The 'how' and 'why' of generative AI innovation
Democratisation, use cases and the concept of data innovation for good
The London leg of the Databricks Data + AI World Tour took place on Wednesday. The theme? “Generation AI”
Whilst the focus on generative AI came as a surprise to no one, the event focused on the 'how' of data and AI driven transformation with an abundance of use cases and industry focused discussion from companies including AXA UK, Legal and General, WPP, Evri, Drax, Haleon and Octopus Energy, blended with multiple panel discussions and more technical product focused sessions and training opportunities.
Before the 'how' comes the 'why' and David Meyer, SVP Product began his update on Databricks product strategy with an explanation of why democratisation is central to recent Databricks product developments.
"We decided we need to democratise AI in every product we build so that it can be in every product you build," said Meyer.
LakehouseIQ which launched in the summer builds on a lot of the industry specific work that Databricks has put in. LakehouseIQ is an AI powered engine which learns company and industry specific terminology and patterns and enables employees to search and query data using natural language. No coding necessary.
Lakehouse AI will enable Databricks customers to build and deploy their own AI models, a key part of which is the integration of MosaicML, acquired by Databricks for its LLM training expertise.
"The thing I'm most excited about is the MLflow AI gateway," said Meyer. "Most companies are playing with OpenAI today and calling OpenAI interfaces and writing code that is bound to a model. But the model they're using today will likely be replaced in a few months. What the AI gateway lets you do is call OpenAI, different models from the gateway but it's abstracted so later, you can swap he model out with no change in code.
"Databricks plus Mosaic is a very powerful platform and makes Lakehouse the best place to do generative AI and the most future proof."
Industry specific applications
Paul Hollands, Chief Data & Analytics Officer, AXA UK & Ireland, provided insight about how the insurance industry, perhaps historically more concerned with building a strong data foundation and not the fastest to innovate technically, was now very much leaning into that prospect.
Hollands said:
"We have a moral obligation to our customers to provide great service and great outcomes, so we need to answer the question 'what's' the value story?' This allows us to make this fundamental step change and lead in terms of how we serve customer support colleagues and ultimately drive value for the business."
One use case for generative AI in AXA that Hollands shared was its use in AXA's retail business whereby customer agents have a tool to enable them to surface policy information much faster, and to establish specific criteria on coverage, excess etc. Hollands emphasised that this was not about outsourcing jobs to chatbots, it was about augmenting what customer service agents could deliver, helping to reduce the mundane work and ultimately improving service levels.
Another use case concerned the ongoing issue with RAAC and its use in commercial properties.
A panel chaired by Databricks Field CTO Robin Sutara, focused on how innovation and data champions could try to balance the risk that public LLMs and use of AI potentially create, without killing the potential of the same tools to propagate innovation, empowerment and creativity.
Emma Duckworth, Head of Data Science at Haleon, emphasised a people-centric approach.
"We're focused on consumer healthcare. We have to provide a clear strategy on how this help our consumers and employees, use cases that we want to explore first, tools we're going to do it with, capabilities that we will help people across the organisation grow in very practical ways. This is the strategy we're starting to roll out."
Others took a blunter approach. Di Mayze, Global Head of Data and AI, WPP commented:
"We say that you aren't going to lose your job to AI, but you will lose your job to somebody who uses it, so get involved and get playing with it."
Data for good
Another panel on the subject of gender biases and data for good took a deep dive into the lack of trust that many consumers and organisations holding back data innovation, and the use of that data to drive positive changes in society.
All the panellists agreed that women could benefit from sharing health data to inform pharmaceutical and medical research on the hormonal changes which occur over a woman's lifetime and the impact of menopause in particular. Retail data collected via loyalty cards could be used to identify patterns which occur before certain cancers are diagnosed. But consumers are worried. At best they simply assume that data will be used for targeted advertising, At worst they fear more nefarious applications. Trust in authorities, corporations and governments is at rock bottom.
Mistrust has been fuelled by a lack of transparency in how data will be used and the fact that both parties tend not to get equal value out the exchange. But regardless of whose fault it is, innovation that could prove beneficial is being blocked.
Speaking exclusively to Computing after the panel Robin Sutara said:
"I don't think that corporates can build the trust that we need. We need to think about neutral data aggregators or stewards. Who has or could build that type of trust across the ecosystem to operate in that capacity? It's going to be NGOs and non-profits so people can associate it with a good outcome, but that puts the onus on corporates and individuals to invest in those organisations."
"During the pandemic people were more willing to share data because they could understand the benefits. The more that we can create those kinds of trust relationships then we could really start to see the power of that data."
More discussion on the concept of data and the common good could be found in a panel on data innovation for Net Zero which included representatives from Octopus Energy, Drax, Scottish Water and UK Power Networks.
Whilst some of the initiatives discussed will require huge quantities of data, panellist David Sykes, Head of Data at Octopus, explained how it will benefit consumers. In this case the value exchange is clear, and rather than asking customers to pay more for the privilege of consuming energy from sustainable sources, the idea is that the customer benefits by being flexible.
"If you can provide incentives for customers to consume their energy at the greenest times that's very powerful. We do things like sending out a real time energy price to phones and EVs so they can optimise their usage to minimise price."
Interviews with both AXA UK and Octopus Energy on AI data innovation in their organisations will be published in Computing later this year.