Aktuelles

23.06.23

Exploring the rest of the map of AI

In Artificial Intelligence (AI), while Large Language Models (LLMs) have undoubtedly captured the spotlight, it's essential to recognise that AI has made remarkable strides in various other domains. This blog post goes beyond Natural Language Processing, uncovering groundbreaking advancements in diverse fields. From image generation and segmentation to gaming and scientific breakthroughs, AI continues to push the boundaries of what's possible.

1. Unleashing Creativity: Image-Generating AI

The emergence of image-generating AI models revolutionised the way we perceive computer-generated visuals. Among the notable pioneers is DALL·E, an AI system capable of generating astonishingly realistic images from textual descriptions. With DALL·E, one can envision and materialise unique visual concepts previously confined to the realm of imagination. Also, models like Midjourney and Stable Diffusion have expanded creative possibilities by enabling users to manipulate and transform images with incredible precision and control. Of course, these models have also started a discussion about Copyright, in which different countries take different sides, the most extreme being Japan, which seems to recently have passed a law making copyright inapplicable to AI training.
More recently, Video and Music have been generated by AI. While Video is still in its infancy, music has progressed. The first reactions have been mixed. AI won’t win a Grammy award soon.

2. Mastering Segmentation: Meta's Contribution

A lot of early research into deep learning was based on the desire to analyse images. The image classification challenge in XXXX was the first time, Deep Neural Networks did something previously unheard of - they classified images on a near-human level.

Segmentation, the process of isolating objects or regions within an image, is crucial in computer vision. Meta, formerly known as Facebook, has made significant strides in this area by introducing Segment Anything, a foundation model. This unique system leverages advanced AI techniques to accurately segment objects, regardless of their complexity or background. Meta's innovation paves the way for enhanced object recognition, augmented reality applications, and advancements in medical imaging.

3. Voicebox: Harnessing the Power of Speech

Voice assistants have become ubiquitous in our lives, transforming the way we interact with technology. However, crafting natural and human-like speech synthesis is an intricate task. Voicebox, an AI-powered speech synthesis system (again by Meta), has made significant strides in this domain. By leveraging generative AI techniques, Voicebox creates highly realistic and expressive voices, opening up new possibilities for personalised interactions, voice-over applications, and accessibility for individuals with speech impairments. However, due to its potential for misuse (see our blog post here), Meta has decided not to release the model behind Voicebox for now.

4. From Games to Grand Challenges: AlphaGo's Triumph

DeepMind's AlphaGo stunned the world when it defeated the reigning Go world champion, Lee Sedol, in 2016—this watershed moment showcased the immense potential of AI in tackling complex strategic games. AlphaGo's success relied on deep learning, reinforcement learning, and powerful computational resources. Beyond the game of Go, this triumph ignited a renewed enthusiasm for using AI to solve grand challenges, demonstrating its capacity to tackle real-world problems with intelligence and ingenuity. The Game of Go was soon followed by Starcraft and other, even more complex games, which the AI learned to excel at in no time. While niche in its game-playing ability at the time, the technology behind AlphaGo (and its successor, AlphaZero) has paved the way for several other spectacular breakthroughs in AI.

5. Unlocking the Power of Fusion: AI in Nuclear Fusion

Nuclear fusion is one of the big hopes for clean and cheap energy. However, it is almost a joke amongst researchers that ‘nuclear fusion is 20 years away’ - independently of when one talks about the topic. A big problem with this technology is that incredible forces must be controlled with fantastic precision. One of DeepMind's follow-ups to AlphaGo is to support this control of nuclear fusion. AI has joined with scientists in this quest by developing AI systems trained to control and optimise fusion reactors. These AI models analyse complex data from fusion experiments, enabling scientists to make informed decisions, improve reactor performance, and expedite the development of fusion technology. The marriage of AI and nuclear fusion offers hope for a sustainable energy future.

6. Decoding the Blueprint of Life: AlphaFold's Impact

In our first blog post, we mentioned how one of the most complex problems of falling to AI was the protein folding problem. Understanding the intricate folding patterns of proteins is a vital pursuit in fields like biology and medicine. DeepMind's AlphaFold has made remarkable strides in this area by employing deep learning to predict protein structures with unprecedented accuracy. This breakthrough can potentially revolutionise drug discovery, bioengineering, and our understanding of diseases, as accurate protein structure predictions unlock insights into their functions and interactions. AlphaFold's impact heralds a new era of computational biology and the transformative potential of AI in life sciences. More than a few of my friends in Biology are actively using the database procured by AlphaFold to speed up their research, and others have used this technology to go from target to hit molecule in 30 days, a process that was previously - at the fastest - measured in months.

While Large Language Models like ChatGPT have dominated AI discourse in recent times, the world of AI is far-reaching and diverse. From the awe-inspiring capabilities of image-generating AI models to groundbreaking advancements in segmentation, gaming, speech synthesis, nuclear fusion control, and protein folding, AI pushes boundaries and reshapes various domains.

If you want to learn more about how AI can help your business, reach out!

Thomas Rost has studied Cognitive Science and Artificial Intelligence all over Europe. He has been an advisor to companies harvesting the benefits of Data Science, Machine Learning and AI for many years, seeing the rise of Deep Learning first-hand and helping his customers derive value from futuristic technologies, generate insights from data and support decisions with evidence.

He is a Senior Manager Consultant at SKAD. If you have questions, reach out to him at t.rost@sk-advisory.com or LinkedIn.
zurück
top