Responsible in AI — Takeaways

Responsible in AI — Takeaways

I invested in attending the responsible in AI conference this year for my career progression. My learning strategy is to learn the big picture, which helps me provide assessments of products. So I hope this investment will pay off in the long run.

As all conferences go there is an overwhelming amount of information to take in. I’ll start by sharing some key ideas that you can takeaway to use, adapt and reinvent.

Different types of AI

• There are different types of AI that you need to think about — narrow, broader Ai, Genetic AI. Each of these possess a different type of challenge. For example Agentic AI has introduced new challenges such as autonomy misalignment, value alignment, memory poisoning or tool misuse. So have a think about the type of AI you are using either personally or in your workplace.

AI and application Lifecycle

• If you’re at the beginning of an AI journey in your company you could use the AI and Application lifecycle ISO/IEC 5338. You have to pay for this, however, if you do a search you can find free information. When I did this I discovered a good diagram on this via here. It could get you started and can amend as you see fit. ISO/IEC 5338: Get to know the global standard on AI systems — SIG

AI Readiness checklist

• Similar to a release checklist or definition of done you could create an agent readiness checklist — example that I’ve come up with (add and amend your checklist as you see fit)

1. Business value — Understanding the business use case
2. Stakeholders needs and requirements validated and verified
3. Security reviewed
4. Data quality reviewed
5. Risks documented and mitigated
6. Evidence of linking up governance to technical requirements

Security — Red teaming

• Red teaming is a practice that simulates real world cyberattacks to test an organisation’s security measures. As an idea why not get together with other professionals and host a simulation. You could do a threat modelling exercise where you review the architecture and explore security risks or do a practical hands on exploratory.
• From talks on security I did my own research and discovered that there is the start of the OWASP AI Testing Guide that you can find here: OWASP AI Testing Guide | OWASP Foundation