Over the years, artificial intelligence has amazed everyone with numerous breakthroughs, and this year it was no different. The whole year, we witnessed awe-inspiring innovations in reinforcement learning, neural networks, among others. Tech companies from across the world benchmarked various leaps in artificial intelligence to further eliminated the doubts people had about achieving true AI.
As a chronicler of the technological progress in the space of analytics, artificial intelligence, data science and big data, among others, Analytics India Magazine was on top of every jaw-dropping development. We bring to you the top 7 amazing AI advancements that changed the world forever.
1. Robot Hand’s Dexterity
OpenAI’s successfully trained a robot hand called Dactyl that adopted to the real-world environment in solving the Rubik’s cube. The robot was entirely trained in the simulated environment but was able to transfer the knowledge into a new situation successfully.
To enhance the dexterity, OpenAI used automatic domain randomisation technique and improved the capabilities of hand for solving the Rubik’s cube. Although Dactyl solved the cube, the key take away was the ability to deliver results in the environment that the robot was not trained for.
2. Deepfake – Bringing Picture To Life
Samsung, in May, created a system that can transform facial images into video sequences. They used the generative adversarial network (GAN) to create deep fake videos just by taking one picture as input. Researchers from Samsung used high-fidelity natural image synthesis for enabling ML models to resonate realistic human expression.
3. AI-Generated Synthetic Text
OpenAI, in February, released a small model called Generative Pre-Training (GPT) to generate synthetic text automatically. The firm eventually released the full version of the model, GPT-2, in November. On writing a few sentences, the model perfectly picked the context and generated text on its own.
The model was trained with over 8 million web pages, resulting in creating content that was difficult to determine whether it was a generic or synthetic text.
4. Gamification Of Memories
Google’s DeepMind changed the way reinforcement learning works to gamify memory. To allow AI agents to make better decisions in the present, they used Temporal Value Transport (TVT) for sending lessons from the future. This allowed the agent to understand the long term consequences of decisions that can be taken at present. Although the methodology was carried out in a game, it was unprecedented in the AI landscape.
5. Solving Three-Body Problem
The three-body problem was one of the longest-standing predicaments in the scientific community. Precisely identifying the future position of objects has numerous use cases, especially in the space, determining the position of heavenly bodies allows scientists their research. The researchers of Edinburgh used neural networks to pinpoint the future location, thereby, opening up the door for extending it into n-body problem.
6. Upside Down Reinforcement Learning
A team from Swiss AI Lab introduced a new methodology calling it an upside-down reinforcement learning. They successfully carried out reinforcement learning in the form of supervised learning. This allowed the team to provide rewards as input, which is contrary to how the traditional reinforcement learning work.
Such a technique also enabled them to imitate and train the robot for carrying out strenuous tasks just by imitating in front of a machine. The command-based methodology assists ML models in expediting the training process, thereby decreasing the time required in the AI workflows.
7. Explainable AI
AI is making great strides, but understanding the methodologies inside the black box is crucial for bringing thrust. Therefore, different companies released services to allow businesses to underline the prime factors that lead to outcomes from their machine learning models. For the first time, firms are able to clear the cloud and gaining insights into the way the black box works. Although one cannot yet obtain all the aspects of a conclusion from models, it has a more prominent role to play in democratising AI.
