As more and more people and organizations turn to AI (AI) for ideas and answers, the leading global AI experts at Ontario Institute of Technology are raising a warning flag for those who assume that the information provided by AI is always reliable and accurate.
Dr. Peter Lewis, Canadian Research Chair of Trustworthy AI, says there could be a quiet AI revolution happening.
Q. At the heart of this shift is questions that continue to challenge professionals such as engineers, ethicists, clinicians, lawyers and educators.
“With current AI systems, we really can't do it. As humans, we know that we're not completely perfect. We expect that level of humility and recognition from each other. So we might expect AI to believe that this same kind of trait can be trusted.
Q. What is the risk if AI doesn't know the answer, but still is creating it? Does AI have essentially human-like urgency and reasoning? Don't you know when you're not confident in the answer?
“Errors have serious consequences when large language models (the driving force behind tools such as ChatGPT) perform tasks, especially in high impact domains such as finance, legal and healthcare. Therefore, important challenges arise. These platforms can arrive in a very different way than humans. AI systems often show that they are confident in what they are telling us, indicating that they are unable to judge their abilities very well, rather than minimizing the uncertainty of AI.
Q. Is it an ethical obligation for AI to be transparent if it is not sure about the answer?
“Yes. It's not just a technical feature. It's a guardian. Rather than aiming for non-fallability, AI systems need to reflexively recognize their own blind spots. They can recommend human surveillance if necessary.
Q. If this form of AI is known to be 100% time accuracy is an unattainable goal, how do you adjust a legitimate level of trust and encourage healthy skepticism in high-stakes AI applications?
“AI is right, but 100% certainty is an unattainable goal. It's a misleading expectation that a machine should always be right. Working with AI experts and users, we need to understand and expose how uncertainty appears in the AI model, and we will tailor these to domain-specific (clinical) risk factors.
Q. Is there any hope for a future AI system that is better at evaluating yourself and reporting uncertainty more accurately?
“Yes, there's hope, so that's a great question. But it's not something that comes down to by building a larger neural network model. In our study, we essentially “package” the engine of AI systems with a reflection layer that evaluates the output of large language models that violate various important criteria, rather than oppose various key criteria. Also in such considerations, such as whether it is acting in a socially expected and responsible way. ”
Q. In addition to ethical considerations regarding AI reliability, is there a responsibility that makes AI fair and accessible?
“Absolutely, this is an emerging area with a lot of work to do. Many AI systems work as what is known as “black boxes.” Here we don't know how they make decisions, what potential bias factors will drive output. Loss Communities, particularly the National Institute of Visually Impaired Canada (CNIB), has developed an accessible version of an explanatory AI tool for people with disabilities, demonstrating why the current state of the technology is not appropriate. ”
Q. AI has many cheerleaders, but many destinies too. When it comes to AI, do people have to be “rational skeptics”?
“AI is a largely polarizing issue. Some say that all-in people are revolutionary. On the other side of the spectrum, there are people who firmly argue that AI is the “end of humanity.” Sometimes stories of right and revolutionary transformation are primarily hype.
– Dr. Peter Lewis is a trusted associate professor of artificial intelligence and Canada's research committee chair Faculty of Business and Information Technology At Ontario Institute of Technology.
Dr. Peter Lewis, Canadian Research Chair of Trusted Artificial Intelligence at Ontario Institute of Technology.
“>
Dr. Peter Lewis, Canadian Research Chair of Trusted Artificial Intelligence at Ontario Institute of Technology.
“>
Dr. Peter Lewis, Canadian Research Chair of Trusted Artificial Intelligence at Ontario Institute of Technology.
“>
