Responsible AI Practices

This image has an empty alt attribute; its file name is ML1-1024x670.jpeg

The development of AI is creating new opportunities to improve the lives of people around the world, from business to healthcare to education. It is also raising new questions about the best way to build fairness, interpretability, privacy, and security into these systems.

These questions are far from solved, and in fact are active areas of research and development. Google is committed to making progress in the responsible development of AI and to sharing knowledge, research, tools, datasets, and other resources with the larger community. Below we share some of our current work and recommended practices. As with all of our research, we will take our latest findings into account, work to incorporate them as appropriate, and adapt as we learn more over time.

General recommended practices for AI

Reliable, effective user-centered AI systems should be designed following general best practices for software systems, together with practices that address considerations unique to machine learning. Our top recommendations are outlined below, with additional resources for further reading.

Recommended practices

Use a human-centered design approach

The way actual users experience your system is essential to assessing the true impact of its predictions, recommendations, and decisions.

  • Design features with appropriate disclosures built-in: clarity and control is crucial to a good user experience.
  • Consider augmentation and assistance: producing a single answer can be appropriate where there is a high probability that the answer satisfies a diversity of users and use cases. In other cases, it may be optimal for your system to suggest a few options to the user. Technically, it is much more difficult to achieve good precision at one answer ([email protected]) versus precision at a few answers (e.g., [email protected]).
  • Model potential adverse feedback early in the design process, followed by specific live testing and iteration for a small fraction of traffic before full deployment.
  • Engage with a diverse set of users and use-case scenarios, and incorporate feedback before and throughout project development. This will build a rich variety of user perspectives into the project and increase the number of people who benefit from the technology.

Identify multiple metrics to assess training and monitoring

The use of several metrics rather than a single one will help you to understand tradeoffs between different kinds of errors and experiences.

  • Consider metrics including feedback from user surveys, quantities that track overall system performance and short- and long-term product heath (e.g., click-through rate and customer lifetime value, respectively), and false positive and false negative rates sliced across different subgroups.
  • Ensure that your metrics are appropriate for the context and goals of your system, e.g., a fire alarm system should have high recall, even if that means the occasional false alarm.

When possible, directly examine your raw data

ML models will reflect the data they are trained on, so analyze your raw data carefully to ensure you understand it. In cases where this is not possible, e.g., with sensitive raw data, understand your input data as much as possible while respecting privacy; for example by computing aggregate, anonymized summaries.

  • Does your data contain any mistakes (e.g., missing values, incorrect labels)?
  • Is your data sampled in a way that represents your users (e.g., will be used for all ages, but you only have training data from senior citizens) and the real-world setting (e.g., will be used year-round, but you only have training data from the summer)? Is the data accurate?
  • Training-serving skew—the difference between performance during training and performance during serving—is a persistent challenge. During training, try to identify potential skews and work to address them, including by adjusting your training data or objective function. During evaluation, continue to try to get evaluation data that is as representative as possible of the deployed setting.
  • Are any features in your model redundant or unnecessary? Use the simplest model that meets your performance goals.
  • For supervised systems, consider the relationship between the data labels you have, and the items you are trying to predict. If you are using a data label X as a proxy to predict a label Y, in which cases is the gap between X and Y problematic?
  • Data bias is another important consideration; learn more in practices on AI and fairness.

Understand the limitations of your dataset and model

  • A model trained to detect correlations should not be used to make causal inferences, or imply that it can. E.g., your model may learn that people who buy basketball shoes are taller on average, but this does not mean that a user who buys basketball shoes will become taller as a result.
  • Machine learning models today are largely a reflection of the patterns of their training data. It is therefore important to communicate the scope and coverage of the training, hence clarifying the capability and limitations of the models. E.g., a shoe detector trained with stock photos can work best with stock photos but has limited capability when tested with user-generated cellphone photos.
  • Communicate limitations to users where possible. For example, an app that uses ML to recognize specific bird species might communicate that the model was trained on a small set of images from a specific region of the world. By better educating the user, you may also improve the feedback provided from users about your feature or application.

Test, Test, Test

Learn from software engineering best test practices and quality engineering to make sure the AI system is working as intended and can be trusted.

  • Conduct rigorous unit tests to test each component of the system in isolation.
  • Conduct integration tests to understand how individual ML components interact with other parts of the overall system.
  • Proactively detect input drift by testing the statistics of the inputs to the AI system to make sure they are not changing in unexpected ways.
  • Use a gold standard dataset to test the system and ensure that it continues to behave as expected. Update this test set regularly in line with changing users and use cases, and to reduce the likelihood of training on the test set.
  • Conduct iterative user testing to incorporate a diverse set of users’ needs in the development cycles.
  • Apply the quality engineering principle of poka-yoke: build quality checks into a system, so that unintended failures either cannot happen or trigger an immediate response (e.g., if an important feature is unexpectedly missing, the AI system won’t output a prediction).

Continue to monitor and update the system after deployment

Continued monitoring will ensure your model takes real-world performance and user feedback (e.g., happiness tracking surveysHEART framework) into account.

  • Issues will occur: any model of the world is imperfect almost by definition. Build time into your product roadmap to allow you to address issues.
  • Consider both short- and long-term solutions to issues. A simple fix (e.g., blacklisting or whitelisting) may help to solve a problem quickly, but may not be the optimal solution in the long run. Balance short-term simple fixes with longer-term learned solutions.
  • Before updating a deployed model, analyze how the candidate and deployed models differ, and how the update will affect the overall system quality and user experience.

Source link

- Advertisment -

Most Popular

Improving Robots’ performance with Machine Learning

A small drone takes a test flight through a space filled with randomly placed cardboard cylinders acting as stand-ins for trees, people or structures....

Combination of Robots, AI and Data Analytics in your local Supermarket

Robots patrolling grocery store aisles and warehouses; so-called dark stores dedicated to online-only orders; data crunched in the cloud that allows retailers to identify and even...

5 (Most Common) Mistakes New Data Scientists Must Avoid

Emerging technologies like data science, machine learning, artificial intelligence are exploding by giving new dimensions to its applications. With business booming into data-driven technologies...

Using Blockchain to manage the supply chain COVID-19 vaccine

Blockchain could play an essential role in the distribution of the COVID-19 vaccine. Tackling COVID-19 will require the first-ever deployment of blockchain in the...

Role of Artificial intelligence in IVF

IVF is a physically and emotionally draining process and success isn’t guaranteed. But machine learning technology could improve the odds for couples trying to...

Machine Learning a major part of Google Sheets

It’s been a while since the first version of BigML’s add-on for Google Sheets. The post announcing it described how one could add predictions...
- Advertisment -