We asked experts: What are you tracking in terms of image veracity? What are your views on image veracity and Gen AI and how it impacts business and society today?
"If 'seeing is believing,' then business and society are in for a lot of trouble because generative AI is capable of producing highly realistic images and video of things that never happened. However, there's a second half to that old adage: 'Seeing is believing but experience is the truth.' People have long known that photographs can be taken out of context and manipulated, and what people need to be aware of today is that the technology to manipulate visual media and take it out of context is readily available. As people consume media, especially visual media, people should pay attention to their experience and consider the media's source and context."
"Globally, there is a great deal of worry about the fate of Democracy and the effect that false information & manipulated content online is having. Countries like Brazil that have experienced their own unrest (the Jan 8, 2023 attack on Brazil's capital) are determined to update their regulations and hold tech companies accountable. Recent discussions, like the UNESCO Internet for Trust meeting, showed growing differences between free speech advocates and those who've been personally affected by online violence and hate speech. These differences may mean that a global agreement will be hard but European governments have moved ahead with the Digital Services Act and that will have an effect on the rest of the world. And just when I was getting used to the DSA now we have LLM and Generative AI to think about!"
– Anya Schiffrin, Director of Technology, Media, and Communications Specialization, School of International and Public Affairs at Columbia University
Featured Content of the Month: Artist Refik Anadol created a new immersive installation Glacier Dreams in the Julius Bär Lounge at Art Dubai 2023.
Headlines to Note
Weaponization of deepfakes on a Local Level?:Three high school students in the New York City area allegedly posted deepfake videos on TikTok showing a local middle school principal and member of law enforcement making racist comments. If truly a synthetic video, this marks one of the first serious examples of the weaponization of deepfake technology at a local level. Reports suggest that the video was part of a potentially concerning trend, which will be challenging for authorities to debunk and address in real-time and scale.
AI Voice Scams: Scammers are using AI to sound like family members in distress, swindling people out of thousands of dollars. The increasing accessibility of AI makes it easier for scammers to defraud victims. According to the Washington Post, a Canadian couple lost $21,000 from an AI-generated voice claiming to be their son's lawyer.
Chat GPT-4:OpenAI has unveiled their new GPT-4 language model that allows ChatGPT to process and recognize images as well as text inputs. Thisnew AI model behind ChatGPT scored in the 90th percentile on the bar exam. GPT-4 is OpenAI's latest large language model, expanding on the capabilities of GPT-3.5, which powers the ChatGPT chatbot. Most notably, GPT-4 is multimodal, meaning that it can accept both images and text as inputs for the first time.
AI x Government:Romania's Prime Minister Nicolae Ciuca has presented a new AI honorary adviser named "Ion." Ion will scan social media to inform the government of Romanians' proposals and wishes in real time. Ciuca described Ion as the first government adviser to use artificial intelligence and an international first.
Responsible Practices:The Partnership on AI is inviting companies and AI experts to back its new ethics guidelines for synthetic media. The Framework for Collective Action provides recommendations for safely creating, sharing, and distributing synthetic media and is aimed at preventing misinformation and manipulation.
AI Accessibility Continues to Increase:Hugging Face and Amazon Web Services have partnered to make AI more open and accessible, with the goal of accelerating next-generation machine learning models. Another example of an effort to increase accessibility is the EleutherAI Institute. EleutherAI is a nonprofit AI research lab dedicated to making advancements in large language models more accessible. Donations and grants from backers and AI startups will be used to fund the institute.
Non-consensual Sexual Deepfakes:Twitch stated that deepfake porn is now grounds for being banned. This was made clear after a January incident in which deepfakes of female streamers were shared by a fellow streamer. Non-consensual celebrity deepfakes have also increased on social media as hundreds of deepfake ads featuring Emma Watson and Scarlett Johansson have been seen on Facebook and Instagram. An app for creating deepfake face-swap videos ran over 230 ads on Meta platforms that showed what appeared to be pornographic videos, starting with a well-known Pornhub intro track playing with the faces of famous female actors.