Artificial intelligence for materials discovery
- PDF / 1,734,900 Bytes
- 7 Pages / 585 x 783 pts Page_size
- 9 Downloads / 232 Views
Artificial intelligence and emergence of intelligent machines Artificial intelligence (AI) is a subfield of computer science, with the ambitious goal of studying and designing intelligent systems. In recent years, we have witnessed rapid progress in AI—these systems are now reaching human-level and even superhuman-level performance for a range of tasks such as speech recognition, image interpretation, machine translation (Google translate), and gameplay (DeepBlue1 and AlphaZero2 for Chess, Watson3 for Jeopardy!, and AlphaGo4 and AlphaZero for Go). There is also a general belief that AI is poised to radically transform many components of our society and economy. One example is self-driving cars, which incorporate real-time image recognition and control, and are close to becoming a reality. This tremendous progress is leading to a radical shift in AI research, from a mainly academic endeavor to a much broader field with increasing industrial and governmental investments. Given the tremendous advances in AI, the broader scientific community has taken note and is exploring the use of AI for scientific discovery.5–13 In particular, the materials science community has started using AI techniques to accelerate materials discovery. The main current trend is focused on using machine learning (ML) techniques. This is understandable, in part due to that fact that much of the recent AI achievements, especially those concerning superhuman capabilities such as
face, image, and speech recognition, are rooted in ML, and more specifically, deep learning (DL), which are described in further detail in the section on “Supervised learning and DL.” Nevertheless, while the tremendous progress in DL is undeniable, in areas such as vision, speech recognition, language translation, and autonomous driving, its limitations are well recognized, in particular, given that DL approaches heavily depend on the availability of large amounts of examples or labeled data, which often are not available. The current state of the art of DL has been compared to what Kahneman terms “System 1,” described in his book Thinking, Fast and Slow.14 System 1 encompasses a human’s routine data processing—it is fast, effortless, a type of pattern recognition that is automatic. In particular, perception, including vision and hearing, are part of System 1. Our perceptual abilities are highly developed with a good part of our cerebral cortex devoted to perception. Humans also possess a more rational “System 2,” which in contrast to System 1 is slow, requiring careful thinking and reasoning, and is needed for solving complex problems beyond reflexive responses. Pure ML and DL are not suitable for such complex tasks. Nevertheless, AI encompasses many other techniques such as search, reasoning, planning, and knowledge representation. These techniques have played various roles throughout the different developmental phases of the AI field. They are poised to become more relevant and play
Carla P. Gomes, Department of Computer Science, Cornell University, USA; [email protected]
Data Loading...
