Generative AI tools like ChatGPT can now create in addition to text, voice, art, and video. people can. I need to know who made what.
(Illustration: Pete Ryan)
Valerie Pisano He is President and CEO of Mila, a non-profit artificial intelligence research institute based in Montreal.
It used to be very easy to know that machines were involved in creating something. The image borders were clearly pixelated, the audio was slightly choppy, and the whole thing looked robotic. With OpenAI’s deployment of his ChatGPT last fall, we’ve reached the point of no return. In other words, artificial intelligence tools have mastered human language. Within weeks, the chatbot amassed his 100 million users and spawned competitors like his Bard at Google. Suddenly, these applications are co-composing our emails, mimicking our speeches, and helping users create fake (yet funny) photos. Soon we will be helping Canadian workers in almost every sector to summarize, organize and brainstorm. This technology not only allows people to communicate with each other. It communicates with us and sometimes better than us. Generative AI tools can now fake people, just as criminals can fake money.
Myra, where I work, is a research institute that regularly convenes AI experts and experts from various fields, especially on the topic of governance. Even we didn’t expect this innovation to permeate our daily lives so quickly. Currently, most countries do not have AI-focused regulations in place, with best practices around usage and clear penalties to prevent bad actors from using these tools to do harm. neither. Lawmakers around the world are panicking. Italy temporarily banned ChatGPT earlier this year due to privacy concerns. And China recently drafted a regulation that would require a security assessment of any AI tool that generates text, images or code.
Here at home, scientists and businesspeople are urging the federal government to expedite enactment of the Artificial Intelligence and Data Act (AIDA), tabled by the Liberal Party in June 2022. The act is part of Bill C-27. Consumer Privacy and Data Protection Laws – Includes guidelines for deploying AI tools and fines for abuse. There is only one problem. AIDA may not enter into force until 2025. Laws usually don’t move as fast as innovation. In this case, you will have to catch up quickly.
Read more: Ivan Zhang, Aidan Gomez, and Nick Frosst build smarter, friendlier chatbots
The European Union has taken the lead in the AI law, the first AI-specific rule in the West, which began drafting two years ago. Canada should consider implementing one of the EU’s key measures as soon as possible, requiring disclosures when developers and companies use or promote AI-generated content. Photos created using text-to-image generator DALL-E 2 may contain watermarks, while audio files may contain disclaimers from chatbots. It can be anything that is immediately obvious to anyone who sees, hears, or otherwise engages with the content. I made it with the help of machines. As an example, a professor we work with at Mila has students use her ChatGPT to summarize a literature review at the beginning of a paper (but noting which parts are bot-generated). provided that They’re also responsible for fact-checking whether AI isn’t citing sources that don’t exist (or are completely bonkers).
The EU AI law also contains a similar provision. Any company that implements generative AI tools like ChatGPT should make publicly available a summary of the copyrighted data used for training in any capacity. Suppose you are using a bank’s financial planning services. In a properly labeled world, that bot would say: Based on that information, my program suggests his three courses of action…” In the creative field, artists have already filed copyright lawsuits for their images being lifted by bots. Mandatory labeling would make it easier to check what those creations were “inspired by.”
One of the main dangers of ChatGPT, specifically, is that ChatGPT says the wrong things in such an authoritative way that it tricks us into thinking we are smarter than we really are. (OpenAI CEO Sam Altman tweeted: “Fun creative inspiration. Great! Reliance on factual queries. Not a very good idea.”) It was trained on a huge amount of information. The upgrade gave me internet access. It gives the impression of having a kind of superintelligence. And although the program generates responses almost instantly, he speaks one sentence at a time in a human-like rhythm. Even people with highly developed intuition can be deceived. ChatGPT is designed to put your trust in us.
It’s not designed to find the correct answer. ChatGPT is not a search engine whose algorithm favors more trustworthy websites. It’s common to ask generative AI questions and spew out errors or “hallucinations” (a technical term for mistakes that the AI confidently returns).in recent times 60 minutes In this episode, Google senior executive James Manika asked Byrd to recommend a book on inflation. None of those proposals exist. Type in “Valérie Pisano, AI, Montreal” and ChatGPT will give you a made-up bio instead of providing a summary of my real bio. Creating fake news is already very easy. Generative AI tools will be able to provide an infinite amount of disinformation.
Related: My students use ChatGPT to write papers and answer exam questions.i support it
In the absence of meaningful guardrails, we must rely on the judgment and good faith of ordinary Internet users and businesses. This is not enough. Canada cannot leave the oversight of this technology solely to the companies that build it. This is essentially what happened with social media platforms like Facebook. (I’m no historian, but I remember it having a negative impact on fair elections.) At some point, the government decided to disguise his AI-generated content as being human-made. It would have to be legal or illegal. and international level.
Penalties also need to be agreed upon. Not all misuses of generative AI carry the same level of risk. Using art generator Midjourney to create a fake photo of Pope Francis in a fluffy winter coat isn’t really a threat to anyone. This is easily managed with a simple “report” button within Instagram’s platform. But in fields such as journalism and politics, using AI to mislead can have disastrous consequences.
Labels also enforce some level of AI literacy in the average person. It’s past the point where you can say, “But I’m not a tech person!” In the future, all Internet users will encounter AI on a daily basis, not just reading articles about it. It will inevitably change the way everyone creates, competes, works, learns, governs, cheats and chats. Watching (or hearing) a “machine-made” disclaimer gives us the opportunity to choose how its output permeates our private lives.
Of all the new tools, chatbots seem to have impressed the scientific community the most. Especially because chatbots feel human. (In fact, I find it hard not to say “please” and “thank you” when interacting with them, even though I know the bots won’t judge my manners.) It is easy to imagine using generative AI for any task. Be more emotional. But while you might ask the new Compose AI extension for Google Chrome to “write an email requesting a refund” to an airline, you probably wouldn’t use it to write a note to a close friend. You can also see the benefits of Snapchat’s new My AI bot. It now greets millions of teens with a friendly “Hi, what’s up?” At the same time, we understand that machines can never replace the deeper kind of support we need to grieve a painful loss. It may be better to leave it to humans. I think you get it.
