It is hard to know when you don’t have the technical background to assess if the problem you want to solve or your idea is a complex one to solve or not. Focusing on the idea alone doesn’t give you enough information to know the different variables and resources you need to succeed.
So thinking about how complex or straightforward your solution to the problem is is very important. It could significantly affect your motivation and what resources you need to get started. It could make you decide that this isn’t for you. It could be a point of reflection to stop and start planning how to bring the appropriate resources you will need to bring your ideas to fruition.
You may be thinking what you don’t know will make you stronger. It is not a good strategy, even if you think you are better off not knowing to prevent you from quitting too soon. When you are not learning, you are not allowing yourself to see the whole chessboard to make your move in the best direction.
You’ll need to review five things to understand if you have a complex or easy problem at hand.
1. Your Available Dataset
You will need data to build your AI solution or model. Data needs to reflect the problem you are trying to solve.
Knowing how much data you need depends on the training method. Are you applying deep learning or machine learning? Each will require a different number of datasets. But at the average minimum, expect you to have about 1000 datasets for each prediction.
Sometimes you may be unable to label each of your data because it is too much work, time, or cost. In that case, you can apply machine/deep learning techniques like unsupervised learning to help you skip data labelling. Unsupervised learning requires more data to get good results. If you have supervised learning, you won’t need as much data.
Where you get that data is another matter. If you don’t have all the data from the start, the problem could be complex when you need consent to share and crowdsourcing data and permission to use and copy the data for training your algorithm.
2. Your Use Case
You’ll need to know when, for whom, and what the problem the solution is for. This is your use case. Solutions with more than one or dozens of use cases create a more complex solution than a simple one.
For example, an example of a simple use case is someone who receives a recommendation for which foodbank to go to based on their Hispanic culture and lifestyle in a specific neighbourhood in an urban city.
A complex use case would be an app providing a recommendation for all cultures and the whole urban community. The number of unique cases we need to satisfy increases and the number of data is required.
3. Parameters or Data Point Involved.
One use case can have multiple parameters involved. Say, I wanted to detect or predict depression in youth. Several factors can cause depression, and each parameter must be accounted for in the data we collect.
You’ll need approximately multiples of ten for each unique data point for your problem to work. That’s why chatbots require extensive data instead of a recommendation engine. A chatbox will have to detect each letter in the alphabet, each word in the English language, and the meaning of each sentence from comprehension to sentiment. This makes it a complex problem. A recommendation engine might have five to ten data points to consider to make a recommendation.
Anything that requires you to mimic the human brain will be complex, especially if you are building it from scratch.
4. The level of risk to your solution
If your idea can put someone at risk or hurt them in ways that are neither reversible nor easily undone, breaks the law of some kind, then you should consider your solution complex. You will need to do more work to ensure safety is applied. This requires more effort than, say, a recreation application.
Solutions that involve people will need to think carefully about ethics. You could be limiting people based on the fact that you haven’t factored their unique needs into the dataset or that the AI model excluded certain groups of people because they were perceived as outliers. Insufficient data existed to make them worthwhile for the model to train.
5. Off-the-shelf solutions
Some models don’t require you to build from scratch. There are existing tools, either available for free or at a cost. This can help cut down the data and tasks you need to get your complex idea into a simple one.
Off-the-shelf tools can be an API that already works for other domains and can also work for your case. Many APIs, like that of ChatGPT, are being wrapped around with a new look and feel and improved usability but do the same tasks as ChatGPT alone.
You can get an API for speech-to-text, image recognition and audio recognition, to name a few, that you can use inside your own application. This is a simple idea that will take no time to get started and launch. You won’t need to build an AI model from scratch.
Another way to use off-the-shelf is to use a model that already exists. It could be licensed, or it could be your own model. Either way, you’re using less data to train on top of an existing model. You are finetuning it to improve the results for your unique situation. You don’t have to start from the beginning with millions of data. Sometimes, you can apply domain-specific knowledge to customize to your needs.
AI is relatively new. It can get resource intensive, and you want a game plan with the right resources to start and grow.
Even if you find out that AI isn’t a path for you right now, it is not to say that later there will be resources accessible that can make it easier for you to achieve what you want.
At the very least, you want to be in a situation where you have the handle on all the possible scenarios. You want to know beforehand to make smart decisions.