Garry Kasparav, the chess legend defeated `Deep Blue’, the powerful computer built by IBM in a historic match in 1996. A year later, in a re-match, much to the surprise of many, a vastly improved computer beat the International Grandmaster.
What happened was `Deep Blue-2’ had used the same heuristics as Deep Blue-1, but it was empowered with more CPU power. Today, two decades later, we see significant advances in Artificial Intelligence (AI).
IBM’s Watson built jeopardy, combined by speech recognition, search and speech generation. It competed against humans without any handicap and defeated the champions, explained M Vidyasagar, an expert talking about advances in game playing and deep learning.
For those of us excited or stumped by Artificial Intelligence (AI) and Machine Learning (ML) and might be thinking they are recent as well as new, wait. Experts say, the present surge in these areas, is at least the third in a cycle of hype and expectation.
“This is the 3rd or 4th cycle of AI hype. Starting from the late 1950s – Perceptron – a precursor to AI & invented by Frank Rosenblatt. In the 1970s and early 1980s – we saw Rule-based “Expert Systems” and in Mid 1980s, 1990s – we saw Artificial Neural Networks and in Current Cycle we see Deep Learning Convolutional networks”, says M Vidyasagar, a leading expert and former Director of the Centre for Artificial Intelligence and Robotics (CAIR), Bengaluru.
In the earlier cycles, only those methods that had a solid mathematical foundation have survived, while others have faded. In the present cycle, researchers have recognised the need for foundational analysis of popular methods like deep learning, he said, delivering the 4th John McCarthy Memorial Lecture 2019 of the IEEE Hyderabad Chapter.
A SERB distinguished Fellow and distinguished Professor at the IIT Hyderabad, Vidyasagar gave a glimpse of AI and its impact in building the machines which can sense, analyze, reason and respond in the Lecture titled “AI: An Unfinished Revolution”. AI Excitement has been driven with advancements in Image processing, Speech processing, Natural Language processing and Game Playing.
The current wave of success is driven by factors like availability of massive amounts of training data, advances in computation, special-purpose hardware such as GPUs and Google’s TPUs, developments in algorithms for machine learning. In addition, AI is now ”democratized” – a lot of open source software is available (ex: Tensorflow, Keras). Anyone can build an AI application very quickly, he explained.
On the challenges facing the growth of these exciting technologies, Vidyasagar said advances in Speech and Natural Language Processing – IVR systems expanded very well on structured queries but accuracy still needs to be handled on unstructured queries. Some of the On-line translation between languages cannot capture the “mood’ – suitable for business contracts nor for poetry.
Current use of massive computational resources shuts out all but a few enormous sized companies from doing research in AI – especially true of ‘deep learning’ research, he pointed out.
Hitendra Sharma, Chair, Computational Intelligence / GRSS Society of IEEE Hyderabad Section said the lecture is organized in the memory of John McCarthy who is the father of Artificial Intelligence.
Atul Negi, Prof of AI at UoH described the connections associated to Control Systems, embedded logic driving into Artificial Intelligence in his remarks.
Credit: Google News
