Can AI safely recreate interactions with your deceased loved ones?
Recently, Joshua Barbeau, a resident of Bradford, Canada, successfully used a website called Project December to create a hyper-realistic AI-enabled (Artificial Intelligence) chatbot that recreated the experience of communicating with his deceased fiancée Jessica Pereira. This might sound like a blessing from a science-fiction film but the AI's creators warn this technology could fuel large-scale misinformation campaigns like deepfake videos. Here are more details.
- Old messages, background information helped replicate interaction with late fiancée
- Microsoft now has exclusive licensing rights to GPT-3
- OpenAI says GPT-3 could start a nearly undetectable misinformation spree
- GPT-3 usage is currently limited by credit-based system
- Barbeau's story triggered immense interest in Project December
- With AI like GPT-3 becoming affordable, associated dangers loom nearer
Old messages, background information helped replicate interaction with late fiancée
Talking to San Francisco Chronicle, Barbeau explained that he modeled a customized matrix on the Project December interface using old messages and some background information of his fiancée who lost the battle to a rare liver condition eight years ago at age 23. Under the hood, Project December uses the GPT-3 language model that can generate pages of text with simple user prompts.
Microsoft now has exclusive licensing rights to GPT-3
Recently, Microsoft gained exclusive licensing rights to GPT-3 after it invested $1 billion into a group called OpenAI that developed the technology. Microsoft explained that it has plans to use the technology to generate computer programs using requirements described in natural language.
OpenAI says GPT-3 could start a nearly undetectable misinformation spree
In a blog post following the release of GPT-2 (GPT-3's predecessor), OpenAI explained that the technology could be used in "malicious ways" to automate the generation of "abusive or faked content on social media," "misleading news articles," and content that "impersonates others online." The concerns are real because GPT-2's generated text is almost impossible to discern from a similar human-written piece.
GPT-3 usage is currently limited by credit-based system
OpenAI released GPT-2 in a phased manner and continues to restrict access to the more advanced GPT-3 technology until people learn the "societal implications" of such technology. Even Project December's creator, a 42-year-old San Francisco Bay Area programmer named Jason Rohrer, said that the website is fueled by $1,800-worth of GPT-3 credits each month. The platform relies on GPT-2 as a backup.
Barbeau's story triggered immense interest in Project December
Following the SF Chronicle article, there was immense interest in Project December since many people wanted to try out first-hand interactions with the advanced AI. On Reddit, Rohrer explained that this drove the GPT-3 credits usage dangerously close to the monthly limit, so customers like Barbeau who created their own matrix would be restricted to GPT-2 temporarily.
With AI like GPT-3 becoming affordable, associated dangers loom nearer
Barbeau's use case certainly rekindled fond memories but it also highlighted a new danger we face. We are well past the point where we could tell computer-generated content from human-generated variations. We believe that with memberships for sites like Project December costing just $5 per head, AI is fast becoming affordable and society has the collective responsibility of encouraging transparency and curbing technology's misuse.
