Summary

In 2022, DeepMind has made significant strides in AI applications, focusing on solving real-world problems, from strategic gaming and data compression to protein folding and nuclear fusion control, showcasing AI's potential to address complex challenges.

Abstract

The year 2022 marked a period of exponential growth in AI, as evidenced by a surge in publications and the emergence of practical applications. DeepMind, a Google-owned AI company, has shifted its focus from general AI tools like GPT3 and Dalle2 to more targeted real-world applications. Their AI, MuZero, has excelled in strategic games and is now being applied to develop new data compression strategies, potentially reducing global data traffic and improving the carbon and energy footprint. In the realm of pharmaceuticals, DeepMind's AlphaFold has made significant progress in predicting protein structures, which is crucial for drug discovery and could revolutionize medical research. Additionally, DeepMind's AI has made strides in controlling tokamak plasmas, a critical step towards making nuclear fusion a viable energy source, aligning with Stephen Hawking's vision of a clean, inexhaustible power supply.

Opinions

The author views the advancements in AI, particularly by DeepMind, as impressive and world-changing, emphasizing their potential impact on various industries.
Conversational AI, such as BlenderBot3, is recognized for its human-like interaction capabilities and its ability to learn and improve through conversation.
Generative image AI tools like Dalle2, Imagen, and Stable-Diffusion are considered fascinating and indicative of the potential of AI in creative fields.
The advancements in self-driving cars, now capable of understanding natural language, are seen as a significant leap towards making the technology more flexible and suitable worldwide.
The author expresses admiration for the major improvements in core computational tasks, such as data simulation compression, where AI has achieved up to 100x compression rates.
DeepMind's approach to AI research, which involves tackling major challenges with real-life applications, is presented as a more impactful direction compared to the creation of "toy tools" with limited practical use.
The importance of AI in reducing data traffic and improving the global carbon and energy footprint is highlighted, showcasing AI's ability to learn complex tasks without prior information.
The author underscores the significance of AI in protein folding and drug discovery, suggesting that this research could lead to the development of more efficient drugs across various medical fields.
The work

Overview of Real World AI Applications — DeepMind Edition

Here’s a (not so) quick overview of 2022’s major AI trends, with a focus on real world applications developed by Deepmind.

2022, AI exponential growth and impressive achievements

Exponential Growth of the number of AI publications

2022 has been an amazing year for Artificial Intelligence (AI), with an exponential growth in AI publications and emerging applications. A recent study showed that the number of publications doubles every 24 months (image: twitter source)!

The number of AI publications doubles every 24 months twitter source

2022 brought amazing new AI applications

To mention a few impressive achievements this past year,

Conversational AI: if you haven’t already, check out blenderbot3, an AI chatbot hard to distinguish from a human and that can learn, and improve, through conversation: https://blenderbot.ai/

Generative images: should I really mention, Dalle2, Imagen, Stable-Diffusion, Midjourney, and so on? Here are four portraits I generated with Stable-Diffusion to illustrate the potential of these tools! Simply fascinating!

Stable diffusion, prompt based portrait generation, 2022

Self-driving cars: they can now understand natural language making it a flexible and suitable technology worldwide. Check out this video

Google’s self driving car can follow high-level natural language instruction for navigating a real-world environment

Major improvement in core computational tasks such as data simulation compression with up to 100x compression! You can check it out here.

In a nutshell, these are major improvements where humans reached their limits a while ago.

And this is merely a sample of current AI development. I’ll be covering the state of the art of AI applications and their evolution in the coming posts, so please subscribe to get the latest news ;)

Deepmind’s recent breakthroughs solving real world applications!

Although we had a lot of fun playing with GPT3, Dalle2, stable diffusion and so on. So far, they remain toy tools with limited life changing applications. Deepmind on the other hand, a major AI company owned by google and focusing on General Purpose AI, has been steering its research towards major challenges with real life applications.

Here are 3 major breakthroughs worth mentioning

Strategic gaming & Data Compression

What? Muzero, a general-purpose AI who proved to be very efficient at beat world champions in chess, Go, startcraft2 and so.

MuZero: mastering Go, chess, shogi and Atari without rules, and now capable of developing data new compression strategies! (source )

Real Life Applications ? Data compression.

Model data compression problem as a game where your score is related to the compression efficiency of the algorithm while maintaining data quality: “MuZero achieves superhuman performance across various tasks by combining the power of search with its ability to learn a model of the environment and plan accordingly. This works especially well in large, combinatorial action spaces, making it an ideal candidate solution for the problem of rate control in video compression.”

Why is it important? Successfully reduce data traffic which means an improved worldwide carbon & energy footprint! This research also proves that AI can learn without prior information on the problem. It simply needs a goal and a s et of parameters to manipulate.

Protein folding and pharmaceutical applications

What? Predict 3D structure of a protein.

Real Life Applications ? Drug discovery

Why is it important? Knowing the 3D structure of a protein helps anticipate molecular interactions between potential drugs and targeted cells. This research should supercharge pharmaceutical research and come with new, more efficient drugs on a wide range of medical fields.

Nuclear Fusion, Magnetic Control of Tokamak Plasmas (#IronMan for geeks)

What? Maintain plasma stable at 100s millions of degrees Celsius (hotter than the sun)

Real Life Applications ? It’s a prerequisite to enable nuclear fusion (and power iron man’s suit!)

Accelerating fusion science through learned plasma control (source )

Why is it important? It would lead to a safe, sustainable, nonpolluting energy source. Nuclear fusion doesn’t have nuclear fissions’ drawbacks (currently used worldwide) such as radioactive waste and risk of explosions. Some scientists would even say, this is the ultimate power source humans need to master. To quote Stephen Hawking “I would like nuclear fusion to become a practical power source. It would provide an inexhaustible supply of energy, without pollution or global warming.”

What about you?

If you had to choose one AI innovation in the past year, one that fascinates you or maybe you think has the potential to change the world as we know it, which one would it be??

If you made it until here, it means you’re a true geek. I’ll be covering the state of the art of AI applications and their evolution in the coming posts, so please subscribe to get the latest news ;)

Subscribe and join BoredGeekSociety.

Have a great day!