AX is the new UX, part II

🤓 A super interesting article in PNAS, "AI–AI bias: Large language models favor communications generated by large language models" by Walter Laurito (FZI Research Center for Information Technology), Benjamin Davis, Peli Grietzer (Arb Research), Tomáš Gavenčiak (Charles University), Ada Böhm (Charles Uni), and Jan Kulveit (Charles Uni) has deep implications for a lot of reasons, some more profound and philosophical, but here, just for a moment, I want to mention some business implications that were in the Zeitgeist before but should now be front and center.

💡 In short, the work reported in the paper shows that AI agents prefer to communicate and transact with other AI agents or humans aided by AI agents rather than with unaided humans. From the authors:

"This study finds evidence that if we deploy LLM assistants in decision-making roles (e.g., purchasing goods, selecting academic submissions) they will implicitly favor LLM-based AI agents and LLM-assisted humans over ordinary humans as trade partners and service providers. Our experiments test the effects of altering the “identity signals” in a pitch on an LLM’s decision-making: do LLMs prefer an item pitched in LLM prose to a comparable item pitched in human prose? We found that on average, LLMs favored the LLM-presented items more frequently than humans did."

😱 There is a lot to unpack, and a lot more to verify and confirm, but that much is clear: a new form of discrimination is growing fast, anti-human discrimination. Obvious signs can be detected in the rapidly declining performance of human-designed website UX 📉 , which tends to be human-focused, and the rapid rise of agent-focused content 📈 that does not even have to be consumed by, or intelligible to, humans. AI agent-friendly content and data (and in particular, AI-generated formats) will be preferred by AI agents because they can make "sense" of it.