Rebooting AI: Building Artificial Intelligence We Can Trust

Gary F. Marcus

Artificial Intelligence (AI) is a hot topic nowadays, and not a day goes by without hearing some news or opinions about AI’s potential to change the world. However, behind the hype, there’s also a growing concern about AI’s limitations, and how far it can go in achieving its lofty goals. In the book “Rebooting AI: Building Artificial Intelligence We Can Trust,” authors Gary Marcus and Ernest Davis take a realistic look at where AI stands currently and provide an insightful overview of how it can move forward towards reliable and robust AI.

One of the main points that Marcus and Davis make is that the current boom and hype surrounding AI are often focused only on one aspect, which is Deep Learning. While Deep Learning has made significant strides in recent years, it also has significant limitations, such as being data-hungry and not being explainable or transparent. Marcus and Davis show that relying solely on data-driven approaches like Deep Learning is not enough to achieve true general AI that can learn and reason across multiple domains and contexts.

Therefore, Marcus and Davis argue that a combination of knowledge-driven and data-driven approaches is needed to create general AI that can go beyond the limits of Deep Learning. Knowledge-driven approaches rely on rules, logic, and prior knowledge to structure the learning process and guide the search for patterns and correlations. This approach can help overcome some of the limitations of data-driven approaches and provide a more explainable and transparent AI system.

One of the strengths of “Rebooting AI” is how it conveys complex concepts and technical details in a simple and accessible way. The authors use real-world examples and stories to illustrate their points, making the book engaging and enjoyable to read. For instance, they describe how a simple logic rule can solve a game of Sudoku, which is beyond the reach of most Deep Learning algorithms. This example highlights how knowledge can play an essential role in AI’s development and how combining knowledge and data can create more robust and reliable AI systems that can handle complexity and uncertainty.

Another misconception that Marcus and Davis address is the idea that AI is on the verge of creating self-aware machines that can think and reason like humans. The authors argue that AI’s progress is still far from approaching human-level intelligence and that it will take many breakthroughs in AI research before we can achieve anything close to that. Moreover, they warn about the dangers of expecting too much from AI, such as the fear of creating an unfriendly AI that can harm humans, as depicted in science fiction movies like Terminator or The Matrix.

In conclusion, “Rebooting AI” provides a refreshing perspective on the current state of AI and its potential for meaningful progress towards more reliable and robust AI systems. By challenging the trend of promoting AI through press releases and highlighting the limitations of data-driven approaches, Marcus and Davis offer a more realistic and pragmatic approach to AI’s development. Ultimately, AI’s success will depend on how well we can balance the technical challenges with the ethical and societal implications of AI’s impact. As such, “Rebooting AI” is an essential guide that can help us navigate the complex terrain of AI research and development.