The world is watching for the next major breakthrough in Artificial Intelligence. The Chinese government has outlined plans to become a world leader by 2030. Russian President Vladimir Putin has predicted that country that emerges as a clear global leader will in turn become “ruler of the world.” In the meantime, companies are vying for dominance in the hardware, software, and sectors where AI and machine learning will live.
Orange Silicon Valley’s recent report, “The New Dawn of AI,” looked at where today’s research dollars are being spent, and our analysts have explored some of the key ethical questions posed by Generative Adversarial Networks (GANs) and robots that are empowered to make their own decisions.
George Manuelpillai, who worked on that report, and Jameson Buffmire, a business analyst at Orange Silicon Valley, offered a few insights from their recent work in a short a interview for this blog. These are the highlights of the AI landscape right now as they see the technology evolving.
Orange Silicon Valley: Software and databases have been around for many years. What distinguishes AI as a new and emerging technology?
George Manuelpillai: AI is nothing but software, and traditional databases are still used to store information. AI is a new and emerging technology because the hardware to parallel process massive amounts of data is now available — through Nvidia’s GPU, Google’s TPU, etc. Using these new microchips, along with historical data and algorithms — modelled after our best guess of how the human brain works — data scientists are able to interpret the world in new ways.
Jameson Buffmire: What we call “AI” today is a way way of using existing technologies in a fundamentally new way. A way that appears to replicate human capabilities, and even human thought. Getting here took decades of research, but most trace the recent advances to a watershed moment in 2012. That year, Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton created a large, deep convolutional neural network and used it to win the Large Scale Visual Recognition Competition (LSVRC). They combined a very large data-set (1.2 million images), with GPU-scale computational resources (two GPU cores), and several new methodological innovations. It’s this combination of factors (abundant data, GPU-scale computing, and new methodological approaches to processing data), that have made AI successful. What’s new is the ability to create learning systems that can perform human-centric tasks. These include language translation, image classification, and even generating novel dance moves.
OSV: What are the most compelling use cases that you’ve noticed recently?
GM: Image recognition, language translation, and speech transcription are the most compelling. Convolutional Neural Networks (CNN) have proven to be very effective at image recognition and speech transcription, while Recurrent Neural Networks (RNN) have been successful at language translation.
OSV: In the race to own this space, what are the benchmarks that tech companies are trying to hit?
GM: The benchmark right now is human competence. Can the AI recognize objects in an image as well as a human? Can the AI translate voice to text as well as, or better than a human?
JB: But these “narrow” applications of Artificial Intelligence are short term goals. While companies with traditional business life cycles (3-5 years), are focused on automating or improving specific tasks, a few, sponsored by lucrative profit centers or independent investors, are pushing boundaries in the field of AI. This basic research is not targeted at solving specific problems, but in generating human level competency across a broad set of tasks. OSV: What sectors stand to benefit the most from AI?
OSV: What sectors stand to benefit the most from AI?
GM: All sectors can benefit from AI, but we are seeing results in e-commerce, cyber security, customer experience, marketing/sales, supply chain, etc. All mundane, repetitive tasks are also good candidates for AI.
OSV: What kinds of opportunities do you anticipate AI opening up for telcos?
GM: Telcos should use AI technologies to improve the overall customer experience: from targeted marketing and product recommendations, to “intelligent” call centers and churn prevention. Fraud is another area where AI has been successful, both in financial fraud and phone fraud instances, including Wangiri, SIM box, and bypass scams.
OSV: Should we worry about our jobs?
JB: The basic level research right now is about creating unsupervised learning systems, not programmed to accomplish a particular task at all. A good example of this is Geoffrey E. Hinton’s work at Google, training an unsupervised agent to interpret the world of YouTube, which so far has resulted in a system that can identify cats without being told what they are. So I think we’re good for quite some time to come.