FOUR TOP GOVERNANCE TIPS FOR PROCURING AN AI SYSTEM

First, the good news, Australia’s new Ai governance standard brings together a set of best practices for procuring, designing, developing, deploying, and using Ai systems.

Now the very good news, the Ai standard works alongside existing procurement processes! As you are well aware, an Ai boom is upon us. So there’s an endless parade of Ai solutions to choose, from to solve all sorts of business problems.

Sourcing an Ai system has the same key steps as traditional systems, but there are some important (deal breaking) factors buyers must be aware of before diving into an Ai solution.

Tip 1: Understand the procurement aberrations of Ai systems

Unlike traditional systems which consist of fixed features and functions, Ai solutions evolve: they may drift, degrade or develop bias. You are actually buying potential not a product. You are not purchasing licenses but a continuous vendor relationship. Ensure your procurement process integrates the Ai governance standard.

Tip 2: Determine who is accountable for applying the Ai Standards

The standard is applicable regardless of whether a company develops an Ai system in-house or buys one off the shelf. Ai standard application may require agreement with a vendor or rigorous testing and monitoring by you. Here's a couple of examples:

If you are procuring a pre-trained large language model, you and the vendor should agree on who is accountable for applying the data and training standards. Fine-tuning the model would actually transfer the some accountability from the vendor to you.

If you are procuring a cloud Ai system, such as Copilot, then you and the vendor should agree on who is accountable for applying the relevant Ai standards (i.e. data, training, evaluate, integrate and deployment). We are not sure how many organisations using Copilot today would be aware of this, let alone have an understanding of what they might be on the hook for if something goes wrong.

Tip 3: Translate the standards into specifications

The challenge here is turning the Ai technical statements into practical procurement requirements. It's best to intelligently group and prioritise those most important to your procurement decision. For instance, Ai output accuracy may depend on training data quality (Statement 17) whereas Statement 40 (Decommission), can be handled by standard contract clauses.

Tip 4: Contracting

Ensure any governance accountability is properly defined in your vendor agreement before deploying anything.

In our experience, projects either end quickly, or end well, when everyone is clear on their respective accountabilities. A good procurement process is the starting point for deploying responsible and safe Ai systems, ensuring a clear pathway from concept to production to decommission.

 Learn more about how we help you implement safe and responsible Ai governance practices by emailing us: info@aigovernancepartners.com.au.