What can we find out about human intelligence by studying how machines “think”? Can we understand ourselves higher by higher understanding the factitious intelligence systems which might be becoming an increasingly necessary a part of our on a regular basis lives?
These questions could also be deeply philosophical, but for Phillip Isola, finding the answers is as much about calculation because it is about reflection.
Isola, the brand new associate professor within the Department of Electrical Engineering and Computer Science (EECS), studies the basic mechanisms of human-like intelligence from a computational perspective.
While the general goal is knowing intelligence, his work primarily focuses on computer vision and machine learning. Isola is especially excited about exploring how intelligence emerges in AI models, how these models learn to represent the world around them, and what their “brains” share with the brains of their human creators.
“I see that every one several types of intelligence have rather a lot in common, and I would really like to know those similarities. What do all animals, humans and AIs have in common?” says Isola, who can be a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL).
For Isola, a greater scientific understanding of the intelligence that AI agents possess will help the world safely and effectively integrate them into society and maximize their potential to profit humanity.
ask questions
Isola began enthusiastic about scientific questions at a young age.
Growing up in San Francisco, he and his father often hiked along the Northern California coast or camped around Point Reyes and within the hills of Marin County.
He was fascinated by geological processes and sometimes wondered how nature worked. At school, Isola was driven by an insatiable curiosity, and although he was drawn to technical subjects akin to math and science, there have been no limits to what he desired to learn.
Unsure of what to check as an undergraduate at Yale University, Isola dabbled until he got here across cognitive science.
“My previous interest was nature – how the world works. But then I spotted that the brain was much more interesting and sophisticated than the formation of planets. Now I desired to know what drives us,” he says.
As a first-year student, he began working within the laboratory of his cognitive science professor and future mentor, Brian Scholl, a member of the Yale Department of Psychology. He remained on this laboratory throughout his studies.
After a 12 months of working at an indie video game company with some childhood friends, Isola was able to delve back into the complex world of the human brain. He enrolled within the graduate program in Brain and Cognitive Sciences at MIT.
“In graduate school, I felt like I had finally found my place. I had a variety of great experiences at Yale and at other times in my life, but after I got to MIT, I spotted that this was the work that I actually loved and that these were the individuals who thought similarly to me,” he says.
Isola credits his doctoral advisor, Ted Adelson, the John and Dorothy Wilson Professor of Vision Science, as a significant influence on his future path. He was inspired by Adelson's concentrate on understanding fundamental principles and not only chasing latest technical benchmarks, that are formalized tests to measure a system's performance.
A computational perspective
At MIT, Isola's research shifted toward computer science and artificial intelligence.
“I still loved all these questions from cognitive science, but I felt that I could make greater progress on a few of these questions if I approached them from a purely computational perspective,” he says.
His dissertation focused on perceptual grouping, which is the mechanisms that humans and machines use to arrange individual parts of a picture as a single, coherent object.
If machines can learn perceptual groupings on their very own, this might enable AI systems to acknowledge objects without human intervention. This sort of self-supervised learning has applications in areas akin to autonomous vehicles, medical imaging, robotics and automatic language translation.
After graduating from MIT, Isola pursued postdoctoral studies on the University of California, Berkeley, broadening his perspectives by working in a lab focused exclusively on computer science.
“This experience helped make my work way more impactful because I learned to balance understanding basic, abstract principles of intelligence with pursuing some more concrete benchmarks,” recalls Isola.
At Berkeley, he developed frameworks for image-to-image translation, an early type of a generative AI model that would, for instance, convert a sketch right into a photographic image or a black-and-white photo right into a color photo.
He entered the educational job market and accepted a school position at MIT, but Isola delay work for a 12 months to work at a then-small startup called OpenAI.
“It was a nonprofit, and I liked the idealistic mission on the time. They were really good at reinforcement learning, and I believed that gave the impression of a vital topic to learn more about,” he says.
He enjoyed working in a lab with a lot scientific freedom, but after a 12 months, Isola was able to return to MIT and begin his own research group.
Study of human-like intelligence
He immediately liked running a research laboratory.
“I actually love the early stages of an idea. I feel prefer it's a form of startup incubator where I can always do latest things and learn latest things,” he says.
Building on his interest in cognitive science and his desire to know the human brain, his group studies the basic computations involved within the human-like intelligence that emerges in machines.
A important focus is on representation learning, i.e. the flexibility of individuals and machines to represent and perceive the sensory world around them.
In recent work, he and his collaborators found that the many differing types of machine learning models, from LLMs to computer vision models to audio models, appear to represent the world in similar ways.
These models are designed for very different tasks, but have many similarities of their architecture. And the larger they get and the more data they use, the more similar their internal structures turn into.
This led Isola and his team to introduce the Platonic Representation Hypothesis (named after the Greek philosopher Plato), which states that the representations that every one of those models learn converge to a typical, underlying representation of reality.
“Speech, images, sound – these are all different shadows on the wall from which you’ll be able to infer that there may be some underlying physical process – some form of causal reality – on the market. If you train models on all these several types of data, they need to eventually converge to this model of the world,” says Isola.
A related area his team is studying is self-supervised learning. This involves the best way AI models learn to group related pixels in a picture or words in a sentence without having labeled examples to learn from.
Because data is dear and labels are limited, using only labeled data to coach models could compromise the performance of AI systems. In self-supervised learning, the goal is to develop models that may independently create an accurate internal representation of the world.
“If you’ll find an excellent representation of the world, that ought to make later problem solving easier,” he explains.
The focus of Isola's research is more on finding something latest and surprising than on constructing complex systems that may outperform the newest machine learning benchmarks.
While this approach has brought great success in discovering modern techniques and architectures, it implies that the work sometimes lacks a concrete end goal, which might result in challenges.
For example, if the lab focuses on trying to find unexpected results, it will possibly be difficult to maintain a team informed and secure funding, he says.
“In some ways we’re at all times at midnight. It is high-risk, high-reward work. Every from time to time we discover a kernel of truth that’s latest and surprising,” he says.
In addition to imparting knowledge, Isola can be enthusiastic about imparting knowledge to the subsequent generation of scientists and engineers. One of his favorite courses is 6.7960 (Deep Learning), which he and a number of other other MIT faculty members launched 4 years ago.
The class has seen exponential growth, from 30 students in the primary offering to over 700 this fall.
And while the recognition of AI means there isn’t any shortage of interested students, the speed at which the sector moves makes it difficult to separate the hype from truly significant advances.
“I tell students to take every thing we are saying at school with a grain of salt. Maybe in a number of years we'll tell them something different. With this course, we're really on the sting of data,” he says.
But Isola also emphasizes to students that, despite all of the hype surrounding the newest AI models, intelligent machines are much simpler than most individuals imagine.
“Human ingenuity, creativity and emotions – many individuals consider these can never be modeled. That might turn into true, but I believe intelligence is pretty easy once we understand it,” he says.
Although his current work focuses on deep learning models, Isola remains to be fascinated by the complexity of the human brain and continues to collaborate with researchers studying cognitive science.
He remained fascinated by the great thing about nature, which sparked his first interest in science.
Although he has less time for hobbies today, Isola enjoys mountain climbing and backpacking within the mountains or on Cape Cod, skiing and kayaking, or finding scenic places to remain when he travels to scientific conferences.
And as he looks forward to exploring latest questions in his lab at MIT, Isola can't help but take into consideration how the role of intelligent machines could change the course of his work.
He believes that artificial general intelligence (AGI), or the purpose where machines can learn and apply their knowledge in addition to humans, just isn’t that far-off.
“I don't think AIs will just do every thing for us and we'll enjoy life on the beach. I believe there will likely be a coexistence between intelligent machines and humans, who still have a variety of agency and control. Now I'm enthusiastic about the interesting questions and applications when that happens. How can I help the world on this post-AGI future? I don't have the answers yet, but it surely's on my mind,” he says.
