Generative AI and Analytics and BI Tools - State of Play

The current weekend in the UK has been marked by the first heatwave of 2023. As I was basking in the sunshine in our back garden, I was thoroughly enjoying the warmth after what felt like a very long winter. I started thinking about how much the Analytics and BI sector is feeling the warmth or potentially the heat from the phenomenally quickly evolving Generative AI technology area.

There is so much going on and clearly Analytics and BI vendors big and small have already started leveraging foundation models or Large Language Models (LLMs) as they are most commonly referred to. Yet, the breakneck pace of progress in this space is making it nearly impossible for anyone to provide a thorough summary and make solid predictions on where the industry is going to be in ten or even five years. I am not going to even pretend that I have all the answers but thought I can put down some notes on what I have read or listened to on this topic.

Overall Generative AI Trends

There is certainly a cause for concern. "The capabilities of the biggest models have outrun their creators’ understanding and control. That creates risks, of all kinds." (The Economist, 2023). The hope (so far that is the best we have), however, is that humanity is going to place the right guardrails in place to maximise the benefits from AI and mitigate the worst of its threats.

One of my favourite sources of analysis is The Economist who reliably pick out the key factors influencing our world at present and in the future, too. They recently dedicated their Science and Technology section on Generative AI across the following three articles:

1. An explanation of what LLMs are and their potential to transform the world - I particularly like the clear interactive visualisations explaining tokenisation and how the models reason and generate text

2. What are some of the predictions for the near future - what is sure is that applications built around LLMs are going to dominate the near future from a commercial standpoint but scientists also continue experimenting with completely new AI approaches.

3. A deep dive in the risks and what to do about them - the "black boxes" which currently represent AI are the biggest concern and interpretability is going to continue to be the focus of a lot of research with regards to mitigations. Certainly, commercial actors have a big role to play as they are the most enthusiastic proponents of releasing and deploying LLMs.

It's hard to come up with anything conclusive at this stage and the above articles confirm that. They do give a few useful clues though on the salient points in AI today like the significance of the risks, and the importance of explainability and funding for research in effective ethical regulation.

For an almost real-time view of the (r)evolution of LLMs I have had a couple of trackers recommended as follows:

• A leaderboard with very insightful comments related to the second Economist article above.

• A collection of practical guides on LLMs again this is something which is useful for the development of applications leveraging LLMs.

For those that have not got the inclination or time for the more serious reads above, I would definitely recommend watching this episode of Last Week Tonight with John Oliver. It is absolutely hilarious but also covers all the risks and the plausible mitigations.

Like almost everyone else, I have also been experimenting with various AI tools like ChatGPT.

I have found Microsoft Designer to be particularly intuitive and useful. The image to the left was one of the outputs of me asking for a picture on Generative AI and Analytics and BI tools e.g. Microsoft Fabric Co-pilot and ChatGPT.

Hardly revolutionary, but I thought it looks quite cool especially considering that it took just a few seconds to make it.

AI and Analytics & BI

Microsoft has arguably become the first Big Tech name which comes to mind when talking about Generative AI. Even though Google came up with the original concept of deep learning, i.e. artificial neural networks transformer, models, OpenAI owned 49% by Microsoft pioneered their ground-breaking application in the form of ChatGPT.

Microsoft are doubling down on the capabilty GPT has armed them with and are integrating it in the form of Copilot across all of the behemoth's ubiquitous products from Office applications to GitHub to supercharge the productivity of coders. Fabric is the new unified analytics platform which naturally has AI at its core.

Another exciting organisation which has kept AI at the heart of its product from the start is ThoughtSpot. Their most recent Beyond conference revealed a powerful GPT integration giving users the power of pure natural language searches. ThoughtSpot bring the rigour of their tried and tested busines-ready SQL which elevates the clunky business analytics capabilities of GPT when used on its own to enable a truly modern and interactive enterprise-ready experience.

Taking the "rough around the edges" pure LLM capabilities and combining them with more established tools and techniques seems to be the clear way forward at least in the short term. This is a theme in a really insightful review of the AI driven trends and predictions for the future role of Data Analysts. The expectation is that the complexity and the rich human-centred context of organisations using analytics and the impact of erroneous outputs is going to demand significant human involvement for a while yet.

All bets are off though in terms of how all of this is going to pan out. The AI optimists seem to outnumber the pessimists though and I naturally side with the former and I will make sure I document here all the exciting new AI stuff I am learning.