Demystifying Buzz About AI

Indicators of AI Technologies on an Accelerated Growth Path

AI technologies require storing, processing and sharing large amounts of data. The impact of AI technologies has increased many-fold in recent times due to (a) reduced cost per byte of storage, (b) reduced cost per byte of network bandwidth, and (c) reduced cost per instruction execution by processor — made possible by evolution in telecom technologies (e.g. 5G), storage technologies (e.g. SSD), and AI chipsets (e.g. NVIDIA GPUs).

According to McKinsey's "Technology Trends for 2022" report, the potential annual impact from AI is $10 to $15 trillion. A few indicators of AI technologies on an accelerated growth path:

• Improved AI Model Training Speed — 100% improvement in training speed for AI models in 2021 versus 2018
• Rapid Innovation in AI Technologies — 30x increase in AI patents filed in 2021 versus 2018
• Substantial Public and Private Investment — $100B private investment in AI companies in 2021
• Market Capitalization of AI Chipset Companies — leading AI chipset company NVIDIA's market capitalization touched $400B recently, doubling in the last couple of years

What Are the Most Noteworthy AI Technologies?

A wide range of technologies and use cases are grouped under the umbrella of AI technologies. Below is a brief summary of some of the most noteworthy:

• Computer Vision — a branch of AI/ML that processes visual data such as images and video, extracting complex information and producing varied interpretations. This impacts many industries: autonomous driving, facial recognition for national security, unmanned quality control for industrial automation, and video analytics solutions like intruder detection.
• Natural Language Processing — processes written and spoken natural language data to produce varied interpretations, powering use cases like virtual voice assistants (Alexa/Siri) and automated voice recognition and response systems (conversational AI).
• Deep Reinforcement Learning — a combination of deep learning and reinforcement learning where a machine/agent learns to make decisions based on live or historical data using algorithms inspired by brain neural networks. This helps build machines/robots that can learn and perform human tasks (e.g. robotic-arm motion control), improving process efficiencies or operating in environments risky for humans, such as fire or flood response.
• Applied AI: Machine Learning, Deep Learning — Machine Learning (ML) uses statistical techniques to enable machines to "learn" from data through a process known as "training" a "model." Deep Learning (DL) mimics layers of neurons in the human brain to learn complex patterns, with "deep" referring to the large number of neuron layers in contemporary models. ML/DL algorithms are applied across industries to both increase revenue and decrease cost — fraud detection in financial services, product recommendations in retail, predictive maintenance in manufacturing, and path/schedule optimization in logistics.

New Frontiers for AI Technologies

While evolution continues at a fast pace across the technologies above, the AI world is opening up to new frontiers being addressed with high energy by stakeholders in both academia and industry. We expect fast-paced movement resulting in new AI technologies for these frontiers.

• Going Beyond Correlation & Finding Causes (Causal AI) — Machine learning and deep learning algorithms learn correlations between input and output variables in historical data, limiting predictions to parameters for which training data is available. Causal inference methods enable modelling of counterfactual or "what-if" scenarios essential for scientific experimentation or business decision-making — such as dynamic pricing for taxi services or discounts for customer acquisition/retention.
• Interpreting Results of ML Algorithms (Explainable AI) — To drive business confidence in AI outcomes, results need to be interpretable. For example, when a bank rejects a loan due to a low credit rating, the bank should be able to explain why — needed both for transparency and, sometimes, by law. Explainable AI (XAI) deals with interpreting the results of AI, with different algorithms and practices helping engineers build solutions that explain/interpret results.
• Building Safeguards from AI Deployments (Responsible AI) — Experts and the public alike suspect potential catastrophic risks from widespread AI adoption. Responsible AI intends to bring policy safeguards to ensure AI technologies abide by laws, incorporate ethics, and implement technical and social robustness to mitigate potential harm. Its tenets include human agency and oversight; societal and environmental well-being; technical robustness and safety; privacy and data governance; transparency; accountability; and diversity, non-discrimination and fairness.