This week in AI and Machine Learning: A controversial new chatbot, a crucial legal decision, and more.
A Note from the Author
Manufacturers of all types can see transformative changes when they introduce computer vision to their production processes. From pharmaceutical manufacturers cranking out thousands and thousands of pills to luxury goods manufacturers sweating over specialty items, all types of enterprises benefit from seeing more.
Monday, August 15th is National Leathercraft Day. Celebrate the occasion with a look at some of the ways Plainsight’s custom vision AI models and professional services are helping luxury goods manufacturers maintain quality standards for items like handbags.
Artificial Intelligence News
Meta’s Problematic New Chatbot
Chatbots learn to effectively interact with people, offer guidance, and answer questions by engaging in regular conversations with users. Over time, chatbots apply the lessons they’ve learned to improve the quality and accuracy of their responses. Unfortunately, AI can often learn the wrong lessons from its interactions with users. Meta has run into these issues with its latest chatbot: BlenderBot 3. Unveiled last week, the bot has already made a host of headlines for its insulting remarks and controversial claims.
Some of what the chatbot has said is strange but benign. Speaking with CNN Business, the bot reported watching anime and having a wife. BlenderBot 3 also described itself as sentient before calling into question the results of the 2020 election. Jeff Horwitz, a journalist for The Wall Street Journal tweeted out an exchange where the chatbot appeared to deal in anti-Semetic tropes. The chatbot has also repeatedly maligned Meta CEO, Mark Zuckerberg, calling him “creepy” and “manipulative.”
When Meta released BlenderBot 3 last week, the tech giant acknowledged the possibility of issues like these. “Since all conversational AI chatbots are known to sometimes mimic and generate unsafe, biased, or offensive remarks,” the statement reads, “we’ve conducted large scale studies, co-organized workshops and developed new techniques to create safeguards.” The bot’s conversations over the last week suggest these safeguards may not be as effective as Meta anticipated.
It’s Official, AI Cannot Secure a Patent in the U.S.
Artificial intelligence can do quite a lot, but it can’t be listed as an inventor on a U.S. patent. This is according to the Court of Appeal for the Federal Circuit’s recent ruling in the case of Thaler v. Vidal. The Court cited the text of the Patent Act which explicitly describes inventors as human beings. This ruling is yet another setback for Stephen Thaler and the Artificial Intelligence Project who’ve spent years submitting patent applications on behalf of the AI known as DABUS.
Thaler has also unsuccessfully submitted patents in jurisdictions including Australia, Germany, and the United Kingdom. Only in South Africa has DABUS been officially listed as an inventor. The National Law Review notes, however, that South Africa imposes a less lengthy review process. Thaler’s multi-pronged argument included pointing to the Constitution’s Copyright and Patent Clause, suggesting that broadening the definition of inventor would “promote the Progress of Science and useful Arts.” Check out a detailed summary of both Thaler and the Court’s positions.
Google Proposes a New Approach to Video-Language Learning
Video data is everywhere and data scientists are leveraging it for an increasing number of applications that rely on models capable of connecting videos and natural language. These applications – which include video captioning and video question-answering (VideoQA) – continue to present a number of challenges. VideoQA remains particularly challenging because models must understand semantic, temporal, and language information all at once.
To address challenges related to VideoQA, a team of Google researchers proposed iterative co-tokenization, a new approach to video-language learning. Iterative co-tokenization was found to outperform other preferred approaches in terms of accuracy and computational efficiency. Efficiency was considered especially important because even short videos can contain dozens or hundreds of individual frames. Check out a summary of the research to learn more about how Google’s new approach differs from existing methods.
Read Plainsight’s recent press release to learn more about our partnership with Google Cloud.
Join our Community
See you next week! Until then, keep the conversation going on Plainsight’s AI Slack Channel.
About the Author & Plainsight
Bennett Glace is a B2B technology content writer and cinephile from Philadelphia. He helps Plainsight in its mission to make vision AI accessible to entire enterprise teams.
Plainsight’s vision AI platform streamlines and optimizes the full computer vision lifecycle. From project strategy, through model deployment, and ongoing monitoring, Plainsight helps customers successfully create and operationalize vision AI applications to solve highly diverse business challenges.