By Sunny Slater
Our interest in AI
In the coming weeks, the Tim Ashwin Consulting team will be using two Generative AI applications to carry out an internal economic analysis project. We will be testing AI’s potential to improve both the speed / efficiency of analysis performed at each stage and the quality of its outputs.
To kick the process off, we have drafted this initial article (with a bit of AI assistance!) It explores some potential benefits of AI in preparing inputs typically required to support the business case for an economic development project.
Why is AI important?
There is a growing recognition of AI’s potential. Many see it as the path to further enlightenment, with unbounded potential to make the world more productive and better, helping to address a plethora of business problems.
BT Group name it “one of the most important innovations of the decade“, Microsoft has witnessed a 55% reduction in programming time, and research by the University of Pennsylvania has suggested a 50% time saving for accountants, auditors, and tax preparers on most tasks.
However, concerns about AI’s risks and its potential for harm are equally prominent with worries about data protection, misinformation / manipulation of the public, and job displacement being expressed.
Regardless of which side of the debate you most agree with, there is no doubt that it’s time we discuss and explore the potential impacts of this powerful tool.
How does Generative AI work?
Generative AI involves computer programmes housing virtual “neurons” organized into “neural networks,” designed to mimic the cognitive processes of the human brain. These programs are then fed vast datasets, spanning diverse mediums, which are used to train the programme.
As it continually processes data its knowledge expands, enabling it to identify increasingly intricate patterns and enhance the quality of its responses.
Using AI to support an economic development business case
Generative AI can help develop a range of key inputs required to support a project business case, as outlined in the schematic below.
Drawbacks and limitations
Although the potential for Generative AI is huge, it is important to consider its potential downsides, such as the following:
- Generative AI can correlate and analyse, but it cannot judge. It has no subjective view of “right or wrong” nor any motivation. Therefore, outputs can appear illogical, incoherent or distorted.
- It can often create superficial or dull content with little underlying depth or substance. Content will often lack a sense of originality, and any meaning or poignancy may be diluted.
- Content generated may often not be factually accurate. Even if what is assembled follows logically defined and mechanically robust processes, the end-product will not necessarily be reliable. Computers cannot tell if something doesn’t feel right or seem right, since they have no feelings or thoughts of their own.
Initial recommendations to make best use of AI
Although GenerativeAI programmes will attempt to respond to any prompt they’re presented with, to get the best out of AI we would recommend the following:
- Context is Key: Always provide the AI with context and persona (i.e. the role / characteristics of the content recipient) to guide its responses.
- Define the Task: Clearly state what you want the AI to do.
- Use Examples: Giving examples can help in clarifying your requirements.
- Set Restrictions: Mention any constraints or restrictions you want the AI to adhere to.
- Define Quality: Specify what you consider as a good response or outcome.
- Break it Down: For better flow, break prompts into smaller chunks, making it more conversational.
- Persona and Tone: Assigning a persona or job title can help in setting the tone and objective of the AI’s responses.
- Identify Your Style: By comparing your writing with others, ChatGPT can identify and mimic your unique style.
- Set Expectations: Clearly define expected behaviour, especially when giving tasks with constraints.
In addition to optimising your prompt design, plugins can be an astonishingly powerful addition to Generative AI programmes. One of the AI tools we are using is ChatGPT (Plus). These plugins can be added to the programme, which we consider may be useful:
- Wolfram: Ideal for solving mathematical problems.
- AIPDF: Can be used to interact with PDF files, with the ability to identify excerpts that are related to specific areas and summarise sections of or entire documents.
- Keymate.AI: Offers a range of features including a personal memory bank and Google searching capabilities.
- Scholar AI: Best for researching academic articles.
- Prompt Perfect: Automatically optimizes prompts for better results.
- Numerous.AI: Enables the functionality of prompting GPT 3.5 inside cells within an Excel spreadsheet.
Chat GPT has been used to assist in preparing this article. Can you tell? Do you have any thoughts, observations or experiences of your own to share? Please get in touch if you’d like to discuss this further.
Generative AI – Artificial Intelligence programmes that can generate new content (this can be in different forms depending on the application).
Training (in the context of AI) – The process of the AI programme learning to detect patterns or relationships from the dataset it’s been fed.
Iteration/Iterative Design – A design process that involves making incremental improvements and refinements to a project or product over time, often based on feedback and user input.