Understanding AI as a Megatrend: Decomposing Its Subfields and Sub-Trends

The AI megatrend can be broken down into five key fields, each with sub-trends that collectively enable and accelerate the evolution of AI. Let's explore these fields in greater detail:

‍

Machine Learning & Deep Learning

This field has three sub-trends:

1. AI Capabilities

2. Scope of Input and Output Data

3. Architectural Convergence

It makes sense to focus on technical development first and understand the driving forces enabling AI to conquer more use cases. The analysis presented in this article does not have the aspiration to be complete and will introduce the relevant aspects rather than deep-diving into technicalities. Do not hesitate to reach out if you want more information on any of the topics that are touched on.

‍

1. AI Capabilities

Bluntly, AI's transformative power lies in its growing ability to recognize, transform, and generate patterns across diverse domains.

1. Pattern Recognition: AI excels in identifying patterns in vast datasets, which has revolutionized fields like healthcare (e.g., detecting anomalies in medical imaging), finance (e.g., fraud detection), and manufacturing (e.g., predictive maintenance).

2. Transformation: AI systems can transform raw data into different forms, such as converting speech to text, translating languages, or summarizing lengthy documents into actionable insights.

3. Generation: AI has advanced into the generative domain, producing text, images, videos, and 3D models. Tools like generative adversarial networks (GANs), diffusion models, and applications like ChatGPT have opened new creative and practical possibilities.

2. Scope of Input and Output Data

Modern AI is no longer constrained by specific types of data.

1. Data Versatility: AI can now process virtually any form of data — text, audio, images, video, tabular data, or sensor readings. This is achieved through more robust vectorization techniques, enabling machines to handle and interpret heterogeneous inputs. This, in turn, feeds the transformation capability highlighted before.

2. Reliable Reproduction: AI can generate outputs with remarkable fidelity, from synthesizing voices to creating lifelike visuals, making it an indispensable tool in industries like entertainment, education, and virtual reality.

3. Architectural Convergence

The development of AI has undergone several waves of architectural innovation and standardization:

1. Neural Networks and Backpropagation: These foundational technologies became the dominant paradigm for supervised learning, underpinning the success of early AI models.

2. GPUs as Compute Powerhouses: With the advent of GPUs (also driven by the CUDA standard), AI training and inference achieved unprecedented speed and efficiency. GPUs became the backbone of AI computing infrastructure.

3. Transformers: This model architecture represents the latest leap, offering unparalleled capabilities in contextualized representations for tasks like language understanding (e.g., the GPT models used in ChatGPT), vision processing (e.g., Vision Transformers), and even multimodal systems that combine text, images, and more.

‍

Data Availability

The digitization megatrend that preceded AI has laid the groundwork for the abundance of data on which AI thrives.

1. Proliferation of Sensors and Robotics: Sensors embedded in manufacturing lines, vehicles, wearable devices, and smart infrastructure collect data at an unprecedented scale and granularity.

2. Robust Foundation Models: The vast datasets now available have enabled the creation of foundation models, such as GPT-4, DALL-E, and BERT trained on trillions of data points (text and more). These models serve as adaptable building blocks for a wide array of applications.

‍

Connectivity & Compute

AI's practical application hinges on its ability to be deployed efficiently and accessibly.

1. Edge Computing: AI systems increasingly operate on devices close to the user, ensuring minimal latency for tasks like real-time translation or autonomous vehicle steering.

2. Cloud and Data Center Advancements: For resource-intensive tasks, centralized computing in data centers provides immense processing power. Innovations in network infrastructure ensure that AI services are delivered with speed and reliability.

The computing landscape is scaling on two fronts simultaneously:

1. Linear Scaling: The number of data centers is growing to meet demand, expanding AI's global footprint. This has reached a point where energy use can almost be equalized with data processing. Recently, Microsoft's Satya Nadella pushed for a focus on "tokens per watt per dollar" to further empathize the energy efficiency attention around AI.

2. Exponential Performance Growth: Hardware advancements, from specialized AI accelerators like TPUs to energy-efficient chips, continually improve performance per watt, server, and data center.

‍

Society

The societal impact of AI is profound and multifaceted:

1. Public Trust: Increasingly, people are relying on AI for cognitive tasks, from virtual assistants to medical diagnostics, as these systems demonstrate reliability and accuracy. Research also shows that trust towards AI and companies utilizing AI increases with knowledge about AI (see illustration below).

2. Hopes and Concerns: While AI fuels optimism about solving complex problems, it also raises ethical concerns regarding data privacy, fairness/bias, and job displacement among other worries.

3. Regulation and Governance: Regulatory frameworks are emerging globally, with differing approaches in regions like the EU (e.g., AI Act) and the US or China, reflecting the tension between fostering innovation and ensuring safety and accountability.

‍

In Conclusion  

We hope this article conveys that AI's progression as a megatrend is driven by the interplay of different fields, sub-trends, and their associated advancements. It is super tough to make a prediction on the AI megatrend and what it will deliver next. One or two levels deeper, looking at the megatrend in a dissected manner helps to understand and predict individual aspects. It is much easier to predict where, e.g., the field of "compute" will go next – and then deduce what this means for the megatrend and other contributing factors. Once the future trajectory is understood, venture opportunities can be identified.