On Thursday evening, scholars from the University of Virginia and leaders well-versed in the world of artificial intelligence gathered on Grounds. Their focus was specifically on the current state of AI, the route going forward, as well as the potential risks that come along with using this transformative type of technology.
Chris Krebs, former Director of the federal Department of Homeland Security’s Cybersecurity and Infrastructure Security Agency, moderated the discussion. Panelists included Dr. Madhur Behl, an associate professor in the Computer Science and Systems and Information Engineering departments at UVA, Ferdinando Fioretto, an assistant professor in the Computer Science department, Katie Harbath, former director of public policy at Facebook, and Yen-Ling Kuo, an assistant professor in Computer Science.
How did artificial intelligence arise?
Starting off the evening, Krebs refuted the idea that artificial intelligence is a recent development. He then asked panelists for their opinions on where the technology came from, and how it has progressed to the state in which it is being used for now.
Dr. Madhur Behl tackled the question first, discussing biological findings of researchers who found that biological neurons in the brain exhibit “spiking behavior” when the brain is tasked with active activity. Years later, the concept of the artificial neuron arose, which Behl described as “a mathematical attempt to replicate the spiking behavior of a biological neuron and mimic…[the] network of the brain.” From there, the concept of an artificial neural network was soon developed, which can be defined as a “method in artificial intelligence that teaches computers to process data in a way that is inspired by the human brain.” Amazon Web Services further explains the concept as “a type of machine learning process, called deep learning, that uses interconnected nodes or neurons in a layered structure that resembles the human brain.”
Following these research findings, the rise of the internet also created new notions of how to “transcribe humankind,” said Behl. Subsequent to the internet, he added, the introduction of GPUs (graphics processing units) transformed efficiency in every aspect of our society: transportation, sustainability, manufacturing, the supply chain, etc.
Bringing the audience to speed on the technicalities of modern day artificial intelligence, and more specifically ChatGPT, Behl described ChatGPT as a special type of neural network (GPT specifically stands for “generative pre-trained transformer”). He compared the technology to that of a child learning from their parents. Children may learn a nursery rhyme from their parents, and from then on be able to recite it over and over; however, how in depth are they truly processing what they have learned? Do they understand it? The same goes for ChatGPT. The system simply has gotten extraordinary at mimicking and reciting patterns of the human language.
What are the critical elements that accelerated AI for the United States?
Ferdinando Fioretto spoke to the audience on various factors that he believed influenced the progression of AI in America, all of which revolve around our capabilities of developing and advancing the way we use and share data. Various models (neural networks) have been trained to recognize certain patterns – all because of our ability to send and transform data into a language for the technology to process. However, operations become more complicated when handling larger quantities of data. This is where GPUs make the transformation of data more efficient- speeding up the development of new algorithms and patterns.
What is the path forward for artificial intelligence?
In answering Krebs’ question, Yen-Ling Kuo argued that moving forward, there are still a variety of unsolved challenges. Telling AI to write your speech is different from having the intelligence to perform particular tasks, such as playing with toys.
For those that fear a future where artificial intelligence takes over the world, Behl shared an encouraging perspective with the audience. AI has its limitations as the technology lacks the necessary perspective of how the real world functions. Kids have the understanding and knowledge that if you push an object, it will fall over and/or break, explained Behl. However, “AI is not even close to having that level of human intelligence.”
The technology also is not sophisticated enough to observe interactions and learn from these actions like humans can. As an example to portray the limits of AI, Behl stated: “I see two people dance, [and] I don’t have to dance before I can just figure out what they’re doing without even interacting with them,” This is one reason as to why he is “cautious” over recent widespread claims of superhuman artificial intelligence.
What are the risks and advantages of democratizing AI so it can be accessible to anyone?
When questioned on her perspective on how to approach the pros and cons of AI, Katie Harbath concisely answered: “panic responsibly.” With extensive experience in campaigns and elections, Harbath discussed just how valuable the technology can be in enhancing efficiency in campaigns – a field of work that has already embraced AI for all its advantages. Whether it be for writing emails and press releases, or curating social media posts, artificial intelligence is going to revolutionize the campaigns world by saving a significant amount of time and money. The private sector will also utilize AI to maximize productivity. However, Katie emphasized the importance of diligence in operating this technology.
“You can’t put the genie back in the bottle,” Harbath stressed. There has to be a balance between discovering new uses for artificial intelligence, but also promoting transparency and regulation so as to not enable bad actors in using the technology destructively.
However, regulation will be tricky. 65 national elections across 54 different countries will take place in 2024. This presents a whole new challenge for regulating AI and how it will be used by campaigns, but also in operating elections. Europe has new regulation with the Digital Services Act. However, Harbath doesn’t see any regulatory reforms coming from Congress any time soon. Meanwhile, the Federal Trade Commission is attempting to tackle regulation on their end. Therefore, the question arises of who has the ability to regulate what- and how?
What would potential government intervention surrounding artificial intelligence look like?
With differing state laws that will come into place and inevitable Supreme Court cases, Harbath believes Congress will have to act after the 2024 election. However, the former public policy director believes AI regulation will have to be focused around transparency and oversight, as if content were to be regulated, First Amendment rights could be brought into question.