What's Next for AI in 2021? Here's What Experts Think

Over the past decade, artificial intelligence (AI) had made advances and huge leaps, but it's clear that it still has a long way to go starting this 2021.

Before 2021, AI already grew in its capacity to collect, store and process large amounts of data.

Now, AI's deep learning function serves a key component in many of today's' everyday applications, but it's not the finish line.

The topic of AI in 2021 and beyond was discusses in "AI debate 2: Moving AI forward: An interdisciplinary approach," where scientists from different fields were in attendance.

Learning Algorithms Could Combine With Rule-Based Software

During the debate, cognitive scientist Gary Marcus said the current standing of deep learning in AI had its own shortcomings, reported Venture Beat.

There were excessive data requirements, low capacity for transferring knowledge to other domains, opacity, and a lack of reasoning and knowledge representation in deep learning.

As a critic of the deep learning-only approach, he published a paper early last year to suggest a hybrid approach where there would be both learning algorithms and rules-based software.

Other scientists in the gathering concurred; they said hybrid approaches could resolve some problems with deep learning.

Computer scientist Luis Lamb also suggested a neural-symbolic AI which will be based on both logical formalization and machine learning.

"...we can also effectively reform neural learning using deep learning machinery," Lamb said.

Beefing Up AI Automated Governance

One key learning from automated governance and controlling AI applications is that the presence of rogue AI is intolerable.

But with the privacy issues that came on various social media companies in 2020, it's no surprise if AI starts moving towards stricter privacy policing.

So enterprises would likely start strengthening their AI governance by instituting strong model assurance, reported Information Week.

This will be important in making sure that AI apps continue to do their intended functions without venturing too much on a person's privacy, demographic biases and other adverse algorithmic issues.

Stopping The Next Pandemic Before It Starts

It could be a centerpiece in tech as people move forward with the "new normal" post-pandemic.

Information Week also predicted that AI would be the "intelligent nucleus" that people can use for automated, robotic and contactless processes that could prevent future outbreaks.

Forbes also noted that AI is geared mostly towards making predictions, so systems that could accurately predict any future outbreaks may not be that far off.

Research on this has been going on for some time now and some of the systems in place during the current outbreak was actually made possible through AI.

Exploring Reinforcement Learning

Computer scientists Richard Sutton said that for the most part, there were shortcomings in "computational theory" in AI, which defines what goal and information processing system seeks and why it seeks that goal.

"Reinforcement learning is the first computational theory of intelligence," Sutton said in a Venture Beat article.

Reinforcement learning is about giving agents the basic rules of an environment and leaving it to maximize the reward.

Sutton recognized that there are other candidates that should be explored about the world of AI but reinforcement learning, for him, was a standout.