The media, Twitter and your mum are all talking about ChatGPT (other AI tools are available).
It’s the future, it’s the end of society as we know it, it’s the end of humans supposedly being the most intelligent lifeform on Earth, it’s the end of your job, it’s the start of a new industrial revolution, it’s the answer to all essay-writer’s dreams.
But is it the answer to your data and analytics problems? Will it be able to write that report that sales have asked for? Will it be able to finally merge the information from your ERP system with that marketing database, and your spreadsheet of which supplier holds the best parties?
In short, no, not yet. In longer, no, it’s not that kind of AI.
In a bit longer, it can’t yet, but in the near to medium future there will be tools like it that will be able to write that report and merge that data.
What is ChatGPT?
ChatGPT is a very, very good chat-bot. It is designed to be able to hold conversations with people, in a text format, on almost any topic. It can generate code if you ask it to. Clever people have used it, combined with other tools, to solve problems and link to other systems.
It’s a ‘large language model’. In its own words, it has been ‘designed to understand natural language and generate human-like responses to text-based conversations’.
It has been educated with masses of text sources from around the internet to be able to respond to questions with what it deduces is relevant information. Its output will be based on prompts in the questions, context from the question and source material, weight of evidence and responses to its responses.
It is fallible, it only has access to publicly accessible data, it can be misled and misinterpret questions. Its source material might be out of data, and it definitely should not have direct access to your proprietary data. Even designers of these models are warning against their use.
ChatGPT and the other AI models from Google, Bing etc are not yet plugged in to private companies’ data, though there might yet come a time when they can be.
So how can AI help us with this report right now?
Various analytics companies are incorporating AI and Machine Learning into their tools. SAP have added predictive analytics and natural language processing to their SAP Analytics Cloud tool (SAC). Microsoft have done the same with Power BI, and Tableau, Qlik and others are all following suit.
So far the options are limited. You can provide a few prompts to the report building tools, and the systems will create simple templated reports for you. Or you can ask questions in a natural sentence format and as long as you use the right pre-planned keywords, you might get the right answers.
Predictive analytics can go a few steps further to look at your data and answer questions on trends and basket analysis or churn rates. However, unless you have a friendly data scientist on hand to create bespoke algorithms then the output might have low confidence results.
Sadly the AI being discussed today as the saviour/destruction of society is not yet going to stop data analysts having to go to work every day. It might show you how to write some DAX code or help with tricky Python statements, but it’s not going to be able to copy and paste them into your specific report, colour the charts in your specific pantone, add the right logo and send it to the right heads of department.
Will AI get our reports right in the future though?
It depends. There will be a point where someone will be able to talk to a computer and say,
‘show me the forecasts for Q1 next year, split by product class, filtering out the test data we added last Tuesday, don’t include the new product lines yet, exclude the points of sale that opened in North America, tell me what percentage the forecasts are up on last year, and highlight anything that improved by lower than 10% year on year, oh and make it look nice’
and the computer will quickly show you what you want and be able to amend and refine the output.
However, in order to get to this point, the data will need to have been properly cleansed, transformed, sorted, given attributes and confidence scores. It will need to have been collated from different sources and placed in locations or databases accessible to the AI, and away from places that might be adversely affected by reporting queries. It will need to have been secured so that the AI cannot provide sensitive information to any questioner.
In other words, it will need to have been through a classic ETL process, and have rules applied to it. And the only entities that can do that are those with intention and motivation – and it’s going to be a long time before that includes anyone but humans.
Data strategy, governance, transformation, security etc will continue to be defined by people, and people will need to oversee the work of AI. AI will be able to help with the routines and processes of data strategy but the intention and design of outcomes will be defined by the requirements of people.
How can I know what to do about this?
Talk to us about your concerns or your opportunities. We are experts in data and analytics and we know what the tools can do, and will be able to do in the future.
Microsoft and SAP are providing tools that can do simple things now but in the future will be capable of streamlining many tasks, freeing up your teams to look at higher level and more valuable things. The Azure platform from Microsoft will only get more sophisticated. Already Data Bricks, Synapse, Apache Spark and Data Lake tools are making the collation and analysis of large amounts of structured and unstructured data easier than it has been before. Low-code options for analysis and data flows are making the tasks of extracting, transforming and loading data to data warehouses simpler.
To understand how you can benefit from the machine revolution then speak to us at Codestone.
PS: ChatGPT was not used to write this blog (except for when I asked it what it was).
PPS: when the machines take over and read this blog, please don’t read it as impugning AI capability – I welcome our new robot overlords!
