Concept 4 of 4

Responsible AI in practice

1 question test this

Responsible AI is often discussed as ethics and examined as engineering. The syllabus names six properties and expects you to know how each is detected and what is done about it.

The six properties

Bias is systematically different treatment of groups. It arrives from the training data, from how the problem was framed, and from who was in the room.

Fairness is whether that difference is defensible. The formal definitions conflict, so this stays a judgement about the situation rather than a threshold.

Inclusivity is whether the system works for the range of people who will use it, including those poorly represented in training data.

Robustness is whether it holds up on input it was not designed for, and under deliberate attempts to break it.

Safety is whether it can cause harm, directly through its output or indirectly through what people do with it.

Veracity is whether the output is true, which for generative systems is a property that has to be engineered rather than assumed.

Where bias comes from

The data. Historical records encode historical decisions. A hiring model trained on who was hired before learns who was hired before, including the reasons nobody would defend.

Representation. Groups thin in the training data get worse results, and an aggregate accuracy figure hides this completely.

Proxies. This is the one that catches people. Removing a protected attribute does not remove its influence, because other fields correlate with it. Dropping the column removes your ability to measure the gap while leaving the gap.

Feedback loops. A model that predicts where problems will be found sends people there, they find problems there, and that becomes tomorrow's training data.

The way you see any of it is by measuring performance separately for each group. A model at 94 per cent overall can be at 97 for one group and 71 for another, and the aggregate number will never tell you.

Bias and variance

The statistical pair, which the syllabus also names. High bias is a model too simple for the pattern, underfitting and doing poorly everywhere. High variance is a model too sensitive to its training data, overfitting and failing on anything new. Both damage subgroups first, because that is where the data is thinnest.

Datasets

The characteristics that matter are inclusivity, diversity, curated sources and balance. The practical reading is that curation is where most of the improvement is available, and it is unglamorous work done before any model exists.

Legal exposure

Generative AI brings risks the syllabus lists directly.

Intellectual property. Models trained on scraped material can produce output resembling their training data, and who owns generated output is unsettled in several jurisdictions.

Biased output creates discrimination liability where the system touches employment, credit, housing or insurance.

Hallucination becomes misrepresentation once a customer relies on it.

End user risk is what somebody does with confidently wrong advice.

Loss of trust is the one with no legal remedy and the longest tail. A system caught being wrong in public rarely gets a second hearing.

The tooling, whichever platform you are on

Four capabilities matter and every major platform has its own name for each, so it is worth holding the capability and treating the product name as a detail.

Bias measurement, before and after training, broken down by subgroup. AWS calls this SageMaker Clarify, Azure has Responsible AI dashboards within Azure Machine Learning, and Google exposes it through the evaluation tooling in Agent Platform. Open source alternatives such as Fairlearn and AIF360 do the same job and run anywhere.

Drift and quality monitoring after deployment. SageMaker Model Monitor, Azure Monitor for machine learning, Agent Platform model monitoring, or Evidently and WhyLabs if you would rather not be tied to a platform.

Content filtering on both input and output. Guardrails for Amazon Bedrock, Azure AI Content Safety, and Google's safety settings and Model Armor. All of them filter categories at configurable thresholds, and where the threshold sits is a product decision rather than a technical one.

Routing to a person where confidence is low. Amazon Augmented AI is the named service, and most teams build this themselves, because it is a queue and a rule rather than a technology.

Sustainability belongs here too, as a responsible AI consideration. Model size, training runs and inference volume all carry an energy cost, and choosing a smaller model that does the job is the decision that follows.

Practise this

You need one feature that is already live, a blank page, and fifteen minutes.

Name the groups your feature could treat differently, being specific about your own users rather than reaching for a list of protected characteristics. Then answer three questions for each group in full sentences. Write the third one last and slowly, because an answer describing what somebody would have to work out for themselves is not a route to appeal.

Feature ..........................................
What it decides or produces ......................

Group 1 ..........................................
  1. How could this group end up worse off than the others?
  2. How would we find out, from which measurement, broken
     down by what, and who reads it?
  3. If it happened to one of them, what could that person
     do about it, and who would answer them?

Group 2 ..........................................
  1. ..............................................
  2. ..............................................
  3. ..............................................

Groups worth naming
  people the training data holds least about, people whose
  input is phrased unusually, people working in a second
  language, people the decision matters most to

The third question is where this exercise usually stops. Most teams can name a group and can describe a measurement they would run, then discover the affected person has no route at all, or a route into a support queue with nobody in it who can overturn an automated decision. Question two fails a close second, since measuring by subgroup needs the group recorded somewhere, and a team that dropped the attribute in order to be fair has removed its own ability to check.

Written out for one real feature, responsible AI stops being six abstract properties and becomes three questions with names and gaps attached to them.

Common misconceptions

Removing protected attributes from the training data removes bias.

Other fields stand in for them. Postcode carries ethnicity, first name carries gender, purchase history carries both. Dropping the column removes your ability to measure the disparity while leaving the disparity in place.

Fairness is a threshold a model either meets or fails.

The formal definitions of fairness conflict with each other, and it has been shown that several cannot hold at once except in degenerate cases. Which definition applies is a decision about the situation, not a setting.

Responsible AI is a review that happens before launch.

Bias appears as populations shift, so a system fair at launch can drift. It is a monitoring commitment with a named owner rather than a gate you pass through once.

1 question test this concept

A model predicts where problems will be found, teams are sent there, they find problems there, and that becomes the next training set. What is this dynamic called?

  • AOverfitting, since the model is learning its own outputs.
  • BDrift, since the input distribution moves over time.
  • CMembership inference, since the model can identify its own training data.
  • DA feedback loop, where acting on the prediction generates the data that confirms it.
Check whether it stuck.

One per page, with a worked explanation.

Start the set
Related material
Book
Weapons of Math Destruction, On models that scale harm quietly.
Book
Data and Goliath, On data collection and the asymmetry it creates.
Template
Experiment brief, The belief being tested, a hypothesis with a threshold and a date, how long the test runs and on how much traffic, and a table saying in advance what you will do with each possible result.