By Alan Zeichick
When the little wireless speaker in your kitchen acts on your request to add chocolate milk to your shopping list, there’s artificial intelligence (AI) working, in the cloud, to understand your speech, determine what you want to do, and carry out the instruction.
When you send a text message to your HR department explaining that you woke up with a vision-blurring migraine, an AI-powered chatbot knows how to update your status to “out of the office” and notify your manager about the sick day.
When hackers attempt to systematically break into the corporate computer network over a period of weeks, AI sees the subtle patterns in historical log data, recognizes outliers in the packet traffic, raises the alarm, and recommends appropriate countermeasures.
AI is nearly everywhere in today’s society. Sometimes it’s fairly obvious (as with a chatbot), and sometimes AI is hidden under the covers (as with network security monitors). It’s a virtuous cycle: modern cloud computing and algorithms make AI a fast, efficient, and inexpensive approach to problem-solving. Developers discover those cloud services and algorithms and imagine new ways to incorporate the latest AI functionality into their software. Businesses see the value of those advances (even if they don’t know that AI is involved), and everyone benefits. And quickly, the next wave of emerging technology accelerates the cycle again.
AI Drives Chatbots to Enhance the User Experience
AI can improve the user experience, such as when deciphering spoken or written communications or inferring actions based on patterns of past behavior. AI techniques are excellent at pattern-matching, making it easier for machines to accurately decipher human languages using context. One characteristic of several AI algorithms is flexibility in handling imprecise data: human text. Specifically chatbots, where humans can type messages on their phones and AI-driven software can understand what they say and carry on a conversation, can provide desired information or take appropriate actions.
In many ways, chatbots are at the forefront of a revolution in creating more-natural human interfaces with machines, explains Suhas Uliyar, vice president of mobile, bot, and AI strategy and product management at Oracle. He explains that there are three trends toward message-based conversation. First, messaging channels such as Facebook Messenger and WhatsApp are familiar to users, so users don’t need to download and learn new apps. Next, users can engage using the language of their choice. And finally, a well-programmed chatbot can respond instantly.
The increase in access to data over the last decade means that machine learning has more to learn from, so it can be more accurate and more deterministic.”–Suhas Uliyar, Vice President of Mobile, Bot, and AI Strategy and Product Management, Oracle
For enterprises to scale with the volume of interactions that comes with such messaging channels, AI is critical for processing these natural-language conversations and returning the appropriate response.
This doesn’t mean that the technology is simple, says Uliyar. It takes a lot of processing power, as well as a lot of data, to compute the sophisticated AI algorithms behind text recognition, including determining context, retrieving information, and, of course, formulating the best reply—all in a fraction of a second.
The challenge is compounded when a user sends the chatbot informal language, slang, tones (emotions), or misspellings. Still, the development of powerful cloud services with impressive computational ability, as well as huge quantities of—and improvements in—text algorithms, has brought the industry to the tipping point of making chatbots ready for pervasive deployment, says Uliyar. The compute power has increased enough to handle the sophistication of the machine learning algorithms behind AI technologies, he says, and “the increase in access to data over the last decade means that machine learning has more to learn from, so it can be more accurate and more deterministic.”
That’s why responsive chatbots are ready for prime time. “With a chatbot, you can apply things like your customer service use cases to it, or marketing or transactional use cases. In just the same way that you enabled mobile interfaces, you can enable these chatbot interfaces,” Uliyar says.
Oracle customers can develop and deploy chatbots using Oracle Mobile Cloud Enterprise. With Oracle Intelligent Bots technology, a key feature of the mobile cloud offering, chatbots can communicate with users on messaging channels such as Facebook Messenger, WeChat, Slack, Apple Siri, Amazon Echo, Google Home, or even a chat widget on a business’s website—or customers can extend an existing mobile app with chat or voice capabilities.
Behind the scenes, chatbots can be trained for natural-language processing, via algorithms that use deep learning neural networks and supervised and unsupervised machine learning, to interact and respond based on an organization’s specific context (such as product offerings or services). The result: the chatbots understand the user’s intent and provide an intelligent response or the appropriate action in real time, 24/7.
Oracle Mobile Cloud Enterprise also uses AI to extract all the necessary information from the user’s natural-language input to shape the payload for integrating with the sources of data to respond to the user. For example, when a user asks the banking bot to transfer US$5,000 from account A to account B, AI not only recognizes the user’s intent to transfer funds but also extracts relevant information such as the From and To account, the currency, and the amount in order to construct the necessary transactional statement to execute in the banking application. AI algorithms continue to evolve with support for additional cognitive capabilities such as image processing and real-time video processing, further increasing the possibilities of an AI-powered conversational user interface.
AI Thrives Buried Deep in the Code
Chatbots are only the tip of the iceberg when it comes to AI and machine learning. Beneath the surface of Oracle Human Capital Management Cloud (Oracle HCM Cloud), for example, AI empowers many hidden functions for human resources, including making recommendations for guiding employees’ careers, advising on work schedules that take childcare into account, and evaluating performance reviews.
Similarly, AI algorithms appear in everything from image recognition (using neural networks) and medical diagnostics (using expert systems) to data analysis (using machine learning) and beyond. AI is hard at work in advanced search engines—even in so-called “recommendation engines” that learn about customers’ buying patterns and offer suggestions about other products or services they might wish to purchase.
Consider cybersecurity, where machine learning algorithms can pore through terabytes of log data from firewalls, application servers, websites, and more and spot outliers. Those outliers could be very subtle patterns indicating that a hacker has breached a system. Alternatively, those patterns might indicate that a server or mobile device is running out-of-date software, or that patches haven’t been made, or even that the hardware might be getting ready to fail.
Where’s the AI? It’s hidden. But it has a big impact nonetheless.
Retail is a practical example of applied AI, says Amit Zavery, senior vice president of Oracle Cloud Platform, citing a scenario in which a customer might upload a photo of a particular dress and then the ecommerce platform might recommend the closest available matches from that retailer. A demonstration application leverages image recognition, big data, and machine learning to present the best choices to the shopper nearly instantly.
Machine learning has been in the Oracle Database, management, and security products for many years. We have a lot of expertise using AI inside products.”–Amit Zavery, Senior Vice President, Oracle Cloud Platform, Oracle
“It’s so much less frustrating than having customers browse through a large catalog,” Zavery explains. “It doesn’t matter if the dress is from a movie or a magazine or wherever; the AI looks through the inventory and suggests matches.” The results: delighted customers who get the dress they want, and happy retailers who make the sale and possibly earn future business as well.
The Tech Is Coming Together
AI isn’t new. It’s not even close to new. For decades, computer science researchers and industry developers have been designing and constructing software that uses AI techniques to solve problems in medical diagnostics, factory automation, pharmaceutical research, text analysis, stock picking, and even playing games such as chess and Go. However, the results were rarely scalable to massive deployments, leaving AI out of the mainstream.
Things have changed, with several breakthroughs in microprocessors, cloud computing, databases, and algorithms coming together.
GPUs. In terms of microprocessors, graphics processing units (GPUs)—chips initially designed for playing computer games and doing 3D modeling—excel at the complex matrix-oriented math required for neural networks, machine learning, and image processing. In fact, GPUs are significantly more efficient, by orders of magnitude, than general-purpose central processing units (CPUs). Today, GPU-based servers form the underpinning of modern AI processing systems.
Cloud technology. Cloud computing has brought massive amounts of processing power, storage, and bandwidth to enterprise developers. With special cloud computing services built with GPU-based servers, high-powered AI has suddenly become accessible both for developers to experiment with and for corporations to deploy as complete applications. Helping to drive both proof-of-concept projects and line-of-business initiatives for Oracle customers are the capabilities built into the new Oracle Artificial Intelligence Platform Cloud Service (Oracle AI Platform Cloud Service) running on GPU-based servers.
Algorithms. Driving the intelligence behind AI are complex algorithms for processing tremendous quantities of structured and unstructured data. Popular algorithms, such as the Caffe deep learning framework, Jupyter Notebook data cleansing and modeling library, Keras neural networks API, NumPy high-performance array processing library, scikit-learn data mining and machine learning library, and TensorFlow library for machine intelligence, make AI accessible to enterprise developers.
Thanks to these libraries, coders and data scientists no longer need a PhD in applied AI to make heads or tails of advanced techniques. Nobody should pretend that AI is easy to code, but it’s becoming easier for enterprises to utilize, thanks to platforms such as Oracle AI Platform Cloud Service and its support for popular algorithms and frameworks.
And then, of course, there is the data itself. The emergence of the internet sparked a huge increase in the amount of digital information being generated, stored, and made available for analysis—in fact, 90 percent of the world’s data was created in the last two years. Data is fodder for statistical analysis, which is the key for higher accuracy in machine learning models.
Where Is AI in Oracle Products? Everywhere.
Oracle has long used AI in its own products, says Zavery, and thus is no newcomer to the field. In fact, AI is pervasive inside Oracle products and services. “Machine learning has been in the Oracle Database, management, and security products for many years,” Zavery says. “We have a lot of expertise using AI inside products.”
Oracle Management Cloud leverages AI to help detect anomalies in very large datasets, such as to indicate problems on a manufacturing line.
Oracle Cloud Access Security Broker Cloud Service uses machine learning to drive its threat-protection and data-protection capabilities.
Similarly, Oracle Analytics Cloud services use natural-language understanding and processing to let business users ask a computer sophisticated questions and get answers in English or the user’s native language—or displayed as charts, graphs, or other meaningful formats.
That’s only the beginning. Oracle Adaptive Intelligent Apps, built into Oracle Cloud applications, bring AI into familiar business domains. Oracle Adaptive Intelligent Apps are embedded into Oracle Enterprise Resource Planning Cloud (Oracle ERP Cloud), Oracle HCM Cloud, Oracle Supply Chain Management Cloud (Oracle SCM Cloud), and Oracle Customer Experience Cloud (Oracle CX Cloud). In addition, the recently announced Oracle Autonomous Database Cloud uses AI to eliminate complexity, human error, and labor-intensive manual database administration.
Build Your Own AI with Oracle AI Platform Cloud Service
Announced in October 2017, Oracle AI Platform Cloud Service helps enterprise developers build new AI applications—and also embed AI functionality into existing applications to make those applications faster, more feature-rich, and more valuable for the business.
There are so many frameworks, and users don’t know which is best or [don’t] want to maintain all of them. We give developers a choice of all the open frameworks in Oracle AI Platform Cloud Service.”–Amit Zavery, Senior Vice President, Oracle Cloud Platform, Oracle
Before Oracle announced Oracle AI Platform Cloud Service publicly, Oracle used that service internally, explains Zavery. “We have been using many of the machine learning frameworks ourselves,” he says. “We started using it first, and now we are making it available to customers.”
Oracle AI Platform Cloud Service is very flexible, Zavery adds, because it includes many built-in algorithms and frameworks, such as Caffe, TensorFlow, and DL4J, and runs directly on existing Spark/Hadoop clusters for big data computation and analysis.
Because developers and data scientists often don’t know which AI models or algorithms are best for a specific application and it’s cumbersome or impossible to experiment with different algorithms if a vendor’s AI platform supports only a single model, this multialgorithm approach is key.
“There are so many frameworks, and users don’t know which is best or [don’t] want to maintain all of them,” Zavery says. “We give developers a choice of all the open frameworks in Oracle AI Platform Cloud Service. We give choice to customers. Oracle maintains and manages the various libraries and keeps them up to date, so developers can focus on building their application and prototyping their models. They can augment their data with other third-party data and Oracle-provided datasets.”
It’s much easier to use Oracle’s toolkit to set up and test four open AI frameworks against the same data, he explains, “rather than setting up and testing four frameworks using services from four different vendors.” Not only that, he adds, but Oracle is committed to developing and testing all popular machine learning algorithms within the cloud service, so the functionality will continue to grow over time.
AI Is Pervasive in Business If you think AI is everywhere today, expect more tomorrow. Oracle’s AI-enhanced software-as-a-service and platform-as-a-service products will continue to incorporate additional AI to help make cloud-delivered and on-premises services more reliable, more performant, and more secure. AI-driven chatbots will find their way into new, innovative applications, and speech-based systems will continue to get smarter. AI will handle larger and larger datasets and emerge in increasingly diverse industries.
Sometimes you’ll see the AI and know that you’re talking to a bot. Sometimes the AI will be totally hidden, as you marvel at the uncanny intelligence of the software, websites, and even the Internet of Things. If you don’t believe me, ask a chatbot.
LEARN more about AI at Oracle.
Illustration by iStock.com/Instants