AI Is Hard But Worth the Investment – PCMag.com

In 1955, scientists behind the first AI research project believed it would take a 10-man team two months to develop thinking machines that could replicate the problem-solving capabilities of the human mind. But six decades, thousands of projects, and billions of dollars later, human-level artificial intelligence remains an elusive goal.

The difficulty of achieving human-level AI has split the field into two subdomains: artificial general intelligence (AGI), the original vision of "thinking" machines; and artificial narrow intelligence, a limited but easier-to-achieve application now found in many industries.

The more we make advances in AI, the more we come to appreciate the complexity of the human brain. But does that mean we should abandon the pursuit of artificial general intelligence?

Many scientists have become disillusioned about cracking the code of AGI. In his latest book, Architects of Intelligence, futurist and author Martin Ford asked 23 prominent AI scientists and thought leaders how long it would take to achieve AGI. Five refrained from giving an estimate, and most of the remaining 18 preferred to guess anonymously. Their mean estimate for AGI was 209980 years from now.

"We have been working on AI problems for over 60 years. And if the founders of the field were able to see what we tout as great advances today, they would be very disappointed, because it appears we have not made much progress. I don't think that AGI is in the near future for us at all," said Daniela Rus, Director of the MIT Computer Science and AI Lab (CSAIL), one of the scientists Ford interviewed.

Other scientists argue that pursuing AGI is pointless. "We don't need to duplicate humans. That's why I focus on having tools to help us rather than duplicate what we already know how to do. We want humans and machines to partner and do something that they cannot do on their own," Peter Norvig, Director of Research at Google and the co-author of the leading AI textbook, said in a 2016 interview with Forbes.

Deep-learning algorithms fail at simple, general problem-solving: tasks that humans learn at a very early age, such as understanding the meaning of text and navigating open environments.

But deep learning is efficient in narrow applications such as computer vision, cancer detection, and speech recognition. In many cases, it surpasses human performance considerably. Most of the current research and funding in AI is focused on these narrow AI or intelligence augmentation applications, the kind Norvig suggests.

While narrow AI makes inroads into new fields every day, the few AI labs still focused on artificial general intelligence continue to burn through mounds of cash and seem to make very little progress toward human-level AI (if any).

Alphabet-owned AGI lab DeepMind incurred $570 million in losses in 2018 alone, according to documents it filed with the UK's Companies House registry in August. OpenAI, another AI lab that aims to create AGI, recently had to shed its nonprofit structure to find investors for its expensive research. Both labs have accomplished remarkable feats, including creating bots that play complex board and video games. But they're still nowhere near creating artificial general intelligence.

So, should we abandon the pursuit of AGI? Or should we focus on finding practical (and profitable) applications for current narrow AI technologies and stop funding AGI research?

Often overlooked in the failure to create AGI are the big rewards we've reaped in six decades of AI research. We owe many scientific advances and tools that we use every day to failed efforts to replicate the human brain.

One of my favorite quotes in this regard comes from Artificial Intelligence: A Modern Approach, the famous AI book Norvig co-authored with distinguished scientist Stuart Russell. "[W]ork in AI has pioneered many ideas that have made their way back to mainstream computer science, including time-sharing, interactive interpreters, personal computers with windows and mice, rapid development environments, the linked list data type, automatic storage management, and key concepts of symbolic, functional, declarative, and object-oriented programming," Norvig and Russell wrote.

We would have had none of those things (nor smartphones, smart speakers, and smart assistants) had it not been for scientists chasing the wild dream of creating human-level AI.

Artificial neural networks (ANN), the main component of deep-learning algorithms, drew inspiration from the human brain and were meant to replicate its functions. Today, ANNs are not nearly as efficient and versatile as their biological counterparts. Nonetheless, they've yielded many important applications in fields such as computer vision, natural language processing, machine translation, and voice synthesis. And many scientific fields, including neuroscience, cognitive science, and other areas that have to do with the study of the human brain have benefited from the research in artificial general intelligence.

So if history is any guide, the pursuit of artificial general intelligence will yield many benefits for humanity. Undoubtedly, we'll encounter many more hurdles, and we might never get to the finish line. But even if we never reach the stars, the journey will be rewarding.

More here:

AI Is Hard But Worth the Investment - PCMag.com