From chat engines and generative apps to self-driving cars, technologies powered by artificial intelligence continue to transform our lives in new ways. But how do we create AI that works for society without disenfranchising some of us? How can we be sure these innovations are fair and trustworthy?
These are the questions we asked at Mozilla’s Responsible AI Challenge event in San Francisco last May. The industry’s brightest thinkers, ethicists, engineers, and builders have come together to celebrate the potential of AI while considering the responsibilities that come with its immense capabilities. Here’s what we learned from this event:
1. Accessibility improves AI
Dr. Margaret Mitchell, a leading AI ethicist, said that in 2011, while working on technology to generate image descriptions, she and her colleagues encountered what they called the “everything is great” problem. The system learned from the photos and captions people shared on social media, and the content was generally positive. So even though the images depicted tragic events, the model continued to generate positive descriptions.
People share “beautiful sunsets, amazing views, amazing skies…”. [So when the computer] Upon seeing this terrifying explosion, the reaction is “this is awesome”. And in that moment we understood how the data we use directly affects the learning of the model,” said Dr. Mitchell. “We are seeing moments where masses of people are potentially injured, but computers don’t understand what mortality is. No, I see purples and pinks in the sky.”
Dr. Mitchell concluded that given the potential of connecting vision to language in assistive technology, the model prioritizes data that does not match the needs of the visually impaired. Blind people need context and actionable information about images rather than explanations based solely on available data that can be distorted.
Since then, AI systems such as image generation have advanced. Lesson learned: Accessibility not only ensures equitable access to technological advances, it also encourages us to innovate so that we can create better systems for everyone.
2. We need (human) thinkers to address the limitations of AI
Despite years of advances in artificial intelligence, Dr. Gary Marcus, a prominent academic who recently testified before the US Senate on AI, argues that deep learning systems still face major challenges. For example, it cannot be incrementally updated with the addition of new knowledge. This can lead to a lot of misinformation.
“We have defamation laws, but what if someone creates a billion perfectly-formed pieces of false information a day?” Dr. Marcus said. “Do you want to treat it like free speech? Or is it more like commercial speech? Should it be treated differently? We just don’t have the law to do that yet.”
That’s where professionals who pay attention to the constant changes in this field come into play.
“The choices we make now will shape the next century,” Dr. Marcus said. “Without scientists and ethicists at the table, our prospects are not great. We cannot afford not to regulate. yeah.”
3. Experts optimistic about AI capabilities, including health car advancese-field
Kevin Roose – technology columnist, podcaster and author – has covered AI for over a decade. He made a few predictions: AI is relatively new and lawmakers’ limited understanding of the technology makes meaningful regulation unlikely by 2030, Ruth said. said Mr. He expects model performance to improve and researchers and companies using AI models to build products will drive more secure systems.
He also predicted great potential for AI, especially in advancing drug development.
“I think there’s about a 30% chance that drugs discovered using AI will reduce deaths from the top 10 cancers by 50% by 2030…I’m not a doctor, but this is about people.” It’s kind of like thinking,” he said. Please advise an expert in this field. i believe in them “
4. AI systems may be able to work together more effectively than humans.
Ruth recounted a conversation with a source who noted a major advantage of AI systems over humans: their tendency to share knowledge.
This expert explained that when one node in a neural network establishes a connection, that information is propagated to all other nodes in the network. For example, in a fleet of self-driving cars, when one vehicle learns to identify new obstacles, it shares that knowledge with all other vehicles in the fleet, including vehicles from different manufacturers. People, on the other hand, don’t have this tendency and often keep research and data to themselves.
Ruth argued that humans can learn from machines in this aspect. “If we want a chance to compete, survive and thrive in this new world of AI, we must be able to do what machines do and share with each other everything we learn. think. “
