The U.S. Patent and Trademark Office just issued its decision that only “natural persons" - not AIs - can be named as inventors, refusing two patents for an AI that created an emergency flashing light and shape-shifting food container. The inventions come from physicist and AI researcher Stephen Thaler's "DABUS," which won't get its name on the patents. The office argues that patent law refers to inventors in humanlike ways, which can't apply to machines.

• The European Patent Office (EPO) and UK’s Intellectual Property Office (IPO) have issued similar rulings turning down Thaler's request and patents.
• The Artificial Inventor Project, which supports Thaler, argues that it could become impossible to patent new innovations if no humans were involved in the work closely enough to take credit.
• The issue of non-human ownership over properties was brought up when PETA argued that a macaque monkey could own the rights to selfies it took; the case was ultimately settled.

The technique - still in the early stages - taps AI agents to simulate how actual people would react to different tax scenarios. The agents collect and trade resources, earning money, while the AI economist optimizes taxes and subsidies to gain the most output and achieve the best income equality. The results could help governments, economists, and others design tax policies that benefit everyone through things like improved productivity and social equality.

• The two-level DRL framework encourages the software agents through a reward system, according to the research paper.
• A YouTube video shows the agents in action.
• The AIs work because in the real world, "you cannot change your policies a million times," Salesforce chief scientist Richard Socher said.
• Scientists have employed similar reinforcement learning techniques to beat players in games like Go and Starcraft.

The rapper's company Roc Nation sent notices to YouTube to remove two AI-created deepfakes that impersonate Jay-Z's voice. The fake audios of Jay-Z rapping Billy Joel’s "We Didn’t Start the Fire" and Hamlet’s “To be or not to be” soliloquy comes from the YouTube deepfake channel Vocal Synthesis, which is run by an anonymous creator. As of yesterday, the copyright strikes were removed because the takedown requests were incomplete, YouTube said, so the videos are back up (though they could be taken down again later).

• The issue raises questions about how much legal sway artists have when trying to remove their likeness from AI-edited videos and songs.
• To generate the deepfakes, Vocal's creator uses Google's Tacotron 2 text-to-speech model trained on Jay-Z's speech patterns.
• Vocal Synthesis has a subreddit where people are talking about the issue.
• Earlier today, Vocal published a new deepfake of Jay-Z rapping the Darth Plagueis Copypasta.

Our daily AI Masterclass feature showcases well-known thought leaders or experts in AI, including their visions and advice for the future.

Microsoft's Ram Shankar Siva Kumar and Harvard Business School's Frank Nagle argue in favor of Al/ML (machine learning) insurance to protect companies from failed systems, whether they're done intentionally or unintentionally. In a column for Harvard Business Review, the two note that most major companies have had their ML systems misled in one way or another, whether it's through attacks by adversaries to steal algorithms, or faulty assumptions made by developers. Here, they offer an action plan to start insuring and protecting ML models...

The state took out a $20.7 million contract with the surveillance company to apply Banjo's AI to live traffic feeds, dispatch logs, and social media, which can detect "anomalies" that could help law enforcement. Now, a OneZero report has exposed Banjo CEO Damien Patton's past as a member of the Ku Klux Klan's Dixie Knights.

The Hong Kong-based AI startup said that the agreement with the People's Bank of China's research unit, which is overseeing the development of the nation's state-backed digital yuan, is aimed at boosting innovation and the introduction of AI in China's technology sector...

To fill the void in interactions between colleagues right now, NLU researcher Josh Eisenberg will host the workshop on July 24 in a virtual space on an Animal Crossing island. The conference is accepting abstract submissions in any AI domain through June 12, though fields like automatic speech recognition, image generation, NLP, conversational AI and others are highly encouraged. 

More:

• Participants will give 15-minute presentations through their Animal Crossing characters, while the virtual conference, audio, and slideshows will be live-streamed over Zoom.
• @Mappletons called it a "fabulously odd and perfectly sensible digital adaptation."
• A similar DevOps conference in Animal Crossing takes place tomorrow, April 30.
• Some less positive feedback from a Reddit user: "Next time you might want to consider choosing a less escapist game as the fulcrum for the workshop."

In this special feature, we summarize 10 of the top daily headlines in AI/ML:

• Facebook has open sourced its Blender chatbot, which it claims is more human-like and possesses empathy. You can find the full model, code, and evaluation setup here.
• A new AI model suggests that wearing face masks has a “significant impact” on the spread of coronavirus if at least 80% of people wear one.
• The FTC released tips for using AI and algorithms to avoid misleading consumers...