Artificial intelligence (AI) and cloud service are big business. Total investment in AI reached $26 billion to $39 billion in 2016 — triple 2013’s numbers, according to a recent survey conducted by Harvard Business Review. And in the next 15 months, roughly 80 percent of all IT budgets will be committed to cloud solutions. That said, despite both industries’ steady growth, just 20 percent of companies say they use one or more AI technologies in a “core way,” and 49 percent say they’re delaying cloud deployments.
IBM blames the dichotomy on the challenges involved in marrying disparate services. In an attempt to address a few of them, the New York company today launched AI OpenScale, a platform that enables customers to build AI on virtually any infrastructure, and Multi-cloud Manager, an “open” solution designed to make it easier to create apps that run on multiple cloud services.
AI OpenScale, which will launch later this year for IBM Cloud and Cloud Private customers, operates through online dashboards and supports AI models developed on a number of open source frameworks, including Google’s TensorFlow, Microsoft’s AzureML, SparkML, Keras, Seldon, and Amazon Web Services’ SageMaker. Additionally, it facilitates the deployment of those models in environments such as IBM’s Watson, Seldon, and other third-party platforms.
That’s just the tip of the iceberg. AI OpenScale offers a suite of autonomous bias detection and mitigation tools, including a logging system that records predictions made by machine learning models, along with the corresponding model version, the training data used, and the associated performance metrics. It continually monitors for prejudicial decision-making in AI applications and, through “de-biasing” technology, makes an effort to mitigate it while providing explanations for recommendations the AI algorithms give.
In that way, AI OpenScale builds on IBM’s earlier work in AI bias detection and model explainability. In September, the company launched an open source toolkit — the AI Fairness 360 toolkit — containing a library of algorithms, code, and tutorials that demonstrate ways to implement bias detection in models. And in a whitepaper this summer, IBM researchers proposed “factsheets” for AI systems that would answer questions about system operation, training data, test setups, results, testing methodologies, and more.
IBM isn’t the only company developing platforms to mitigate algorithmic prejudice, it’s worth noting. At its F8 developer conference in May, Facebook announced Fairness Flow, an automated bias-catching service for data scientists. Microsoft and Accenture have released similar tools.
IBM also today announced Neural Network Synthesis Engine, or NeuNetS, a new system that automates AI development for business tasks and datasets. The company claims that in tests, custom AI models autonomously designed, trained, and deployed by NeuNetS have achieved accuracy “comparable to human-designed neural networks.”
At AI OpenScale’s launch, NeuNetS will be available in beta.
“Our strategy is to use an open, interoperable approach to fuel the AI economy,” said David Kenny, senior vice president of IBM Cognitive Solutions. “We believe AI OpenScale represents a new technology category and the start of a new era in the mass adoption of AI for business because it is open — making any AI much easier to operate and fully transparent.”
IBM Multi-cloud Manager
IBM’s Multi-cloud Manager — which runs on IBM Cloud Private, a platform based on Kubernetes, an open source container orchestration system that automates the deployment, scaling, and management of containerized apps — provides a dashboard interface for juggling up to “thousands” of Kubernetes apps spanning volumes of data in multiple locations.
“With its open-standard approach to managing data and apps across multiple clouds, the IBM Multi-cloud Manager will enable companies to scale their many cloud investments and unleash the full business value of the cloud,” Arvind Krishna, senior vice president of IBM Hybrid Cloud, said in a statement. “In doing so, they will move beyond the productivity economics of renting computing power, to fully leveraging the cloud to invent new business processes and enter new markets.”
Multi-cloud Manager’s dashboard shows Kubernetes clusters on a single, unified control panel, and uses an “integrated compliance and rules engine” to ensure those clusters remain compliant with enterprise policies and security standards. Additionally, it offers backup tools that protect deployed Kubernetes apps and data in case of a security breach or hardware failure.
IBM Multi-cloud Manager, which is already in use by clients like Bendigo and Adelaide Bank, Australia’s fifth largest retail bank, will be available in October 2018.
Its unveiling comes after IBM reported four-quarter revenue of $18.5 billion, up 23 percent from $15.1 billion a year ago.
To Read Our Daily News Updates, Please Visit Inventiva Or Subscribe Our Newsletter & Push.