 2017 Year in Review report today, the fourth installment of the company’s attempt to educate the public about Android’s various layers of security and its failings. One of the most interesting learnings to come out of the report is that 60.3 percent of Potentially Harmful Apps (PHAs) were detected vi…Read More
Pokémon Go took the world by storm in July 2016 when it debuted with game characters embedded in real-world maps. Now Google is making it easy for any game developer to do the same thing with Google Maps through the Unity game engine. With the new Google Maps application programming interface (API), it will be much simpler to create games based on…Read More
South by Southwest (SXSW) is a melting pot of techies and artists showcasing their product, band, or movie to more than 150,000 participants in Texas’ buzzing capital city of Austin. So it’s the perfect place to announce a project like The Music Fund. The team behind it — Geoff Cross, Thomas Jerde, and Nick Smith — are curre…Read More
OPINION: We’re nearing the end of the beginning for the battle among virtual assistants in the U.S. With the launch of the HomePod, Apple joins Google, Amazon, and Microsoft in having a smart speaker available at all times to answer queries and perform actions. But the market for intelligent virtual assistants is far more complex than just determin…Read More
Machine learning is making substantial impacts on businesses around the world, but many organizations struggle to understand where and when to optimally use ML. To enable successful deployments, businesses must first recognize which problems are most amenable to ML and, second, ensure the right processes are in place to evaluate its impact.
In general, ML algorithms build relationships between inputs and outputs by leveraging statistical properties of the data. As researchers expose the algorithm to more data, it learns and adapts. Eventually, the relationship becomes accurate enough that the algorithm generalizes to predict outputs from new inputs. Companies can use these predictions to uncover new insights and power business automation.
Above: Vertical sample use case
Finding problems ML can solve
So how can you identify a business goal you can address with ML? Here is a typical set of steps to consider.
1. Is the goal substantive, quantifiable, and measurable?
First, the goal you are looking to achieve using ML should be meaningful to your business. Identifying substantive goals typically requires engaging business owners who have a broad understanding of the value generated by an ML solution.
Next, the prediction goal should be quantifiable and well-defined. For example, one of the most common types of ML frameworks is supervised machine learning. In supervised machine learning, researchers give the algorithm an input ‘X and an output Y, and ask it to find the functional mapping Y=F(X) between the X and Y.
If you’re looking to maximize engagement on your site, the target Y that you optimize might correspond to onsite click-through rate, total time on site, or a combination thereof. A supervised ML problem requires the business owner to explicitly quantify what they’re optimizing the function for.
Finally, the output of the ML algorithms should be measurable on an ongoing basis. The best ML algorithms adapt as researchers expose them to new data in order to minimize the error rate of predictions. This enables the system to continually learn and adjust its algorithms to optimize business goals.
2. Is machine learning the right approach?
Broadly speaking, businesses can approach developing intelligent systems (AI systems) in two ways: expert systems and ML. In expert systems, humans explicitly program actions often based on “if this, then that” rules. Such systems, in general, don’t have the same data requirements as ML algorithms, and they benefit from the developer having a more explicit understanding of the final algorithm. ML, on the other hand, uses data to learn rules. For nuanced problems whose solutions analysts cannot encode in rigid rules, machine learning can uncover relationships that expert systems may miss. But this flexibility comes at a cost: You need data.
Above: Expert systems vs. machine learning: Medical diagnoses
3. Do you have the necessary data?
If ML is a viable path to solving your business problem, then data is required. Both quantity and quality are important. Historical data establishes a reliable input and output relationship to train the model. Beyond the model-training phase, infrastructure is typically needed to collect new data from which to learn over time.
Evaluating machine learning in your business
After a company identifies an ML project, it is important to evaluate the broader impact of ML on the business.
An active area of ML research focuses on interpreting why an algorithm derived the output that it did, and researchers can decompose many algorithms to provide these insights. For instance, knowing that a user will churn is valuable, but knowing why that user churns allows businesses to enhance their products or develop automated mechanisms to prevent them from churning. Ensuring that your business is actively engaged in the “why” is important to optimizing the broader impact of ML on your organization.
It’s important to measure how the ML algorithm affects broader business goals. For example, a Facebook feed focused on maximizing engagement might initially focus on maximizing content clicks. This ML goal may have unintended consequences, namely adding click-bait to a content feed, ultimately leading to a decrease in CLV across the user base. Over time, it may be necessary to modify the engagement goals to a quantifiable metric associated with long-term customer retention.
A recent survey of 1,000 business leaders found that while 66 percent of organizations use AI to automate routine tasks, 80 percent of C-level executives indicated that the future of their businesses “will be informed through opportunities made available with AI technology.” It is crucial that business owners are able to effectively identify and measure the impact of ML in order to maximize the business opportunities the technology brings.
Alex Holub is the cofounder and CEO of Vidora, a real-time machine learning platform used to optimize marketing and product automation.
Densify, a company that helps enterprises make sure that they’re using their compute resources to the fullest extent possible, announced a new service today that uses AI to cut down customers’ cloud bills. The Cloud Learning Optimization Engine (Cloe for short) analyzes workloads using machine learning to determine how much CPU, RAM, and storage they need, then suggests ways to save money.
Cloe has helped customers like Bank of America, Honda, and IBM save an average of 40 percent on their cloud bills, with some customers seeing savings of more than 80 percent. After analyzing the needs of each workload, it suggests compute instances companies can shut down, workloads that can sit on the same instance to save money, and ways to optimize which types of virtual machines customers use in order to reduce their spend.
As more companies move their workloads to the public cloud, the sort of optimization work that Cloe helps with is an important component of ensuring that businesses aren’t spending too much on their computing infrastructure. Densify CEO Gerry Smith said that the company’s product was aimed at solving the core problem in infrastructure optimization: Customers don’t know what resources their applications actually need to perform well. That can be costly when they’re paying by the hour for compute capacity their applications will never use.
The service works across Amazon Web Services, Microsoft Azure, Google Cloud Platform, and IBM’s cloud offerings, so customers can see which environments can handle their workloads most efficiently. Cloe has a normalized understanding of each platform’s offering, so it’s possible for customers to compare how different providers stack up.
Customers can connect Cloe to their cloud instances, and the service can use the already-available log data (like AWS’ CloudTrail) to begin optimizing right out of the gate.
“Usually within two to four weeks, you’ve got 50 percent of the savings,” Smith said in an interview with VentureBeat. “Depending on where the savings are, within another two to four months, [you’ll get] 100 percent of the savings.”
After that, Cloe will continue to review cloud providers’ pricing changes, applications’ needs, and new products to find where customers can save money further.
For companies that are looking to move their workloads to the public cloud, Densify’s technology can analyze the contents of virtual machines running in private datacenters and suggest the best cloud environments for each workload.
Virtualitics, the pioneering startup that merges AI and VR/AR with big data, recently announced the close of its $7 million Series B. The round was led by by Centricus, a global investment platform, and with participation by existing investor the Venture Reality Fund (‘The VR Fund’) and other private investors. This follows a $4.4 million Series B that was closed only last April, which was led by The VR Fund.
“The reason for the short time between A and B was the strong market response, interest and adoption, which attracted new investors to lead the new round to provide acceleration of the company’s original growth model.” Marco DeMiroz, general partner at The VR Fund, told me. “We are fortunate to have Centricus as a new lead since they have an extensive global presence to help the company.”
Early wins for the early stage
Why is this a big deal? Last year, we saw early-stage VR/AR investments fall, even if the spikes in Q2 and Q4 among a very few mammoth-sized deals in specific investment categories made it appear like it was an improvement from 2016. It’s an important, positive, and well-timed indicator that a VR startup with as strong fundamentals as Virtualitics raises a healthy new round that is triggered not purely by the need for funding in order to survive, but because they’ve effectively struck oil and need to scale in order to satisfy market demand. To have it happen now also helps to encourage and strengthen investor confidence at an important time of the year.
This last round brings Virtualitics’ total funding to date to an excess of $11 million to supporting their vision of leveraging emerging technologies in order to enable and unlock the potential of big data in ways that are otherwise hindered or altogether impossible with simple 2D tools and platforms, including the 3D tools that are only represented on 2D panes and windows. Indeed, VR/AR-driven reporting powered by machine learning promises to do what has defied the reach and grasp of all of its predecessors.
“Turning big and complex data into useful insights requires new ways to analyze and interact with it. We’ve solved for this by coupling AI with immersive environments.” Ciro Donalek, CTO and co-founder, wrote in the press release. “Moreover, business intelligence platforms need to be 3D and collaborative by design in order to help companies gain a deeper level of understanding in the stories being told by the raw data. This is how we’re approaching big data and helping evolve the next generation of data analytics and intelligence platforms.”
Above: Virtualitics’ reporting tools in motion
The kind of interactive data visualizations that can be presented and digested for human consumption as an interactive medium that you can literally step into and manipulate in either a calculating or intuitive way is why Virtualitics has been able to consistently reel in, retain, and grow its relationships with several major global Fortune 500 businesses across multiple industry sectors, including consumer goods, healthcare, energy and finance.
Donalek shared with me some of the interesting and varied kind of use cases that their clients have put into play with their immersive solutions:
Extracting insights through the machine learning models to quickly find and visualize relationships among hundreds of variables in order to understand the ones that drive the problem at hand.
To better explore and interact with their data through ad hoc visualizations in 3D and VR.
To let teams collaborate across different geographic locations, interacting and analyzing data together, which is much more effective than exchanging PDFs or screen sharing.
As a storytelling tool to convey results to upper management, decision makers or clients.
While last month I covered the mixed reality analytics platforms that cater for the needs specifically of VR/AR startups, which for their part suffer from a widespread lack of market adoption and traction, it’s clear that Virtualitics’ novel breed of Sci-Fi reporting solutions are much more readily received by traditional industries. And that’s due to the fact that this digital sorcery allows these industries to finally approach and confront all of the myriad of extraordinary challenges that rest inside the scrambled nests of big data sets, which have been all the while lying dormant and can now be stirred awake to deliver bleeding-edge breakthroughs.
A bellwether for the sector
Other startups in this space, like DatavizVR, are welcoming the news as it adds fuel and a sense of validation for their sector as a whole, raising the profile of the technology’s applications in remedying the most pervasive big data challenges faced by the majority of enterprises today. Their WebVR-powered platform, 3Data, allows users to create, collaborate, and present 3D graphs seamlessly to an unlimited amount of viewers on any hardware or any device via the open web, reducing what would normally take hours down to minutes.
Above: 3Data platform by DatavizVR
“The recent raise from Virtualitics helps validate the market for VR/AR B2B enterprise tools like data visualization.” Nick Walusayi, co-founder and CEO of DatavizVR, told me. “We are solving big data problems using cutting edge technology and enterprises need to adopt these solutions to stay competitive.”
This reason above all others may define our fascination with emerging technologies like VR/AR and AI in the first place. Their talent to transcend the barriers that we as humans simply cannot cope with either in scale or scope is simply awe-inspiring. Indeed, these type of collaborative platforms enable data scientists and laymen alike to dig up and surface multidimensional relationships in big data that just cannot be discovered by any other means.
Amir Bozorgzadeh is cofounder and CEO at Virtuleap, the host of the Global WebXR Hackathon and the startup powering up the Gaze-At-Ratio (GAR) XR metric.
Making a video game is a hard endeavor even in the best of circumstances. Developers have to anticipate an audience’s desires years in advance, determine the best way to monetize their creation, and execute on a vision amid a cascading series of strict deadlines. And with so many new gaming formats becoming part of the gaming landscape, such as mobile, AR/VR, esports, and streaming, developers also have to consider the type of game they want to make.
It will never be an easy job, but it doesn’t have to be quite as hard as it is today. Market research has become a critical part of the design process in recent years. And there are more data streams than ever feeding into that research pool. Too few game makers are taking full advantage of the information that’s available to them –and even those that do generally focus solely on the success of a single title, rather than an entire brand.
This is partly because, for years, market research was limited to assembling player focus groups and siloed product data.
Where the problems are
User feedback is problematic because people’s stated opinions don’t always match up to objective facts or their behavior. Take Super Mario Run. When it launched in December 2016, it had over 50 percent negative (1 and 2 star) reviews. By that metric, it was a failure and you might predict poor sales, but it was downloaded over 200 million times! 12 years ago (before mobile games as we know them) those negative reviews might have tanked the whole project, but here we are, a little more than a year after launch, and Super Mario Run is still in the top games charts.
An additional factor for non-data-driven production is likely that studios and publishers had limited product data to work with, and the data they did have, was siloed by teams dedicated to to build different parts of a game: level design, texture mapping, character animations, menu interactions, etc.
The success for each of these pieces was frequently measured independently with only a select few — if any — executives able to see the entire picture. This could lead to gaming experiences where one small issue, say with a menu load time, wound up stacked on top of another small issue where a certain enemy randomly generated too often. Looked at independently, each feature might be within their design specifications, with development teams giving the “green light” to move to production. But, when these issues are experienced together, players found themselves frustrated with constant, time-consuming interruptions leading to a poor game experience, reviews and sales.
Either way it was a long and not a particularly data-driven process. But have no fear — instead, keep reading for some advice on how you can focus on utilizing the right data to increase success.
Taking a look back
By 2010, mobile gaming was impossible to ignore — and suddenly developers and publishers had a lot of concrete information to work with, including user reviews, engagement stats, download totals, and more. And because it cost substantially less to create mobile games than console titles, developers and publishers could rapidly deploy smaller games in short order. From there, they would monitor the feedback and find areas for improvement, then introduce new downloadable content that was tailored to the playing habits of customers.
You don’t have to look far for examples. Rovio’s Angry Birds made its debut in 2009 and quickly became a smash hit. Over the next four years, the game had 14 different versions and got 3 billion downloads worldwide.
Other titles like: Clash of Clans, Candy Crush and Temple Run continued the agile, data-driven approach to game development, parsing user data and releasing new versions weekly, as well as new spin offs that were largely the same as the original, but tweaked enough to entice players to grab them. Many can still be found on the top app charts today.
That new style of game making quickly proved to be a success. Mobile games became the fastest growing segment of the industry. In 2017, revenue from those titles hit or exceeded $40 billion. It was also the beginning of a more data-driven approach to development — one that would become essential for future, costlier development projects.
Hey, developers, use my data!
Delivery platforms have become invaluable partners for game makers, not only as storefronts but also as data providers. Steam, the leading digital distribution platform for PC games, offers players a number of options beyond its retail roots, including multiplayer gaming, video streaming and social networking services. In these offerings, as well as its storefront, Steam collects massive amounts of data, including Steam page traffic sources, visitor behavior, page visits and purchase information, that developers can analyze to develop the right games for the right audiences. Data from Steam, via APIs to the Steam platform, is sent daily, weekly or monthly to the game publisher. It’s not raw event-level data. Rather, it’s aggregated or summarized information that outlines trends, but doesn’t provide deep insights.
VR, AR, and esports
Developers are now trying to apply the lessons of online streaming to AR/VR. Gamemakers aren’t sure what kind of controller works best, or what kind of animation resonates with players. So while there’s no previous technology model to follow, there is a data-driven marketing model.
That model works with esports as well. Beyond capturing gameplay data points, competitive gaming has the added opportunity for measuring community engagement, via observer camera angles, team-chat, peer-to-peer betting and trading, and even fantasy teams. All these new interaction types supply data that can be used to improve the game experience and the viewing experience, and, ultimately, boost revenue.
The inclusion of game viewers is something new and potentially huge. Developers now not only have the opportunity to bring players into their brand, but also a large audience that didn’t download or purchase the game. It’s a situation where marketing models are even more critical.
Challenges with data and data aggregation
The challenge developers and publishers face, though, isn’t access to data, it’s access to the complete data picture. Creating a true brand experience doesn’t come from making standalone successful games, it requires getting players of one title to consider trying another they might normally not be aware of (or not normally be interested in playing). And that’s where data aggregation can really help.
Ninety percent of all of the world’s collective data (both inside and outside of the entertainment industry) was created in the last two years — with an additional 2.5 quintillion bytes generated each day. The sources continue to multiply. That’s means there’s a lot of information for developers to parse — and most of it gets lost. Creating a 360-degree view of each player and understanding their buying behavior across channels can increase long-term loyalty and maximize a company’s ROI.
How does that play in the real world? Consider this hypothetical: Plants vs. Zombies continues to be a phenomenally popular title on mobile and console. And between those two platforms, EA and PopCap Games have oodles of data to mine. But PvZ has grown into an industry, with plushies, action figures, books, board games and even slot machines. Those other categories can highlight, for example, which character in the game is popular or trending with players, That data can be used in determining where to focus DLC and expansion efforts and increase user engagement (which, in turn, makes them more likely to complete in-game purchases).
The same sort of data can be used with the Star Wars franchise, though as EA learned with the recent release of Star Wars Battlefront 2, it’s critical to not use the data in a way that is obviously meant to boost a title’s bottom line. Knowing the characters players love is one thing. Asking them to pay extra to access it is a riskier proposition.
Today’s game development studios, big and small, have access to more data about their customers than any creator ever has. But by failing to tie all of that information together, they’re making their jobs more difficult and sacrificing potential profits. Making confident, informed decisions in a timely manner requires a dynamic, scalable way to acquire and aggregate the fast-moving data streams. Companies that do so will find that the key to future success in the video game industry is having a firm grasp on the habits and preferences of their consumers, even if the consumers aren’t aware of those habits themselves. Those who ignore the opportunities that today’s data-rich environment presents do so at the risk of being left behind.
“Deep learning” has become a hot topic in the general rush to launch AI products. But many of these products will fail because companies are putting branding ahead of functionality. Success depends on understanding what deep learning is, how it works, and what its most effective applications are.
Deep learning 101
Traditional machine learning algorithms are typically linear, in that they can be represented by only one node that linearly transforms input to output. Previously called artificial neural networks or neural networks, deep learning uses multiple such nodes, organized like the neural networks originally invented in 1943 to model how human brains work. The more nodes and layers in a neural network, the more sophisticated its learning capabilities can become. Although people still use the term “neural networks,” today’s deep learning networks represent how information flows across nodes more than how information in the human brain flows across neurons.
Deep learning requires ample data and training time. But while application development has been slow, recent successes in search, advertising, and speech recognition have many companies clamoring to get in on the action.
Mislabeling and overuse
Vendors’ tendency to label almost anything “deep learning” is a recipe for disappointment because the technology is less effective without sufficient data and domain expertise.
A key issue for machine learning algorithms is selection bias. In sound research, you can define the population, have access to all available population data, and sample a portion of that data. With deep learning, you start with sample data, deploy the model, and then expose it to the real world. But models that work well on training data often perform poorly on real data. Deep learning provides the ability to accurately determine the classification function from inputs to an output. However, there is no guarantee that the model will perform accurately on input data from the population if the training data is not representative.
This data failure is more common when training data isn’t developed by domain experts. While deep learning might eliminate the need to have domain experts in the feature extraction part of the classification process, it still requires expertise in the data extraction process. In fact, deep learning might be overkill when a domain expert can explicitly describe the linear or nonlinear function using logic and rules.
For example, if a baker applied deep learning to making bread, a robot’s action, such as telling the automated bread maker to stop kneading, could be more explicitly defined by a domain expert (i.e., a baker) based on input values. In this case, those would be the attributes of the bread dough, like consistency and temperature.
In scenarios such as this, companies that focus on collecting data points might be better served by speaking to an expert. The bottom line is that much of what is marketed as “deep learning” is likely to be ineffective or difficult to manage properly. And “deep reinforcement learning,” as implemented in autonomous robots, self-driving cars, and creation of images, voices, and videos, is far from being widely available. Buying into deep learning hype without doing due diligence could lead to general disillusionment and another AI winter.
Achieving greater accuracy
We may someday reach the point where AI and deep learning will help us achieve superintelligence or even bring on the singularity. But our challenge, and duty, as artificial intelligence professionals today is to ensure that deep learning applications live up to their billing and deliver benefits to users and society.
Dr. Sid J. Reddy is chief scientist at Conversica, a company that provides AI software for marketing and sales.
Read Full Article
Read for later
Articles marked as Favorite are saved for later viewing.
Scroll to Top
Separate tags by commas
To access this feature, please upgrade your account.