2020 has been a year of unprecedented events, mostly those that made us want to fast-forward instead of rewinding. But believe you me, while the world grappled with the pandemic, the magic of artificial intelligence continued to spread far and wide – giving all of us AI enthusiasts some reason to cheer and smile.

Let's end the year on a high note by looking back at some of the pathbreaking AI innovations of 2020.

• One of the biggest problems in biology solved by AlphaFold AI: How can we determine the structure that each of the over 400,000 proteins will fold into? This is called the Protein Folding Problem which has been intriguing scientists for half a century. Through a better understanding of protein folding and unfolding errors, we will be able to better treat diseases and produce more effective medicines and vaccines. On 30 November, for the first time in 50 years, an AI called AlphaFold by DeepMind has finally cracked this grand challenge of biology. The latest version of Alphabet’s AI system has been recognised as an effective solution to protein folding by the organisers of the biennial Critical Assessment of protein Structure Prediction (CASP). This breakthrough demonstrates the impact AI can have on scientific discovery and its potential to dramatically accelerate progress in some of the most fundamental fields that explain and shape our world. 

Read also: Solving one of the biggest problems in biology is AI's latest landmark moment

• Natural language processing took a giant leap with OpenAI’s GPT-3: NLP capabilities have been tremendously advanced with OpenAI’s GPT-3 that has close to 175 bn trainable parameters. Generative Pre-trained Transformer 3 is an autoregressive language model that uses deep learning to produce human-like text. It is the third-generation language prediction model in the GPT-n series created by OpenAI, a San Francisco-based AI research laboratory. GPT-3’s parameter capability is the biggest out there in the market, better than Microsoft’s ZeRO-2 which has 170 bn parameters. GPT-3 also moves away from the traditional approach of the pre-trained language model, which trains a model on specific datasets for task-specific model architectures. GPT-3 has generated news articles that human evaluators found hard to distinguish from human-written articles.

Read also: GPT-3: The double-edged AI sword

Listen: GPT-3 Explained | Last Week on INDIAai Ep #8

• India’s PARAM Siddhi AI ranked #63 globally in supercomputing: India’s fastest and largest supercomputer PARAM Siddhi AI, commissioned by C-DAC, has been ranked #63 in the latest TOP500 Supercomputing list. PARAM Siddhi AI can achieve 210 AI Petaflops and is highly capable of addressing largescale challenges in healthcare, manufacturing, defence and urban planning among others – enhancing India's potential to become the AI garage of the world. The Center for Development of Advanced Computing (C-DAC) is intent on setting the benchmark as the facilitator and manufacturer of indigenous high performance computing capabilities. PARAM Siddhi AI’s superior compute capabilities, a mature HPC AI framework and AI libraries in SDK will jumpstart research efforts and support startups to execute largescale projects.

Read also: C-DAC – A Make In India force that’s powering India’s AI revolution

• Graphcore unveiled the world’s most sophisticated IPU for AI: Amidst the race for developing IPU processors that can train Artificial Intelligence (AI) systems, the British chip designer Graphcore unveiled the Colossus MK2 GC200, which boasts of 59.4 billion transistors. This chip has superseded NVIDIA A100 as the world's most complex processor. This pizza-box sized Intelligent Processing Unit is “completely plug-and-play” and users will be able to connect up to 64,000 IPUs together for a total of 16 exaflops of computing power. This will accelerate machine intelligence and enable innovators to make AI breakthroughs.
• AI has enabled robots to feel pain and self-heal: The scientists of the Nanyang Technological University, Singapore have developed an AI-based ‘mini brain’ for robots which gives them the ability to sense pain and heal themselves when damaged. The system has AI-enabled sensor nodes on the robotic skin to process and respond to 'pain' arising from pressure exerted by a physical force. The system also allows the robot to detect damage and repair its own mechanical functions with a self-healing ion gel material, without the need for human intervention. Unlike other sensors that send information to a single large CPU, the new NTU approach embeds AI into the network of sensor nodes. This reduces wiring requirements and response time by five to ten times when compared to conventional robots. By perceiving their environment to learn and adapting accordingly, these sensors can make effective decisions. This innovation is necessary for the next generation of robots to interact effectively with humans.

Read also: Four recent research advancements that will give AI human-like traits

INDIAai wishes all our readers a very healthy and happy start to 2021! Keep following us as we trace India's journey to global prominence in artificial intelligence.

Image from Shutterstock