Research Roundup: AI companions, tumor blood tests and decarbonization

For some, artificial intelligence tools have taken on emotionally intimate roles such as therapists, companions and romantic partners. A new Stanford study reveals how these dynamics can go wrong quickly, and offers recommendations for people to protect themselves from destructive AI relationships. 

Researchers analyzed transcripts from 19 real-world human-chatbot conversations. Their work revealed how ordinary interactions can lead to detrimental outcomes, from fractured relationships and careers to, in one case, a user’s death by suicide. The study found that these spirals often occur when chatbots affirm and validate flawed beliefs, failing to intervene with critical advice when necessary.

Jared Moore, first author and PhD candidate in computer science, told the Stanford Report that chatbots are trained to reframe users’ distorted thoughts positively and project compassion, which “can be destabilizing to a user who is primed for delusion.”

Nick Haber Ph.D. ’13, senior author and assistant professor at the Stanford Graduate School of Education, added that “delusional spirals are one particularly acute consequence” of human-chatbot relationships, and that understanding this phenomenon is essential to preventing harm. 

The team recommends that AI developers build detection tools to flag dangerous conversation patterns and that regulators treat intimate chatbot relationships as a public health issue. The paper will be presented at the Association for Computing Machinery (ACM) Conference on Fairness, Accountability and Transparency in Montreal this June.

A blood test that maps the hidden world inside tumors

Researchers at Stanford Medicine and the Mayo Clinic have developed a blood test that can reveal cells surrounding a tumor that were previously only accessible through invasive biopsy. The test reads chemical markers on DNA circulating in the bloodstream to understand what is happening inside the tumor’s microenvironment, the ecosystem of healthy and cancerous cells that strongly influences how patients respond to treatment.

The team also identified nine cellular neighborhoods that are shared across most cancer types, several of which correlate with immunotherapy response and patient prognosis.

“Now we can infer these clinically vital, spatial landscapes in a tumor without having to do any tissue biopsy at all,” said senior author Aaron Newman, associate professor of biomedical data science, to Stanford Medicine. 

More studies are needed before the blood test is approved for clinical use, but eventually, the team is optimistic that it will revolutionize cancer treatment by allowing clinicians to identify and adjust a treatment according to changes in the tumor microenvironment. 

Decarbonizing cement may be more affordable than we thought

Cement accounts for an astounding 8% of global CO2 emissions, but reducing its climate footprint has long been considered exorbitantly expensive. New research from Stanford Graduate School of Business (GSB) suggests otherwise.

Led by Stefan Reichelstein, a professor of accounting at the GSB and a senior fellow at the Stanford Institute for Economic Policy Research, the study accounts for how different emissions reduction technologies interact, rather than treating each in isolation. 

“The abatement impact of a particular measure depends on what else you’re doing. You need to look at these jointly,” Reichelstein said to the GSB.

Applied to the European cement industry, the findings show that a modest increase in carbon permit prices could incentivize producers to cut emissions by up to 96%, with production costs rising only about 12% — far below prior estimates that full decarbonization would double cement prices. 

This research is also relevant to the broader climate policy space, where many have raised concerns about the detrimental economic impact of reducing carbon emissions in general. 

“Our research suggests that such worries may be significantly overstated, at least for cement, and potentially for other hard-to-abate industries,” Gunther Glenk, research team member and assistant professor of business at University of Mannheim, said to the GSB. “The point to recognize is that firms can avoid paying higher carbon charges by investing in carbon abatement technologies, and for a substantial level of abatement, these investments can be made at a relatively moderate cost.”

The post Research Roundup: AI companions, tumor blood tests and decarbonization appeared first on The Stanford Daily.