An Inconvenient Truth About AI: AI won't surpass human intelligence anytime soon

TL;DR The history of AI showcases different waves of progress, with the current era being driven by neural networks and massive data. Successful AI deployments have typically involved human supervision or low-cost failures. While AI has potential, it still relies on human intervention in critical situations. The risks of AI systems without human involvement, such as flawed face-recognition technology, highlight the need for cautious implementation. However, in less consequential scenarios, AI can enhance user experiences while requiring their input and adaptation. It is essential to remember AI's role and not be unwittingly influenced.

The 3 Waves

AI has gone through three waves of significant investment. In the 1960s, researchers predicted human-level intelligent machines within 10 years but focused on symbolic reasoning, ignoring neural networks. In the 1980s, rule-based expert systems and neural networks resurged, again raising expectations of dominance in intelligence. The current AI age emerged in the early 2000s with symbolic reasoning algorithms and simultaneous localization and mapping. Neural networks gained momentum in the 2010s with massive data sets, leading to hype and profitable applications.

Human Involvement

Successful AI deployments usually involve human involvement or low consequences of failure. For instance, Roomba's first home-cleaning robot had limited AI to prevent major failures. In military applications, human supervision was crucial to avoid fatal consequences. Advertisements, search engines, and dating sites rely on AI but offer room for human intervention. Self-driving systems are only at Level 2, requiring human attention, and fatal accidents have occurred due to lack of attention.

The AI Failure

AI systems have failed without human involvement, such as wrongful arrests using flawed face-recognition technology. However, AI powers smart speakers, car systems, and language understanding in less critical scenarios. Users adapt to AI agents and provide feedback, becoming the "people in the loop."

Image Source: @Google NGRAMS;

This post is a Precis and the TL;DR, an opinion of "An Inconvenient Truth About AI"