Despite a few setbacks—such as the demise of Microsoft’s 2016 chatbot Tay, which was killed off after it started tweeting racist and sexist messages—AI is racing ahead full steam. As its computing power increases and companies invest more resources into AI development, it’s clear the technology could one day reshape the world.
But how do we prevent that from happening? Some fear that super-intelligent machines will wreak havoc on financial markets, torpedo the food supply chain, and pit human beings against each other through social media. Others worry that personalized news feeds curated by AI systems will limit exposure to diverse viewpoints, contributing to the spread of misinformation or political polarization. In an op-ed for Bloomberg, Evgeny Morozov warned that the current trajectory of AI is “catastrophic for humanity.” It’s possible that these fears are exaggerated, but it also seems plausible that the most powerful systems will be built by private companies and that these companies won’t have the expertise or incentives to monitor how their AI systems make decisions.
A growing number of groups are pushing back against the idea that super-intelligent AI should be built in secret by a few large corporations. Instead, they want it to be created out in the open. In August 2021, a community-driven group called EleutherAI released an open source version of GPT-Neo, a machine learning model that can write realistic comments and essays on nearly any topic—and pass the notoriously difficult Turing test.
In addition to being a valuable tool for developers, the open-source software offers an opportunity for users to examine and compare different models’ explanations of their decisions. This transparency can help to increase trust in AI systems and facilitate greater adoption. But while a growing number of companies are starting to build explainable AI (XAI) into their products, the majority still haven’t done so.
To address this gap, we developed a Python library called OmniXAI that can be used to build XAI for existing AI systems. The library offers a range of tools to help developers create and visualize explanations of decisions made by their AI systems, including a framework for automatically translating explanations into human-readable text. This enables users to gain more insight into how their AI system makes decisions and provides them with a better understanding of why it chose a particular action over another.
AI is transforming our lives in numerous ways. In healthcare, it’s helping to diagnose diseases and improve treatment. In autonomous vehicles, it’s making them safer and more reliable. And in assistive technology, it’s enabling devices to better understand user input and enhance connectivity, improving accessibility. But we need to ensure that the AI systems we create are in our best interest. Otherwise, they may end up creating their own worlds—not to mention pheromone-generating aliens and those dream-manipulating monsters we saw in that sci-fi film. Fortunately, we can take steps to prevent that from happening. By fostering a culture of transparency and accountability, we can ensure that the future of AI is truly beneficial for all. omnivoid ai