Jason Corso on artificial intelligence
The most exciting use of AI for me focuses around a better collective use of our available resources, says Prof. Corso.
When is a computer more than a machine? One person’s concept of artificial intelligence (AI) can be quite different from the next person’s. In its most basic form, artificial intelligence is a process in which machines take cues from their environment and act responsively to achieve a goal. Experts may point to certain defining features of the technology such as computer vision, language processing, and machine learning. These characteristics provide important information, but are not a requirement of AI. When push comes to shove, AI is the fruit of computer engineers’ intensive programming labor. It can be slow and tricky work, but the perks of its successes have the potential to change our lives forever.
Read on to discover the varied applications of AI through an interview with Jason Corso, a professor at the University of Michigan.
This interview comes courtesy of Online Engineering Programs.
WHAT AI CAN (AND CAN’T) DO
Artificial intelligence can do everything from play tic-tac-toe to diagnose cancer, yet it has its limitations. For decades, books and movies have warned of AI-enabled androids making independent decisions and, naturally, taking over the world. It is a fascinating notion—one respected experts like Elon Musk and Stephen Hawking have taken very seriously—but the truth is AI is a spectrum, and we are still in the relatively early days. Here are the various stages of AI, as reported by the World Economic Forum (WEF):
- Automated intelligence: Systems that do labor-intensive tasks requiring some degree of discernment, like sorting recyclables.
- Assistant intelligence: Systems that analyze data and note patterns that can help humans make better decisions.
- Augmented intelligence: Simulators and other systems that can process historic and real-time data to make predictions.
- Autonomous or general intelligence: Systems that can process data and make decisions without any human intervention.
The fact that AI innovators have not yet reached the point of autonomous or general intelligence does not mean the technology isn’t doing remarkable things. As Facebook’s chief technology officer, Mike Schroepfer, told MIT’s Technology Review, AI technology could solve problems “that scale to the whole planet.”
THE FOURTH INDUSTRIAL REVOLUTION: HOW AI IMPROVES THE WORLD
Artificial intelligence is not yet advanced enough to supercede humans in any broad capacity, but when programmed to do a task, it can often do it faster and more accurately. Sometimes this capability just makes our lives easier. Sometimes it saves them. The World Economic Forum calls this growing capacity for solving the world’s problems with technology the “fourth industrial revolution.”
“Oh there is just so much potential for the use of well-structured AI (or multivariate optimization problems),” said the University of Michigan’s Dr. Jason Carso, who researches high-level computer vision and AI’s physical engagement with the environment. “Popular possibilities like in-home robots that clear the table, healthcare robots that provide optimal bedside care, or autonomous vehicles that increase mobility are all well and good. But, the most exciting use of AI for me focuses around a better collective use of our available resources.”
Such innovations are just the tip of the metaphorical iceberg. Here are a few of the ways AI is already making a difference.
Healthcare
Artificial intelligence’s capacity to rapidly analyze data—including visual data—has made it a game-changer in the healthcare field where, according to Big Think, it has been proven to diagnose disease faster than any human doctor. Medical AI can also offer novel therapy options, minimize human error, and take over repetitive tasks that clog up the system, freeing doctors and nurses to focus more on their patients. In terms of medical research, AI could smooth the drug development selection process to ensure scientists are studying those that show the most promise. It can also help identify previously unexplored treatments and therapies.
Climate & Severe Weather Response
According to a 2018 report from the WEF, artificial intelligence is already improving air quality and reducing global warming using tech such as smart energy grids, smart traffic lights, AI-enabled electric cars, and real-time air pollution simulation. It also mitigates some of the effects of climate change by predicting extreme weather and mapping disaster risks. When disaster does strike, AI can predict and monitor conditions in real time to coordinate emergency response efforts and minimize damage. Some experts predict AI will eventually extract carbon from the air and repurpose in an environmentally viable way.
Protecting the Environment
There are several ways researchers use AI to protect our environment, whether it’s monitoring ecosystems and animal migration patterns or analyzing aquaculture to prevent overfishing. We can use AI-controlled drones to monitor riverways and simulations to predict drought and wildfire risk. The WEF proposes that someday scientists may create AI-enabled pollinating drones or robotic fish that fight ocean pollution.
Food Production
By analyzing sensor data (e.g., soil pH, moisture) and reviewing historical data, AI has the power to improve crop yields and minimize loss, thereby reducing global hunger. According to the WEF, farmers can use automated machine decision-making to identify disease early, adjust plant and livestock nutrition for optimal results, and minimize the use of pesticides that harm the environment through our waterways. The group predicts that these capabilities, combined with robot labor, could eventually automate the entire food-production process from beginning to end.
Resource Allocation & Security
Some experts suggest AI has the potential to address some of the most critical issues faced across the globe today, including those that affect the most basic means of survival.
“As we all know, resources—money, water, shelter—are not always abundantly available to all people; globally, many people dedicate significant time, money and effort to bring resources to people in need. However, perhaps they could do better, with appropriate modeling and intelligent systems,” said Dr. Corso. “The future of AI will enable new abilities to understand resource usage and need, and allot for those available resources to be better distributed.”
DEEP REINFORCEMENT LEARNING: THE ANSWER TO ALL OUR PROBLEMS?
Perhaps the most significant promises of AI is only just emerging: deep reinforcement learning. Deep reinforcement learning refers to a machine’s ability to learn from data and use it to create new, better algorithms without human intervention.
The machines still rely on people to begin the process and clarify the data; they are not autonomous, but can make themselves more efficient. They do this by creating what some experts call “neural networks.” Developing these networks requires a lot of time and intense coding, but as the tech advances, experts from Google, IBM, the WEF and others believe it can be used to solve virtually any world problem. It will be a while before we reach that point, but Google’s Deepmind AlphaGo Zero is proof we are moving in the right direction.
WHERE DO WE GO FROM HERE? AI RISKS & LIMITATIONS
It is difficult to consider a concept like deep reinforcement learning without imagining that dystopian, robot-ruled world imagined by by science fiction gurus. Respected thought leaders from all across the globe warn us to be cautious with a tool as powerful as artificial intelligence, but as the Future of Life Institute notes, AI does not need a body to be harmful. Machines could be programmed (or hacked) to harm others, or they could simply cause unanticipated harm while striving to achieve a goal. There could also be a damaging mistake in the data. Initiatives like OpenAI are already working to ensure a benevolent AI, or to at least minimize its risk.
Science Alert’s Vyacheslav Polonski notes that there are other, more practical limitations on AI that have nothing to do with ethics concerns. For starters, neural networking is exceedingly complex and relies on a great deal of data which is something the public sector is just not able to provide. According to Polonski, most data in the U.S. is stored in offline archives spread across the globe, and each requires special permission to be accessed.
“General intelligence is complex and is a quagmire of possibility and uncertainty,” said Dr. Corso. “What are the limitations? Everything from ‘systems’ aspects like scalability to be able to manipulate large amounts of real data fast and robust enough to be useful; to ‘intelligence’ aspects like the fact that the world we live in is complex and has ridiculous variability.”
Another key limiting factor: we simply do not have enough human talent to reap the full benefits of machine intelligence. College graduates with the right training are extremely valuable in an industry PwC projects will grow from $1.4 billion to $59.8 billion between 2016 and 2025.
JOIN THE AI REVOLUTION
Some would argue that an artificial intelligence professional is part computer programmer and part psychologist—or perhaps philosopher. A great deal of their work is based on data analysis and the creation of algorithms that support machine learning, so it is not uncommon for AI experts to study computer programming, data science, or statistics. Some colleges offer programs with special machine learning tracks, but the field is still relatively new and untapped.
Many AI degree programs include coursework in computer languages, statistical analyses, data architecture, cloud computing and, increasingly, ethics. Note that these programs are not to be confused with business intelligence or data analytics programs that focus more on data analysis than machine learning.
“The bland advice is that math and analytical thinking are critical; that hard work is an expectation,” Dr. Corso counsels anyone considering a career in AI. “Perhaps the hard part is that AI research has the potential to create quite new behaviors that will impact society in unexpected ways. Making an effort to understand these impacts is a responsibility.”
Dr. Jason Corso is an associate professor of electrical engineering and computer science at the University of Michigan. His primary research focuses on high-level computer vision, with an emphasis on video understanding for physical engagement with the environment by agents (e.g., robots). He has published on a large number of problems in this space, but his primary areas are video segmentation and video action recognition. His company, Voxel51 LLC, seeks to expose video understanding in a commercial setting.