Do AI Models Really Need to Forget Everything? Apple Researchers Say No
A new study finds that much of what we ask AI to 'unlearn' barely mattered to the model in the first place, which could cut the cost of privacy fixes dramatically.
Key points
- Apple ML Research identified subsets of training data that have so little influence on a model's outputs that removing them may be unnecessary.
- Current AI unlearning methods treat every piece of data equally, even when some data had almost no effect on how the model behaves.
- Skipping the removal of these low-impact data points could reduce the computing time and cost of privacy-related model updates.
- The findings apply to both language models (the kind that power chatbots) and vision models (AI that analyses images).
When you ask a company to delete your data, you probably assume the AI they trained on it will change. Researchers at Apple ML Research now suggest that assumption is often wrong, and that gap between expectation and reality could actually save a lot of money.
Here is the background. AI models learn by processing enormous amounts of data. After training, some of that data leaves a strong fingerprint on how the model behaves. Other data barely registers at all. The new research focuses on that second category.
The technical term researchers use is machine unlearning, which means teaching a trained AI model to behave as if it never saw a specific piece of data. Privacy laws in several countries already push companies toward this capability. The problem is that current unlearning methods are expensive. They treat every item in the "forget list" the same way, whether that item shaped the model heavily or barely touched it.
The Apple team used a tool called influence functions, a technique that measures how much any single training example actually shifted the model's final behaviour. Think of it like checking receipts after a party: some purchases moved the needle on the total bill, others were so small they were noise. The researchers found that a meaningful share of training data falls into the noise category.
Their argument is direct: if a data point had negligible impact on the model, skipping its formal removal produces almost identical results to doing the full deletion, at a fraction of the computing cost. GPUs, the specialised chips that do the heavy number-crunching AI needs, are expensive to run. Any method that reduces GPU hours translates into real money saved.
The finding held up across two different types of AI tasks: language and image recognition.
Does this mean companies can ignore deletion requests?
No. The research does not give companies a legal pass to ignore privacy requests. What it does suggest is that engineers building unlearning tools could prioritise high-influence data points and safely fast-track low-influence ones, cutting costs without harming the outcome.
For ordinary people, the honest implication is this: the data you contribute to an AI system does not affect that system equally. Your interaction might be formative, or it might be background static.
Survivorship bias is worth flagging here too. The researchers identified the low-impact cases, but the hard and expensive work remains for high-impact data, exactly the sensitive records most likely to matter in a real privacy dispute.
Takeaway: If you work in a business that handles AI models and user data, ask your technical team which data in your training set actually moves the needle. You may be paying to "forget" things the model never really learned.
