Risk & safety · Research field
Present-day AI harms
Catalogued:
What it means
An umbrella for observable harms associated with current systems: discriminatory decisions, privacy failures, fraud, unsafe outputs, exploitation and unequal burdens. These are different problems, not one mechanism. Attribution requires examining the specific system and the institutions deploying it.
The strongest case
Concrete cases make responsibility and remedies more tractable. People already affected should not have to compete with speculative future scenarios for the right to be taken seriously.
The difficult part
Not every social problem involving software is caused by the model alone. Evidence must distinguish system failures from baseline failures and identify affected groups. A focus on current harms need not deny future risks; the two can share causes and remedies.
Sources & further reading
Sources support the factual context. The jokes are our responsibility.