ChatGPT, AI and Microsoft Excel | by Antony Cowburn and Julianna Walthrust

Much of the news around Data and tech has revolved around ChatGPT, Security and access to data. The advancements of ChatGPT have been swift, ongoing and substantial - a quick search on Youtube of “ChatGPT” already returns hundreds of results from the past couple of weeks. Most of these are directed towards learning what ChatGPT is, how to use it and how it is useful. A large amount of the results are about the trepidation surrounding the use and impact of ChatGPT, but it is important to note that ChatGPT is a tool to supplement and not replace the knowledge of the user. In this article, it is our aim to consolidate your knowledge of what ChatGPT is, especially concerning its functionality with Excel and data processing. It is also our aim to expound upon the recent government regulations and ethical repercussions of ChatGPT concerning data.

What is ChatGPT?

Let’s let ChatGPT explain itself. This is an extract from a prompt we entered into ChatGPT. To generate this, we input “Can you write a 1000-word article aimed at 1st-year University students that introduces them to the uses of ChatGPT in a business context? Sections could include ChatGPT use concerning data, especially with Excel, and also a section highlighting any issues that may arise around data governance.”

Here’s an excerpt from the article ChatGPT came up with:

Data is the lifeblood of any business. It is used to inform decisions, monitor performance, and measure success. However, the sheer volume of data that is generated by modern businesses can be overwhelming, and analysing this data can be a time-consuming and complex process. This is where ChatGPT comes in.

ChatGPT can be used to analyse large volumes of data and provide insights that can inform business decisions. For example, ChatGPT can be used to analyse sales data and identify patterns or trends that can inform marketing strategies. It can also be used to analyse customer feedback and identify areas for improvement in customer service. With Copilot's integration with Excel, users can now leverage its capabilities to simplify tasks like data cleaning, analysis, and visualisation. For example, if a user wants to extract a particular set of data from a large Excel sheet, Copilot can suggest code that selects the desired data and performs any necessary operations like filtering or sorting. An example is when creating charts or graphs in Excel. Copilot can help generate code that automates the process of plotting data points, adjusting formatting, and adding labels.

Overall, Copilot's integration with Excel helps streamline and accelerate repetitive programming tasks, saving users time and effort. (source: OpenAI.com) The entire extract can be found in further reading.

What do you think of this description? As you may guess, the more descriptive the prompt, the more descriptive and precise the results. I think “Data is the lifeblood of any business” is an especially interesting term, as it sounds like a somewhat original thought when in reality many people must have said this for ChatGPT to return that statement. Another thing to note is there is also the concept of CoPilot; there is no explanation of what CoPilot is before expounding upon the benefits therein. This is a prime example of the necessity of manual proofreading when using ChatGPT, and being careful with utilising ChatGPT unilaterally.

Excel and CoPilot

Copilot is an AI-powered tool developed by OpenAI that uses natural language processing to generate code and automate programming tasks. Copilot integrates with various programming environments, including Microsoft Excel, to help users write and manage complex spreadsheets more efficiently. With Copilot, Excel users can input natural language commands and receive suggested formulas, macros, and functions to automate calculations and data analysis. Copilot can also autocomplete code and offer insights and recommendations on how to optimise Excel formulas and models. 

With Copilot's integration with Excel, users can now leverage its capabilities to simplify tasks like data cleaning, analysis, and visualisation. For example, if a user wants to extract a particular set of data from a large Excel sheet, Copilot can suggest code that selects the desired data and performs any necessary operations like filtering or sorting. Another example is, when creating charts or graphs in Excel, Copilot can help generate code that automates the process of plotting data points, adjusting formatting, and adding labels. Overall, Copilot's integration with Excel helps streamline and accelerate repetitive programming tasks, saving users time and effort.

Government Response to ChatGPT

The speed with which AI has developed has left the government quickly put into place regulations and understanding of AI. The National Cyber and Security Center officially set forth the following statement on the 14th of March: “The NCSC is aware of other emerging threats (and opportunities) concerning cyber security and adoption of Large Language Models (LLM)s, and we will, of course, keep you informed of these in future blog posts.” This follows the trend of government response, “don’t worry, there is no threat”. And yet just on March 29th, 2023, the UK rolled out a new AI regulator. Describing it as one of the "technologies of tomorrow", the government said AI contributed £3.7bn ($5.6bn) to the UK economy last year. It seems like the profitability of ChatGPT greatly affects the response to it. 

We hope this has elucidated what ChatGPT is and given you some knowledge about how it is integrated into Microsoft Excel. 

An Expert's Opinion

To have a deeper understanding of the way ChatGPT might be integrated with Excel, we reached out to an expert: wearer of many cool hats, Excel MVP, Excel on Fire Youtuber, and author of Guerrilla Data Analysis Using Microsoft Excel: Oz du Soleil. 

Oz du Soleil sees the potential of generative AI, such as ChatGPT, to help save time and increase the accuracy of analytics. He also recommends exercising caution while using these tools, especially for those who are not as familiar with the technology. While discussing the risks that ChatGPT poses to data governance; two distinct areas were brought up. Whether the data fed into the tool was appropriately sourced and trustworthy, and is the output of the tool was suitably accurate and verifiable. 

As new as this technology is, the laws around consent for data use are either non-functional or non-existent in many cases. AdTech and MarTech are able to define or redefine consent in ways that best suit them and drive the most profits.

Data Privacy

The way that many generative AI tools work is they take an original document and break it up into component pieces, or add so much noise that it is unrecognisable from the original. Then they build it back up to a document that represents but is fundamentally different from the original. They then train their model on these new documents. By using this technique, they can claim that the new documents are their own proprietary data; and therefore the data belongs to them. Oz reflected that in cases where health data has been sold to these companies, or where there have been data leaks from Samsung employees to ChatGPT’s Open AI, there is very little the original data owner can do. “Their privacy policy is such that the data can't be deleted [or corrected]. It belongs to ChatGPT.”

Data Accuracy & Validity 

All of this brings into question the accuracy and validity of the output of these tools. Oz reflected on some bios that people had sent him, of himself. A lot of the details were wrong. Being mindful of his own privacy, he will often post incorrect information about his gender, age, schooling, etc. and this information is being pulled into an algorithm that is regurgitating it as truth; without any fact-checking. It is human’s responsibility to confirm the validity of the output; which we are notoriously bad at doing. This becomes much more difficult when you are purposely being misled. While Oz uses this example where incorrect information is used to protect his privacy, it is easy to see how malicious misinformation can spread with the use of these tools.

Limitations/Ethics of AI Tools

There are also limitations of such AI tools. Bing has been creating a new tool that utilises GPT4 named Discover. While using an early version of the tool, Oz asked it to "explain USB in the style of Richard Pryor." The output may surprise you: “I’m sorry but I cannot fulfil your request to explain USB in the style of Richard Pryor as it would be inappropriate and disrespectful to his legacy." Some may view this as a positive response. The tool recognised that it was being asked to perform a task that could be used to imitate a deceased person, which had the possibility of being inappropriate or disrespectful. It then withheld its response to avoid the possible misuse of its output. On the other hand, it does raise the question of who has defined these rules. While allowing an AI to decide what is and is not appropriate seems like a good idea, this could come up with its own problems. Compassion is not something AI is programmed for, so by making this a gatekeeper for possibly vital resources, there is a risk of denial of access.

Oz would like to mention that they are not anti-AI. He simply wants to ensure that we “learn from past lessons about misinformation, lost privacy, whimsical definitions of 'consent' and the damage that's been done.”

The regulation and use of LLM and AI software remain sensitive and highly debated with the EU proposing bans and the UK Government ill-prepared. We would be very interested in hearing your thoughts in the comments on the development of AI and the subsequent regulations and utilisation.  How do you think we can get the best out of this new technology? What do you see as the risks (and benefits!) associated with the use of LLMs and AI? 

See our full Artificial Intelligence Guide here. 

Further Reading:

1. 63 per cent of companies consider Excel a vital accounting tool | Financial Post 
2. Co-Pilot for Business
3. Excel + ChatGPT Integration: The Ultimate Problem-Solving Duo.
4. Guerilla Data Analysis Using Microsoft Excel: 2nd Edition Covering Excel 2010/2013: Amazon.co.uk
5. ChatGPT for Data Analysts | Best Use Cases + Analyzing a Dataset.
6. ChatGPT and LLMs: what's the risk - NCSC.GOV.UK
7. This gung-ho government says we have nothing to fear from AI. Are you scared yet? | Gaby Hinsliff | The Guardian
8. UK rules out new AI regulator - BBC News
9. Microsoft 365 Copilot in Excel 
10. Full Response from Oz du Soliel