Safety and Best Practices

We know many people are smart contract developers used to working in EVM environments that are fully deterministic, and may not be as familiar with working on systems that are not (like machine learning models). Working with ML models requires a change in perspective when it comes to integrating the systems in environments that previously have been deterministic.

ML products are stochastic in nature, therefore they might in the extreme exhibit unexpected behavior. As such, here are some standard best practices for integrating ML products into any system.

Human in the Loop (HITL)

Wherever possible, we recommend that a human review the outputs of our model. This is critical for high-stakes domains where bad decisions may have severe consequences.

Red hat testing

We recommend that you test for adversarial actors. Think about how a bad actor might exploit possible vulnerabilities. A good general rule is to run through your systems from the perspective of someone trying to cause harm.

Allow users to report issues

Users should generally have an easily-available method for reporting improper functionality or other concerns about application behavior (listed email address, ticket submission method, etc).