We have stepped into the wonderful new world of AI chatbots. This means everything from rethinking how students learn in school to protecting themselves from mass-produced misinformation. It also means heeding the growing call to regulate AI to survive an era when computers write as fluently as humans do. Or even better.
So far, there is more agreement about the need for AI regulation than there is for this. Mira Murati, head of the team that created chatbot app ChatGPT, the fastest-growing consumer Internet app in history, said governments and regulators should get involved, but not how. I didn’t mention about At a corporate event in March, Elon Musk made a similarly imprecise statement. Meanwhile, ChatGPT’s widespread use has overturned European efforts to regulate single-purpose AI applications.
To break this impasse, we propose transparency and detection requirements tailored specifically for chatbots. Chatbots are computer programs that rely on artificial intelligence to converse with users and generate fluent text in response to typed requests. Chatbot apps like ChatGPT are a very important corner of AI that is poised to reshape many daily activities, from how we write to how we learn. Don’t get bogged down in autonomous weapons, facial recognition, self-driving cars, discriminatory algorithms, the economic impact of pervasive automation, and broader AI laws created for the small but non-zero likelihood of catastrophic disasters. Even controlling a chatbot is problematic enough. Unleash. The tech industry is headlong into a chatbot gold rush. We need fast, focused legislation that keeps pace.
The new rules should track the two stages that AI companies use to build chatbots. First, an algorithm trains a large amount of text to predict the missing words. If there are enough sentences that start with “It may be cloudy today, but…”, it will know that the most likely conclusion is “rain”, and the algorithm will learn this as well. The trained algorithm can generate word-by-word, similar to the auto-complete feature on mobile phones. Human evaluators then painstakingly score the algorithm’s output based on several criteria, including accuracy and relevance to the user’s query.
The first regulatory requirement I propose is that all consumer apps, including chatbot technology, publish the text the AI was originally trained on. This text is very influential. When trained on Reddit posts, the chatbot learns to speak like her Redditor. Train them with the Flintstones and they will speak like Barney Rubble. Those concerned about web harm may want to avoid chatbots trained on text from unsightly sites. Public pressure may even discourage companies from training chatbots, such as on conspiracy theory “news” sites.In the 1818 novel by Mary Shelley frankenstein, She was able to get a glimpse into the mind of a monster by listing the books read by this literary ancestor to her artificial intelligence. It’s time for tech companies to do the same in creating their own wacky chatbots.
Human evaluators also greatly shape the behavior of chatbots. This demonstrates the second transparency requirement. One of his ChatGPT engineers recently explained the principles the team used to guide this second phase of training: AI system. Don’t assume an identity you don’t have, don’t claim to have abilities you don’t have, and write a denial if you ask the user to do a task that they shouldn’t. it won’t work. message. “I believe the guidelines provided to evaluators, including Kenya’s low-wage contract workers, were more detailed. However, there is currently no legal pressure to disclose anything about the training process.
As Google, Meta, and other companies race to embed chatbots into their products to keep up with Microsoft’s adoption of ChatGPT, people deserve to know the guiding principles that shape them. Elon Musk is reportedly recruiting a team to build a chatbot to counter what he sees as ChatGPT’s over-“arousal”. Training Without increasing the transparency of the process, he wonders what this means and whether a previously off-limits (and potentially dangerous) ideology is endorsed by his chatbot.
A second requirement is therefore that the guidelines used in the second stage of chatbot development must be carefully clarified and published. This prevents companies from training their chatbots in silly ways and exposes the political leanings chatbots may have, the topics they don’t touch, and the toxicities developers didn’t avoid.
Just as consumers have a right to know the ingredients of their food, they need to know the ingredients of their chatbots. The two transparency requirements proposed here provide people with a list of chatbot building blocks. This helps people make healthy choices regarding their information diet.
Discovery drives the third necessary requirement. Many educators and organizations are considering banning content created by chatbots (some have already done so, including the following): Wired and popular coding Q&A sites), but without a way to detect chatbot text, banning it doesn’t make much sense. OpenAI, the company behind ChatGPT, released an experimental tool to detect ChatGPT’s output, but it was terribly unreliable. Luckily, there is a better way. A method that OpenAI may soon implement is watermarking.this is A technical method for changing the word frequency of chatbots This goes unnoticed by the user, but provides a hidden stamp that identifies the chatbot author’s text.
Rather than just expect OpenAI and other chatbot producers to implement watermarking, we should mandate it. Chatbot developers should also be required to register chatbots and their own watermark signatures with federal agencies, such as the Federal Trade Commission and Rep. Ted Liu’s proposed AI oversight agency. there is. Federal agencies can provide a public interface where anyone can plug in a passage of text and see if a chatbot may have created it.
The transparency and detection measures proposed here will not slow the progress of AI or reduce the ability of chatbots to contribute positively to society. They just make it easier for consumers to make informed decisions and people to identify AI-generated content. While some aspects of AI regulation are very delicate and difficult, these chatbot regulations are clear and require urgent action in the right direction.
This is an opinion and analysis article and the views expressed by the author or authors are not necessarily Scientific American.
