Artificial Intelligence (AI)
Discuss current events in AI and technological innovations with Intel® employees
544 Discussions

Next, Machines Get Wiser

0 0 1,333

Published July 15, 2020

Gadi Singer is Vice President and Director of Emergent AI Research at Intel Labs leading the development of the third wave of AI capabilities.


Where is machine intelligence headed in the next five years? All signs indicate a categorical jump in artificial intelligence (AI) cognitive competencies with machines growing markedly wiser. To DL or not to DL — that is the cardinal research question on the path to higher intelligence.

Deep learning (DL) will continue to make significant progress in technical capabilities and scope of deployment across all aspects of life, including revolutionizing healthcare, retail, manufacturing, autonomous vehicles, security and fraud prevention, and data analytics. However, to build the future of AI, it is necessary to define a set of goals and expectations that will drive a new generation of technologies.

Many of the current DL applications focus on tasks related to visual recognition, natural language processing (NLP), translation, and recommendation systems. These tasks provide exceptional results based on differential programming and sophisticated data-based correlation and manipulation. However, they broadly lack the ability to understand, conceptualize, and reason, still awaiting the capabilities I noted in my blog from Jan 2018 that will allow AI to move from recognition to understanding

Yoshua Bengio, a co-recipient of the 2018 ACM A.M. Turing Award, provided a compelling articulation of this upcoming new phase of AI in his keynote talk at NeurIPS 2019: From System 1 Deep Learning to System 2 Deep Learning. Using the paradigm defined by Daniel Kahneman in his book, Thinking, Fast and Slow, Bengio equates the capabilities of DL today to what is characterized as “System 1” – intuitive, fast, unconscious, habitual. In contrast, he asserts that what the AI field needs to solve and provide next are the capabilities of “System 2” – slow, logical, sequential, conscious, and algorithmic, such as the capabilities needed in planning and reasoning.

If you accept this assertion that a fundamental new class of AI capabilities is about to emerge, the next question divides the field: Would these considerate, logical, well-reasoned capabilities be powered by further improvements in DL, or will they require a new set of technologies altogether? In particular, one proposal is the integration of DL with symbolic reasoning and deep knowledge. While Bengio charts a path expected to accomplish these new competencies by advancement within the domain of DL, others, myself included, believe that the best path leads well beyond just enhanced DL.

Leslie Valiant, a 2010 ACM A.M. Turing Award winner, highlighted the complementary aspects required to achieve “intelligent cognitive behavior.” He writes about the two most fundamental aspects of intelligent cognitive behavior:

1.     The ability to learn from experience.

2.     The ability to reason from what has been learned.

According to Valiant, we are seeking an understanding of knowledge that can computationally support the basic phenomena of intelligent behavior. Differentiable programming is extraordinarily equipped to learn from facts and patterns, but struggles with the machine faculty to reason while effectively partnering with humans. 

Neuro-symbolic AI (also known as neural-symbolic AI) is gaining recognition as a highly promising direction that combines the strengths of neural networks and symbolic reasoning. This led to media headlines earlier this year declaring it to be the “next big thing” (for example: Neuro-symbolic AI is the future of artificial intelligence and What are neural-symbolic AI methods and why will they dominate 2020?). A good part of the attention was sparked by research at the MIT-IBM Watson AI Lab and work around a new, large-scale video reasoning dataset called CLEVRER. While the neuro-symbolic direction presents high potential, I believe that a few key innovations are required to enable the breakthroughs needed for real-world application.

Intel Labs has established Cognitive Computing Research to drive innovation in machine intelligence and cognition. In an upcoming series of blogs, I will outline some of the key aspects underlying this advanced research, combining deep learning with deep knowledge structures and symbolic reasoning to materially improve efficiency, explainability, extensibility, and reasoning capabilities of AI systems. A broader question will also be evaluated – are we in the transition time from the Information Age to an emerging Age of Knowledge?

We look forward to building next-generation AI systems that will one day understand this blog post and other informative content – and deliver even greater benefits to our lives.


Read more

Tags (1)
About the Author
Gadi joined Intel in 1983 and has since held a variety of senior technical leadership and management positions in chip design, software engineering, CAD development and research. Gadi played key leadership role in the product line introduction of several new micro-architectures including the very first Pentium, the first Xeon processors, the first Atom products and more. He was vice president and engineering manager of groups including the Enterprise Processors Division, Software Enabling Group and Intel’s corporate EDA group. Since 2014, Gadi participated in driving cross-company AI capabilities in HW, SW and algorithms. Prior to joining Intel Labs, Gadi was Vice President and General Manager of Intel’s Artificial Intelligence Platforms Group (AIPG). Gadi received his bachelor’s degree in computer engineering from the Technion University, Israel, where he also pursued graduate studies with an emphasis on AI. ​