It has been about 5 months since I last blogged on AI and so much has happened, there is a lot to ponder and learn.
In my specific context, which is relatively low volume but high complexity work (think thousands of transactions or processes not tens to thousands or millions) there are some specific things that we are feeling more and more confident about.
- New capabilities
AI is enabling us to do things that weren’t previously possible, or makes the cost of doing them within reach. It is driving more in terms of capability, than it is in terms of efficiency. For example, on business.gov.uk we’re now able to show businesses government funding opportunities, personalised to their sector and region. This is made possible by AI tools aggregating and personalising what was previously disparate, unstructured data across multiple websites and sources. It would have been too difficult and expensive to do this before. - It’s all about the data
Data quality, infrastructure and governance are absolutely critical. Getting these right unlocks more value than anything else. These are difficult to build, hard to persuade people of their importance and slow to achieve. If you don’t get these right your AI work will be drawing on incorrect, out of date information. - What does sovereignty mean for you?
Managing sovereignty has many facets and many approaches. We are just at the beginning of thinking this through in a meaningful way. Does it mean who owns the models, or where the data is processed, or who has their hand on the ‘off’ button for the data centre? All of the above or none. Consensus is some way off, but discussions should be happening now in organisations, before it’s too late to make strategic choices. - Cost control
Understanding cost in this space is incredibly tricky with pricing changing fast, and token usage highly variable between models and use cases. Rigorously evaluating value for money is vital. Don’t try doing this on the back of an envelope, get the analytical professionals in to make sure you look at this robustly. - AI doesn’t mean LLMs for most people
When people say “AI” it’s not just about large language models, it’s about so much more. For most it is now a generic term that means automation, machine learning, modelling big data and general technology transformation. This is an opportunity for digital, data & tech people to influence at the top table, use it! - We still need to build right
Through all of this, we need to keep building in the right way. We still need multidisciplinary working that puts user needs at the core of our work that focusses on outcomes. Building secure by design remains as important as ever. This is still a team sport, and we’re still building critical systems which need to be trustworthy and reliable.
Those are my thoughts at the moment. What are yours?
Jason Kitcat 