AI Has 'Dark' Side but Also Offers Benefits for Telecom Networks: Rosenworcel

Rosenworcel announced FCC commissioners will vote Aug. 3 on a notice of inquiry on using AI and other tools to better understand the actual usage of nonfederal spectrum bands (see 2307130025).

“So much of the news about AI, let’s acknowledge, it’s been pretty dark ,” Rosenworcel said: “How do we rein in this technology? What does it mean for the future of work when we have intelligent machines? What will it mean for democracy and elections? What happens when AI models inherit the prejudices of the systems that they are trained on, and they get to determine who gets a job and who gets a loan?” Some warn of an existential threat to humanity, she said.

However, she said, AI offers the promise of making telecom networks more efficient. “The day is not all that far off when we will be able to use this technology to help self-configure, self-optimize and self-heal” networks, she said.

“Where the United States has succeeded in technological development, it has done so through a mindful attempt to cultivate and potentiate innovation,” Simington said. “The American success stories contrast with the stories of Soviet, and more recently European, technological stagnation,” he said.

U.S. regulation similar to the EU’s general data protection regulation isn’t the answer, Simington said. GDPR is “an impossibly complex, pan-European regulatory and enforcement scheme that varies by member state implementation, the requirements of which are brain-curdling and chill the behavior of even the most intrepid entrepreneurs,” he said. A U.S. version of GDPR “would be devastating for America’s leading position in the global order,” he said.

“We have to be optimistic,” said NSF Director Sethuraman Panchanathan. “We have to identify the problems but more importantly we have to find how to solve those problems and such solutions don't happen unless we bring like-minded people [together], who are all interested in solving the problem, from different sectors.”

Panchanathan and other speakers noted AI may have appeared to come out of nowhere, but it has been decades in development. AI required more than 50 years of “sustained investment” in research, he said. “We face some challenges” that “require us to go back and think deep and hard about what more explorations and discoveries need to happen,” he said.

Congress, the administration and the military are already “deeply engaged” on AI, said FCC Commissioner Geoffrey Starks. AI has the potential “to impact, if not transform, nearly every aspect of American life,” he said. One potential use will be to help people with disabilities “benefit from modern-day digital communications,” he said.

Risks Remain

AI also has the potential to “amplify data-driven bias and discrimination,” Starks warned. Equity issues “have to be fundamental as we consider technological advancement,” he said. AI also poses new security threats to U.S. networks, he said: “AI has raised the stakes of defending our networks … not by a little, but truly by a mile.”

The workshop also included two panels of speakers on AI challenges and how it’s already being used in the telecom industry.

AI today presents a “freak-out moment,” said Public Knowledge Senior Vice President Harold Feld. AI has been around for a long time, mostly as a “background conversation,” he said: “It wasn’t until people were like, ‘Oh, my God, my computer is hitting on me’ that suddenly this became a major discussion point here in Washington. … Panic is bad when it comes to formulating policy.”

AI isn’t new, agreed Lisa Guess, Ericsson/Cradlepoint senior vice president-global sales engineering, who also co-chairs an FCC Technological Advisory Council working group on AI. Companies “develop systems that can learn,” Guess said. “They have contained variables, and they are doing a very specific thing -- this has been going on for a long time,” she said. That led to “broader, more general AI, super AI, if you would,” she said. Networks are changing, and are now “closely intertwining compute, storage and processing power all the way out to the edges,” she said.

AI will be needed to design the intelligent networks of the future, “networks that are self-healing, that are robust,” said Ness Shroff, director-NSF AI Institute and engineering professor at Ohio State University.

Controlling and designing next-generation networks requires AI that takes into account “the very specific constraints that networks have, things like non-stationarity, things like hard constraints,” Shroff said: “If you basically tell a network to transmit at a certain power level, and it doesn’t have the battery capacity to do it, no matter what your AI algorithm says” it’s not going to happen. Many of the devices consumers use already use AI, but “in a stand-alone way,” he said. AI can be used to better manage the radio access network, to better understand the traffic and user needs, he said. Billions of dollars have already been invested in open RAN, which incorporates AI, he said.