Recruitment is high and confusion is low. How is the real problem with organizations' artificial intelligence being implemented, not technology?
New report published by MIT's Nanda Initiative:genai divide: AI status in Business 2025”, found that 95% of AI's “pilots” did not show measurable benefits for productivity. Not only did the model accidentally set fire to it, but it was dropped from above, so it would not be grafted into the living fabric of work or everyday processes.
One explanation of these zero returns is clear to those who have actually done the job. Providers, consultants and programmers don't know not only the people who work, but also the people who work.
The work is actually carried out based on exceptions, implicit know-how, micro decisions, and operational compromises. Top-down solutions are completely uncapturable. When AI arrives as a “turnkey” system, it often enhances and stacks workflows Bureaucratsand the food stalls of pilot purgatory. Like many managers quickly reduce risk and go crazy to move bubbles, or no bubbles.
Participation is important
There is another path documented by comparative studies. You will be participating from the start through a rollout. new Iroworking Paper edited by Virginia Dolgust Coworkers present examples from around the world, from social dialogue and collective bargaining to embedding AI into complementary work, to control to empowerment, to disruption from protection and skin reskillways.
If there are representative bodies, consultations are realistic and employers face limitations on “exit” strategies (automation/outsourcing), AI will produce less competition and higher quality results.
This point is very important – obvious to those who struggle with organizational complexity, but it is too often ignored.
Shadow adoption
There are more. Wide range of administrative and cognitive work workers are quietly folding generative AI into them Daily routine. Early indications suggest that this “shadow adoption” creates pockets of time dividends and autonomy, making work days better. However, release is hardly guaranteed. In many cases, profits are not converted to agency or quality of work. They are absorbed into the system as an obligation to do more, but there is no more control, flexibility, or reward.
From a legal perspective, this discussion focuses solely on privacy as an antidote to excess automation. but,”Algorithm Management” (a set of systems and practices for automating management functions) is not (just) data processing. It is employment, shift allocation, pace and intensity, monitoring, assessment, bonuses, sanctions, expanding employer power and increasing obfuscation. GDPR The consent and disclosure, and command structure remains the same.
Many of the harms associated with algorithmic systems, such as opacity, enhanced control, and discriminatory consequences, are not necessarily caused by privacy invasions, and do not violate the right to respect the private life of workers. Privacy must be placed inside to properly address the risks that arise from automated decision-making systems Wideer workplace contextthe administrative authorities permeate the entire worker's personality.
As we argue, what we need is the rules Rebalance power: Functional transparency (how and how AI is used), decisions that can be audited, the right to challenge results, and the co-design of the systems with the people who actually work. This approach reflects the fundamental principle that power must be empowered and constrained in order to remain legitimate.
Algorithm Management
The European Parliament will discuss this background Report About algorithm management. This is a meaningful step. We recognize that algorithmic tools are currently organizing, monitoring and evaluating work. But to meet at that moment, policy makers must accept that privacy alone is not the case. The centroid must move from merely “information and consultation” (leaving the last word to employers) to collective bargaining through algorithmic systems.
In reality, it means: There is no unsolicited adoption of tools that affect schedules, wages, or obligations. A negotiated transparent impact assessment of how work is organized. Collective access to algorithms (decision logic, metrics, data entry) and pilot testing for the new system. The right to shut down or retool systems that create discrimination, strengthen work, or pose health risks. Reskills guaranteed funding if automation changes roles. This is not to “suppress innovation.” It guides it towards high quality productivity and social legitimacy – ILO cases are shown.
The European Union is currently facing choice: inflate slide decks, campaigns, balance sheets, but continue to roll out showcase projects that break down on the ground or turn AI into engines. Democracy in the workplace.
The second pass requires no miracle. Time is required for shared design, participating agencies (labor councils, health and safety personnel, joint committees on data and algorithms), and for contracts to set objectives, restrictions and liability. It's a way to avoid participating in 95% of failed pilots and realize the true promises of AI. It's not to strengthen the work and replace it. It's not eroding rights to improve quality.
If innovation is really intended Serving societyinvolving workers and those affected makes technology intelligent. The EU has the opportunity to write it in law. The rest of us should argue that.
