exhausting

Imagine that you hire someone to help you manage your email. Of course, the first week you have to explain to him how you like him filter messages and tell you the most urgent thing. The second week, you correct the mistakes he is making and, for the third week, you have to explain again what you already taught him the first week because he forgot the instructions. At the end of the month, you have a helper, but it takes longer than before because not only do you have to be aware of what he is doing, but you also have to manage your email. That is, in essence, what is happening right now with AI at work, according to the report. Work AI Index of the Glean Institute, carried out by researchers from the universities of Stanford, Berkeley and Notre Dame. According to their findings, employees spend an average of 6.4 hours per week making AI work. Almost one day of work lost every week. Time is not saved, it is transformed. 87% of workers who participated in the study acknowledge that they use AI at work. Of these, 75% affirm that AI makes them more productivesaving them approximately 11 hours per week with automation alone. However, only 13% of companies claim to obtain a real increase in productivity. The gap between what the individual perceives and what the companies’ results show is enormous, and the report has an explanation: those hours do not disappear, they are only redirected towards a new layer of work that no one was taking into account. The authors have called this new task botsitting (a play on words that translates as “bot care”) which consists of little more than “AI kangaroo” to give context to the tool, review errors in the results it generates, relaunch prompts that do not go well and clean up results that seem correct, but are actually full of invented data or hallucinations. As Rebecca Hinds, director of the Work AI Institute, describes, this guardianship is “often tedious and exhausting work”, which no one measures or rewards, so the time that AI saves ends up being a loan that you have to return a few hours later. Too many tools and context switches. The researchers highlight that part of this excess time spent using AI not only comes from reviewing its results but also from how each tool is used. 77% of respondents use multiple AI tools each week, and a third of participants combine four or more. Each jump from one app to another has a cost of time which is rarely countedbut that implies repeating the same instructions or rewriting the prompt on another system because the previous model did not deliver the expected result. Nearly half of workers (46.5%) have to jump between two or more AI tools to complete a single task. Researchers call it the “toggle tax”, the cognitive tax of constantly changing context. Harvard Business Review already calculated the cognitive cost of changing applications and the consulting firm McKinsey calculated that workers waste an average of almost two hours a day searching for information between tools, inboxes and chats. AI, which is sold as the panacea of ​​productivityhas only added a new layer to that chaos instead of reducing it. Of the botsitting to the botshitting. The study found that when the worker spends too much time fixing AI bugs and maintains its delivery deadlines, begins to skip reviewing the results, generating something that the report has called botshitting or “bot crap” which would be delivering work generated by AI without having previously verified. 69% of the participants admitted to having done this at least on occasion. The consequences go beyond the quality of the work itself, when that content reaches the next link in the production chain. without anyone having reviewed itsomeone who did not produce it you have to clean it. That is, both the cost and the time are transferred to another person, but you don’t save that much as it seems. To no one’s surprise, more AI doesn’t solve the problem. Bob Sutton, professor emeritus at Stanford and founding member of the Work AI Institute who produced this report, has pointed out On other occasions, one of the solutions usually taken from management positions when a process generates friction is to add more of that element. In this case, trying to solve a problem of misuse of AI…with more AI. The data in the report suggests that the organizations that are ahead are not those that use AI the most. They are those that have built what the authors call “human infrastructure.” 53% of workers say that the information they need does not come through their AI systems. In companies where it does arrive, employees are 64% less exhausted and are 52% less likely to deliver works that have not been reviewed. In Xataka | OpenAI assures that AI has not had that much impact on employment. Anthropic believes just the opposite and therein lies the problem Image | Unsplash (Flipsnack)