peshkov -

The power and limitations of enterprise AI

A panel at CES 2021, held virtually this year, discusses the areas in which modern-day AI and automation shine, and where they still struggle.

Most companies, whether they realize it or not, are likely using some form of AI. Machine learning, deep learning, robotic process automation and other forms of AI are baked into hardware and software, enabling users to optimize and streamline their workflows.

Modern-day AI works well at taking over repetitive tasks, handling them behind the scenes to make software interactions more effortless. It still struggles with higher-level thinking, however.

Ubiquitous AI

"AI is all around," said Bridget Karlin, global managing director, CTO and vice president of IBM's Global Technology Services business.

Speaking at a panel on AI at the 2021 Consumer Electronic Show (CES), Karlin pointed out that the technology is widely used in most industries, including healthcare, supply chain and education. Still, she said, people are only "at the tip of the iceberg" when it comes to AI, with advances in software, increased computing power and access to more data driving AI and accelerating its adoption.

Karlin described AI as able to do three things: predict outcomes, improve automation and optimize cost, performance and user experience.

Summing it up, panel speaker Kevin Guo, CEO at machine learning startup Hive, explained that AI models exist to reduce low-level, repetitive labor that humans have to do.

Machine learning models have very concrete parameters, so they are well-suited for handling repetitive processes, he said.

AI is all around.
Bridget KarlinGlobal managing director, CTO and VP, IBM Global Technology Services

Eric Cornelius, chief product architect at BlackBerry, added that "not every problem is best solved by the use of artificial intelligence, but it seems that a growing number of problems are, at least, able to be solved by AI."

In the healthcare industry, for example, AI can quickly and accurately view medical images to rule out certain conditions. IT security, in which machine learning and deep learning algorithms can be programmed to block attacks automatically, can reap almost unlimited benefits from this technology, Cornelius said.

Automating and optimizing workflows, AI lets human workers free up their time to think more creatively, he continued.

Providing an example, Cornelius said, "AI is never going to build a bridge."

"AI may build really great plans; it may give us the best blueprints for a bridge and do all the soil analytics; it might even file the paperwork for us, but it's not going to build a bridge," he said. "AI will always be there to supplement human ambition."

Power of AI panel during CES 2021
Participants on the panel 'The Power of AI' at CES 2021 talk about the uses and limitations of AI.

Doubts on performance

Still, while AI shines at handling these low-level tasks, it's unclear when, or even if, AI could perform high-level thinking.

Take fake news, for example, Guo said. Artificial intelligence performs well at finding explicitly wrong things, such as nudity and hate speech, but has difficulty discerning the intent behind longer pieces of writing.

Cornelius said he doubts if AI will ever eliminate "the really big problems." Take malware, he said. The models have proven incredibly good at detecting common pieces of malware, but it's difficult, if not impossible, for models to keep up with the constant flow of new malware and threats.

The threat industry is a $20 billion market, and a massive amount of people are constantly working on methods to scam people and infiltrate networks, he said.

"As long as there's motivation by smart humans to continue pushing the playing field, AI -- by definition, because it's built by people -- will always be built to solve problems that exist," Cornelius said.

Current AI models are also highly vulnerable to bias. While unintentional, bias, introduced through skewed data, can dramatically affect predictions and cause models to malfunction.

"Applying AI is hard; it's really hard work," Karlin said. Besides getting access to massive data sets and forming and training knowledge bases, enterprises must be aware of biases and prejudices in their models and data and work to eliminate them.

"Each of us is a steward of the technology," she said of technology vendors. 

Creating fair, unbiased AI that is properly calibrated and programmed will remain the responsibility of the vendors who create them, Karlin said.

The panel "The Power of AI" was held Tuesday during the virtual CES 2021 conference and featured Karlin, Cornelius, Guo, and moderator Jeremy Kaplan, editor in chief at Digital Trends.

Next Steps

A look at AI trends and bias in AI algorithms

Language models and the metaverse top AI stories of 2021

Dig Deeper on AI infrastructure

Business Analytics
Data Management