ChatGPT and generative artificial intelligence (AI) dominate headlines and conversations. We’ve seen it when posting weird and intriguing screenshots and images of chatbot conversations on social media, allowing search platforms to “talk” to chatbots. But what is behind this technology? Who feeds the data and decides where it comes from? What does this have to do with human rights? Senior Web Producer Paul Aufiero sits down with Anna Bacciarelli, her Program Manager in the Technology and Human Rights Division at Human Rights Watch, to discuss the issues at the center of this new debate as companies race to develop and implement generative AI. This section explains.
What is Generative AI and ChatGPT?
ChatGPT is a type of technology called “Generative AI”. Generative AI is a fairly state-of-the-art technology that allows users to generate new content (words, images, and even videos) by entering prompts into a system that guides them to create specific outputs. will do so.
ChatGPT, perhaps the most famous generative AI product, was developed by Californian tech company OpenAI. The generative AI race has been going on since OpenAI released his ChatGPT in November 2022. Since then, multiple technology companies such as Google, Amazon, and Baidu have released their own generative AI products. Microsoft has invested billions in OpenAI and uses his ChatGPT in the generated version of search engine Bing. Elon Musk is also reportedly gearing up to launch a new company focused on generative AI.
Because of the significant financial and technical resources required to create and run these products, including access to vast amounts of data and computing power, they are generally large enterprises or very well-financed. It’s important to note that only companies can build these products. If generative AI is the future, it is a vision put forward and realized by a handful of powerful technology companies and individuals, each with their own commercial interests at stake. This raises important questions about corporate power and accountability.
What is a generative AI feed?
Generative AI models are trained on massive amounts of data. It is problematic when the training data and training process are not publicly available, as is the case with GPT4, OpenAI’s newest image and text generation AI product. We know that OpenAI’s previous model, GPT3, was trained on text scraped from numerous internet sources, including Reddit boards.
Using billions of images and text from across the internet as training data, without careful filtering and moderation, risks perpetuating the worst content-related problems we’ve already seen online. This means presenting opinions as facts, creating unbelievable and deceptive images and videos, reinforcing structural biases, and generating harmful and discriminatory content, just to name a few.
Research shows that content on the web simply doesn’t represent the reality most people live in. Wikipedia, for example, still has a big bias in showing information about men and the so-called Global North, despite great efforts to address content inequities. A generative AI system trained on non-representative data simply reproduces unfairness from existing internet content.
There are also obvious privacy implications for companies scraping people’s content without their knowledge or consent about their training data. How do I know that an image or text has been used to train a generative AI system? How do I request its removal from the system?
How is this technology different from what is already available online?
This is the first time that advanced and creative AI applications are accessible to anyone using a computer or smartphone. However, both text and image generation AI applications are still dominated by English, so accessibility in that sense is very limited.
For the past five years, Human Rights Watch has been working on the impact of AI issues on human rights. Generative AI is, in many ways, an extension of well-known concerns about AI and machine learning techniques. Discrimination, and lack of accountability when things go wrong.
What are your privacy and data security concerns?
Be careful what you enter into the generative AI tool. We have to assume that everything we put into a generative AI product is used to some extent to train and “improve” the model. It may also appear in technology companies that own or use the system. Companies across multiple sectors are now asking staff to refrain from entering confidential or personal information into generative AI systems.
Even if you enter seemingly mundane information into a generative AI search or chatbot, it can be used to figure out who you are. AI is particularly good at noticing patterns, and searching and typing in chatbots can over time reveal more and more information about our identities and behavior, from content that we don’t particularly think reveals information. A high degree of insight can be revealed.
At this time, we do not have enough information to know to what extent our information is being used and can be associated with an individual’s identity. I need an answer from It’s worth noting that OpenAI tightened his ChatGPT privacy controls last week. This follows the product being “banned” in Italy due to data protection concerns. This is a very underdeveloped landscape.
Are these systems trustworthy?
Systems known to contain falsehoods and inaccuracies cannot be given too much weight, especially if their design is opaque. For example, Google’s chatbot Bard returned answers containing factual errors when launched. You can use Google search to find the correct answer, but you may have to sift through the wrong answers before finding the correct answer. However, some generative AI systems cannot easily see the source of the information. This helps critically assess whether people trust the output.
It is worth noting that there are additional safety concerns regarding generative AI chatbots. design and conversational tone. This can lead users to over-trust them.
The consequences of this are serious. The Colombian judge said he consulted ChatGPT while preparing his sentencing. To what extent could this technology affect court decisions?In Belgium, a woman says she committed suicide after her husband interacted with a generative AI her chatbot.
Technology makers and regulators should stop and consider some of the big issues. How could this be abused? Even with the best of intentions, what could go wrong? What can you do about it?
Companies are rushing out products that are unsafe for general use. Previous versions of generative AI chatbots produced problematic and biased output. We know that the current industry competition is not based on human rights policies and practices, but rather a race to the bottom.
Technology companies have a particularly important human rights responsibility when creating powerful new exploratory technologies. Prior to product release, it must be clearly demonstrated that human rights risks are identified and mitigated. You should also be held responsible for any damage caused by the product. This requires opening training data, design values, and content moderation processes to independent scrutiny.
It seems that this technology can lead to rampant misinformation. are you worried about that?
Yes it will probably be a big problem. Generative AI has the potential to significantly change how we use the internet by changing the way we find and trust information online. How do you trust what you see? With developments such as the text-to-image model and text-to-video conversion, we are opening Pandora’s box for Internet content authenticity. We already have a problem of knowing what is real, and it is about to get bigger.This could be for example political events such as elections or confrontation and critical situations.
There are some generative AI chatbots impersonating historical figures and government officials. People may publish texts that can be used to cover up serious crimes and human rights violations.
What about the people sifting through the data to feed these chatbots?
Building AI involves a lot of invisible human effort. Someone has to label the training data, and someone has to decide if the machine is right or wrong. Because this process relies on humans inputting information, technology inevitably includes human bias.
Exploitation of workers in the creation of AI is another important human rights issue. A Time magazine study showed how OpenAI outsourced data labeling for ChatGPT to Sama.
Tech companies outsource this workforce, primarily to the Global South workforce. There is a huge gap in working conditions between the headquarters of US technology companies and the places that drive these technologies. Workers building AI often see the worst of the internet. All businesses have a responsibility to provide a living wage and dignity at work.
What types of regulations exist to govern this technology?
Technology companies have tried to stay ahead of regulation by attempting to self-regulate. For example, creating and adopting principles and guidelines to be nominally followed. For example, Google states under its AI Principles that it will not release any of his AI products whose purpose is against human rights. But expecting technology companies to follow their own principles puts too much reliance on self-governance. AI is too powerful and its rights implications too severe for companies to regulate themselves.
All companies have a responsibility to respect human rights standards and to take steps to identify, prevent and mitigate the human rights risks they cause, contribute to or are associated with through their own operations or operations in their value chains. I have. These responsibilities are set out in the United Nations Guiding Principles on Business and Human Rights and the guidelines of the Organization for Economic Co-operation and Development. We need to translate the UN Guiding Principles into binding law for all technologies, not just AI.
Legislators have struggled to keep pace with new tech industry developments over the past few years, but we are now seeing the emergence of AI regulations at the national and regional levels. The big one is the European Union’s AI law, which passed its first ballot last week with a proposal to address copyright issues in generative AI. However, let’s see if it’s good enough to address the human rights issues this technology raises.
How will this technology affect human rights investigations?
Human rights groups have a key role to play in revealing and challenging how new technologies are being developed, including products that use generative AI.
But there are also real threats to human rights investigations using generative AI. Advances in this technology have made it faster and easier to create authoritative content. This undoubtedly poses new challenges for human rights groups documenting abuses and gathering evidence to explain injustice. Their ability to do so will become increasingly important in the future of generative AI, as false or misleading information circulates online, including highly credible photos and videos generated by AI. .
