Comments
If AI can hear chickens suffer, is the industry obliged to act?
2
This week, whilst preparing my AI x Animals class, participants were asked to explore a real example of AI already affecting animals. I did the exercise myself and expected to end up with something rather speculative. Instead, I found a system that is already listening to chickens inside commercial farms.
It raised a question I keep coming back to: if a farm's monitoring system flags that its birds may be in distress, what are the moral and practical implications of having that data?
My view is that having the data strengthens the obligation to act. But that obligation lies mostly with whoever can change the birds' conditions, and it will mean little unless it is made formal. There is also another (perhaps less obvious) issue: what the system treats as "normal" may already be a state of distress.
AudioT places microphones throughout broiler farms and analyses the sound continuously. According to the company, the system follows how chickens' calls change as they grow. The technology picks up unusual patterns in different parts of the farm and alerts farmers when something seems wrong.
It is tempting to say the system can "hear suffering" (yes I have used that shorthand in my title to catch your attention). Strictly speaking, it detects sound patterns (and changes in them) that may be linked to welfare or behaviour. How much those sounds tell us about a bird's experience depends on how well the system has been tested... and on what people do once an alert comes in.
The funding is worth noting. AudioT was supported through SMART Broiler, a programme run by the Foundation for Food & Agriculture Research with McDonald's. It received a $300,000 Phase II award, with in-kind support from Tyson Foods. The programme says its technologies could improve welfare for 9 billion birds a year in the US... as well as improving producer profitability.
Two caveats. 1) The 9 billion figure is the programme's potential reach, not the number of birds AudioT monitors today (I couldn't find public deployment figures). 2) I also couldn't find evidence that it has improved welfare at scale. What exists is the company's account of what the system is designed to do. This is enough to take it seriously, but not to call it proven.
AI does not create the obligation to care about chickens. That obligation already exists. And the industry has long known about the welfare problems in broiler farming. What a microphone in the shed adds is precision: a change in this house, in this zone, at this exact moment. That kind of information is harder to dismiss, and it makes "we didn't know" a much weaker excuse.
Receiving an alert, though, is not the same as being able to fix its cause. The alert may reach a farm worker, whilst the breed, the number of birds per house and the age at slaughter are decided further up the supply chain (by integrators and large buyers). If responsibility follows the power to change conditions, it cannot fall solely on the person in the shed.
So, if AI makes suffering harder to overlook, who should be responsible for acting on what it detects? My answer is that it has to be shared, in line with who controls what. Farm workers and growers should respond to acute alerts, such as a failing fan or a sick flock. Integrators, who set the breed, stocking density and slaughter age, should be accountable for the chronic problems the data reveals. Retailers and food companies can make fast responses to alerts and improvements to underlying welfare conditions a requirement of their procurement contracts. Developers should design baselines that don't treat chronic distress as normal. Independent auditors and regulators should check that alerts lead to real improvements for the birds (and not just to better production figures...).
There is also the uncomfortable possibility that, if installing a system makes producers more accountable for what it detects, some may decide not to install one. Others may configure it so that only costly problems trigger an alert. I haven't seen evidence that this is happening, but it is the obvious risk with voluntary monitoring: a duty that applies only to those who sign up is easy to avoid. This is a large part of why I think it needs to be formal and incentivised.
The idea of an "anomaly" also concerns me. AudioT describes flagging changes from what is expected, which works well for sudden events, such as an unusual sound pattern in one zone. But the baseline itself is not neutral. Most broilers are fast-growing breeds, raised to reach slaughter weight in around six weeks (or less). That growth rate is linked to lameness, leg disorders, heart problems and more. As a result, many of these birds are already in some degree of distress before anything unusual happens. If "expected" is learned from these flocks, chronic suffering becomes part of the baseline. The system could then be good at noticing what is unusual, and blind to what is routinely harmful. In other words, the technology would measure distress relative to an already compromised normal.
None of this means the technology is without value. For the birds already in these farms, continuous monitoring could make a real difference. Detecting respiratory illness or heat stress earlier than a few daily walk-throughs would allow farmers to respond faster. Catching a broken fan or a water problem early could prevent birds from suffering and dying. And welfare assessment would depend less on what one person happened to notice during one visit.
In those cases, welfare and profit point in the same direction (often they don't). However, even if AudioT works exactly as intended, the underlying conditions remain: fast-growing breeds, crowded houses, short lives and very little opportunity for natural behaviour. Monitoring could reduce some acute harms whilst making intensive production more efficient. Whether that leads to fewer birds suffering, or to a system that is cheaper to expand, remains an open question...
The problem, then, is not the monitoring itself. It is the risk that "AI-monitored" comes to stand in for "conditions improved". An alert is only valuable if it leads to a meaningful response, and addressing one bad night cannot replace addressing the conditions that cause harm every day.
The action I would prioritise is linking welfare monitoring to clear, enforceable outcomes. Examples would be lower rates of lameness, footpad dermatitis and hock burns, fewer heat-stress events, lower mortality, as well as a maximum time between an alert and a response.
In practice, this could include:
Now is a good moment to push, whilst these tools are still being developed and validated. SMART Broiler's funders have said publicly that the initiative is about both welfare and production efficiency. Advocates can hold them to that: ask which welfare outcomes the technology actually improves, and make sure the data is used to change conditions (not only to manage production).
For now, I think a system like AudioT could help individual birds, provided its alerts are reliable and acted on. Whether such tools improve welfare overall depends on the rules built around them, and on whether they are judged by what happens to the birds rather than by how much more the farm can produce.
As we develop tools that can detect acute distress in real time, what would it take for them, and the industry using them, to also address the chronic suffering already built into these systems? I would especially welcome views from people who work with poultry producers or on precision livestock farming (PLF).
I used AI to help draft and edit this post. The ideas and arguments are my own, and I made the final edits.
