Let me explain my reasons for the AI revolution.
This was publicly articulated during a fireside chat with SGX and CSOP in February this year.
Harnessing the Power of Technology and Innovation in China (part 1)
In 21:55 onwards:
I believe that we are in the midst of an AI revolution that begun in 2020s. Previously, there has been two AI “winters” in the past 50 years [1974 – 1980 and 1987 – 1993].
But, this time is different.
Why?
There are three main reasons:
Massive data availability
Cheaper cost of computing
More advanced algorithms
Hence, I was not surprised with the rise of ChatGPT.
Let me dive deeper into these reasons.
Massive Data availability
The digital era has generated an unprecedented amount of data especially from social media, internet-connected devices, and other digital platforms.
This is important.
Because data is the fuel that powers AI systems.
AI models extract valuable insights from this vast trove of information, enabling them to make decisions with remarkable accuracy.
The sheer volume of data is crucial in their performance over time. In another words, more data means better AI.
Consider the domain of image recognition.
Here, AI model, such as convolutional neural networks (CNNs), is able to learn from millions of meticulously labeled images, allowing them to discern and identify objects with astonishing precision.
.
Cheaper Cost of Computing
Throughout history, AI has its fair share of challenges, with computing power a formidable barrier. In fact, there were two AI "winters" because the required computing power is simply too expensive.
Today, advancements in technology such as GPUs and TPUs have made computing power more affordable.
The training of AI models, particularly the deep learning models with their numerous layers, can now be accomplished efficiently on these affordable processors.
The lower costs of sophisticated processors have democratized AI allowing organizations of all sizes to leverage powerful computing resources.
Training time is reduced significantly, resulting in improved AI performance and accomplishments of difficult tasks such as natural language understanding.
.
More Advanced Algorithms
Researchers have made significant breakthroughs in neural network architectures, optimization techniques, and learning algorithms. This resulted in more sophisticated AI models in areas such as reinforcement learning, generative models, and unsupervised learning.
These advancements have pushed the boundaries of AI with the ability to now handle complex tasks, process unstructured data, and learn from limited labeled data.
One remarkable advancement enabled by these algorithms is the introduction of transformer models, including GPT. These models have revolutionized language understanding and made thee hugely popular ChatGPT a possible innovation.
Another noteworthy example of advanced algorithms is reinforcement learning. By interacting with their environments and receiving feedback in the form of rewards, AI systems can learn optimal strategies. This has found applications in autonomous vehicles and game playing, where AI systems learn to make decisions that maximize desired outcomes.