If you have tried out some of the generative AI tools such as ChatGPT, Gemini, Claude, etc. you have probably been quite amazed at how it feels just like speaking to a person. So much so that it can even convince you that it can do everything and anything for you, writes Dorothy Molloy, co-director of Climbing Turn.
How does AI actually work?
In a nutshell?
Well, even a nutshell of a summary would be more than you have time for right now but here’s a super quick explanation of generative AI.
AI works by taking parts of information and splitting them into tokens. If it is a word then it will be part of a word, for example “baked” could be split into “bak” and “ed” so that you could also use other parts of a similar word e.g. “bak” “ing” => “baking”. Then all of the tokens are mapped to other tokens depending on how likely they are seen together creating multiple dimensional connections.
The multidimensional connections are very complicated, more complicated than a human can hold in their head.
OK, I won’t go further into this as it gets pretty complicated but what I did want to get across is how these connections make the AI able to predict the next likely token. It predicts the next likely token and this is how it creates the text, code, images, etc.
But predicting the next token isn’t intelligence; it’s just quite remarkable!
Push rubbish in get rubbish out!
If your systems has poor data or it is in a mess, then building on this is going to end up with even more mess.
The same will be true when you implement AI, in fact more so as the AI builds upon the data that you feed it. If you build code on mess, much like building a house extension on a ramshackle old building, you are going to end up with a poor result.
You will need to train the AI on your own data so before you begin you need to clean your data first.
What do I mean by “clean your data”?
I mean you need to go through your systems looking for all the incorrectly entered records. Here are some common examples:
• A person’s full name entered into a field that is specifically for their first name.
• A product SKU in the ID field
• A link between records that relies on a text field matching exactly
• Fields that allow free text when really the user should have been given a limited choice of options (preventing spelling mistakes, etc.)
If you think that nothing like that exists in your data then you probably just haven’t looked hard enough! I’ve never come across a system without these little beauties!
How to start
The first place of call is to look for anything that is broken or to use a technical term “shonky”.
This sounds obvious but many businesses put up with less than optimal systems for so long that they no longer notice it. Before you spend a penny, you can prepare the way by doing a bit of research yourself.
Here are some signs of things that need attention:
1. Processes that fall over when someone is not available to tend to them.
2. Repetitive manual processes
3. Entering information that originated elsewhere manually or semi-manually
4. Processes that often fail or go wrong
Sometimes these things are not so easy to see at first glance because your staff may be alleviating the symptoms and smoothing out the rough edges.
Interview your staff to find out if they are performing dull repetitive tasks or if they are regularly fixing the same issue.
Of course, hiring an external technical consultant is great for spotting the things that you can’t and the things that you have just got used to dealing with but you can get the ball rolling on your own.
AI can be my technical resource
No. AI can be your technical tool.
So long as you really know your business and you have someone that can verify that the AI is creating what you need, you can leverage AI.
Is AI the best way to increase the productivity of my business?
AI is an excellent solution for some things but you need to understand what you are trying to achieve before you can honestly say that AI is the solution.
Where AI really shines is where you have millions of records and you need to have a way to make predictions based upon this data. For example, you have a huge data source that holds information about the product. You can use AI to predict how your sales might change if you introduced a new line into your stock. The AI can be used to map this product against existing data and predict how well it could sell.
AI is also great for brainstorming and summarisation. For example, you have a long document to digest; by asking AI to summarise, you can be more sure that you haven’t missed something out.
It helps with finding things. Use it to find out when something that is on your server and you don’t know the document name or what folder it could be in. You can describe what you are looking for and tell it how the AI should respond: e.g. a summary of the path to the document itself.
The next step in AI automation is the use of AI Agents, they can make decisions on your behalf. There are a lot of pros and cons to this and this is a whole topic of conversation in itself.
Common Sense
Beware of using a sledgehammer to crack a nut. Often the solution is with tried and tested software development and that may be the most cost effective and accurate solution.
AI is great, but accuracy is not its strong point. You’ve probably noticed how easily it gets things wrong or makes up plausible stories! If you need accuracy, then a traditional software solution or an AI mix is going to be the most appropriate for you.
Cost
This is really the moment to get the most out of AI. OpenAI is fighting with Claude for supremacy, and that could mean some great price cuts for businesses using AI.
But will these low prices remain low?
As more and more businesses start to take on AI, resources are getting stretched and the large AI firms are already considering changing the pricing accordingly.
Privacy and business integrity
It is important to read carefully the Privacy and Terms and Conditions when deciding to implement AI. After all, your data is going to be extracted onto a server outside of your business (unless you have the budget to host your own large language model).
You might want to consider who your data will belong to once the AI has been trained on it.
• Will this be the property of the tech company that supplied the AI?
• Will your hard won data be shared with potential competitors?
Innovation
Well, that comes from you. AI copies, it doesn’t create.
Comments are closed.