The rapidly evolving space of artificial intelligence has received growing interest, investment and scrutiny.
With all the hype surrounding AI’s future and what it might mean for society, Jeremy Kedziora, Ph.D., the PieperPower endowed chair in artificial intelligence at the Milwaukee School of Engineering, has a few ideas for what’s to come – including the emergence of agentic systems and integration into daily life. Kedziora was a speaker at BizTimes Media’s annual Economic Trends event on Jan. 23 at the Italian Community Center in Milwaukee.
When looking to the future of AI, Kedziora said he speculates there may be a “plateauing” of large language models. Large language models like OpenAI’s Chat GPT or Meta’s Llama require a tremendous amount of data in order to learn and improve.
“There’s a reason why this AI stuff is happening now and didn’t happen 30 years ago,” Kedziora said. “The reason it’s happening now is because the internet kind of digitized all this information and made it accessible. So, you can have somebody web scrape and collect a billion documents, and you have Google’s attempt to digitize books and that sort of thing. It renders those things available for consumption by an AI model so that it can learn.”
But the data available to the AI models is finite, and this will eventually prevent substantial increases in the performance of large language models, Kedziora said.
“The thing is that we’re kind of running out of information,” Kedziora said. “There’s only one internet, and it’s being mined out. If you don’t have any new information, you can’t really encode any new capabilities. So, my sense is that there’s going to be a plateau in large language model performance.”
If Kedziora’s theory proves true, there will likely be a focus over the next few years to increase the efficiency of AI systems in terms of energy consumption and cost, he said. A trend to create agentic systems, or “effective, focused agents based on these large language models that have already been built,” may emerge, Kedziora said.
“I think they’re also going to turn towards trying to use existing systems to solve problems,” Kedziora said. “If the models aren’t going to get substantially better as general purpose models, I think people will turn towards trying to adapt them for specific use cases and make them really good at specific use cases.”
Integration into daily life
As AI becomes normalized, it will become a tool for productivity, similar to when computers were brought into the workplace, Kedziora said.
It’s a common concern that a widespread use of AI could cause people to lose their jobs, but some research suggests these developments in AI will have a limited impact on the U.S. labor market, Kedziora said. Economists find that “a relatively small number of jobs” could be heavily affected, he said.
“No one’s going to be replaced by an AI, but somebody who uses AI as an effective partner might out-compete somebody who doesn’t use AI at all,” Kedziora said.
There’s a growing pool of contradictory research seeking to predict how AI may affect the labor market.
In July, the Council of Economic Advisors, a U.S. agency that advises the president, released a report stating that AI won’t negatively affect the labor market. However, the report also found about 10% of employment could be vulnerable to AI implementation.
One professor and economist at the Massachusetts Institute of Technology predicts AI will have a sweeping impact on office jobs focused on tasks like “data summary, visual matching, pattern recognition,” noting those kinds of jobs make up about 5% of the economy, according to a December article by MIT News.
A separate MIT research group published a paper predicting that the high cost of implementing AI would create a gradual, limited disruption to employment.
For the past two years, MSOE has hosted hackathons allowing students to develop AI solutions to support a nonprofit organization. In November, MSOE partnered with Discovery World and had students working to use AI to help make the nonprofit’s exhibits more dynamic or expressive, Kedziora said.
“I think this is the kind of thing that you’re going to start seeing all over the place, this kind of heavy integration of these AI capabilities in your daily life,” Kedziora said. “So, what they did was, they basically wanted to make, like, ‘Pokémon Go’ for Discovery World. And I think you’re going to start seeing that kind of stuff all over the place as really creative people start to engage with and normalize and integrate AI into their lives.”
Navigating the hype
With the emerging normalization of AI, new developments frequent the news cycle. Kedziora thinks everyone should take AI headlines “with a big grain of salt.”
“The basic paradigm of optimizing loss functions to make these kinds of agents has been around for 200 years, and it’s been at least 12 years since the deep learning revolution,” Kedziora said. “So, it’s not really all that new, and there’s a lot of hype, and people have interests.”
New scientific discoveries often happen in academia, but it’s different for the world of computer science, where the greatest breakthroughs are likely to come out of corporate research labs at big tech companies, Kedziora said.
“I think what you have to remember when folks from big tech and corporate America talk about this stuff is, they have interests,” Kedziora said. “They want to promote the interests of their company. They want to promote a certain regulatory environment that is favorable to their goals. And a lot of what they say is aimed at that kind of stuff, busting up their prestige, their image.”
