Collaborative Intelligence Week in Review - 17Nov2023

Best business article(s) I read this week…

• Implementing and scaling artificial intelligence

  This is a long academic article, so it may not be for everyone. The authors view the challenge of implementing and scaling AI in organizations through the lens of socio-technical systems theory, an approach I use myself.

Best technical article(s) I read this week…

• Conversational AI Threads for Visualizing Multidimensional Datasets

  Those interested in or attempting to use LLMs for data analysis and visualization may be interested in this article. The authors present an approach for developing prompting threads to improve outputs and provide their assessment of the strengths and weaknesses of LLMs for this application.

• On the Exploitability of Reinforcement Learning with Human Feedback for LLMs

  The authors describe how they used poisoning attacks on LLMs to highlight critical security challenges for the Human-in-the-Loop reinforcement learning algorithms used to train LLMs.

Other item(s) of note…

• Morgan Stanley Research Note on Generative AI

  Morgan Stanley released a research note on the opportunity for generative AI in specific industries and at a macroeconomic level. They expect “more than 40% of occupations, with associated labor costs of $4.1 trillion in the U.S. alone, will be affected by generative AI in the next three years.” I think this is too big, too soon, but you never know.

• New data poisoning tool lets artists fight back against generative AI

  The path to broad usage of generative AI and other CI technologies will not be smooth. This article describes tools artists can use to break the training algorithm of generative AI models that scrape their images from the web. I expect there will be many efforts by humans in general and creators specifically to “fight back” against generative AI.

• Enhancing human-AI collaboration: The case of colonoscopy

  Description and findings of a CI use case in medicine. As the authors point out, many of the principles applied are relevant to many areas of care.

Coolest thing I saw…

• DIRFA algorithm for creating videos from a photo and an audio track

  This advancement may be problematic because it makes creating deep fakes easier. But from a purely technological viewpoint, it’s pretty cool.