It won’t be long before we are up to our virtual necks in AIs and at NVIDIA’s GTC this week IBM stepped up with a new platform called PowerAI which targets AI developers and sort of performs the role of midwife in the birthing of these ever more intelligent thinking systems. IBM—known for its Watson platform—came to the conclusion some time ago that most firms and developers can’t afford something as comprehensive as Watson, and that other lower cost systems that were specifically designed for the massively increasing opportunity for ever smarter systems, would better address this market. PowerAI is one of the solutions that was created as a result.
IBM has been showcasing for some time that its unique, close relationship with NVIDIA has resulted in systems that use GPUs that are uniquely tuned to optimize performance for loads that favor NVIDIA’s technology. The result has been targeted servers that provide up to twice the performance of similar x86 products according to firms like Elinar Oy Ltd that use them.
Let’s talk about some of the core elements.
The Return Of The Midwife
Midwife is an old term popular when and where there was a shortage of doctors. More focused on just helping a mother produce a healthy baby the midwife was involved in assuring the viability of the fetus, assisting with the actual childbirth, and heavily involved in the postpartum process. AIs are a new kind of children and IBM, with this offering, is positioning itself as a new kind of AI-focused “midwife” to ensure the development and birth of the new AIs that are being born all the time.
At the heart of this AI birthing effort is a new software tool called “AI Vision”, which is an application developers can easily use without much training to train—and eventually deploy—computer vision focused offerings. These use deep learning models and this focus on ease of use should substantially reduce the time to market for what they develop.
IBM Spectrum Conductor
A big part of what current generation AI systems do is analyze unstructured data. IBM Spectrum Conductor cluster virtualization software, which is integrated with Apache Spark, is designed to transform both structured and unstructured data into data sets conducive to deep learning, and it is also a part of this offering.
Developed by Google as a deep learning open-source framework, TensorFlow is built into this offering in distributed form and designed to massively reduce the deep learning training time. Current estimates for this reduction suggest a massive improvement—taking the typical training time for a new AI down from one measured in months to one measured in hours.
DL Insight, also included within the framework of this offering, is a new software tool. This tool enables data scientists to analyze their deep learning models and find and correct their qualitative weaknesses. Bad data leads to bad decisions in every segment and this part of the solution is to ensure that the quality of the results doesn’t adversely impact that final ouput.
The Smarter AI Platform
The end result of all of this is a smarter AI platform—one that doesn’t require an excessive amount of training, easily integrates with both structured and unstructured data sets, aids in model development, and ensures that the quality of the result is acceptable to the needs of an enterprise.
What this all amounts to is kind of a lower cost “midwife” for the coming age of deep learning and AI systems. An offering that helps not only husband a new product to term but assists in the birth of the AI and assures that the result will perform in a high-quality fashion. I expect, as a result, PowerAI will be a critical part of many AI development projects and also the cause of our eventually being up to our necks in ever smarter and ever younger intelligent machines. Apparently, IBM is working to become the biggest “midwife” of that next batch of intelligent machines.