Now that you are more familiar with artificial intelligence’s key concepts, let’s find out what goes on behind the scenes of an actual artificial intelligence project.
To help you get a better sense of how to develop an AI project, let’s imagine that you are a project manager at a major automobile manufacturer.
Your goal is to lower energy use. A research group has already conducted several in-depth analyses and has discovered that one way to tackle the problem is to use AI to anticipate future energy consumption.
Here are the main stages of an AI project you might organize:
Put Together the AI Project Team
To make this project a success, you will need a team with many different skill sets:
Industrial skills: People who specialize in your specific business sector (in this case, manufacturing). Their knowledge will be crucial to finding the right solution.
IT skills: In particular, a software architect and some software developers.
AI skills: An AI expert (senior profile), plus data scientists (career data and AI professionals).
Data governance skills: A DPO (data protection officer) - someone from the legal department (to ensure data management compliance), and a CSR officer (corporate social responsibility) to ensure that environmental objectives are met.
Define the Parameters of Your AI Project
Now that your team is in place, you need to ensure that you properly manage the project. For an AI project, this means you must complete the analysis:
Define the relevant business parameters: What problem are you addressing? Who is it for? What's the budget? What is the cost? What is your return on investment (ROI)?
Define your objectives: For example, what are the solution's CSR (corporate social responsibility) goals? What is the solution's economic model?
Evaluate impacts: Ethical, societal, personal safety.
Provide data governance: Define the data, its availability, usability, consistency, and user controls.
Design the solution: Define the architecture and the level of data security.
Industrialize the solution: This includes results tracking, user management, communication, and change management.
Collect Data
Your project will only be as good as the data you use! In this phase, it’s important to collect as much data as possible.
Remember, your goal here is to improve the energy efficiency of a manufacturing plant. So you could collect data concerning electricity consumption over specific periods of weeks and months. You may also want to obtain data from the production equipment itself (how much energy each machine uses, how long it runs per day, etc.).
Your plant is also equipped with many sensors that measure and record various types of information, such as the number of times someone enters the factory, the temperature of different rooms, elevator use, etc. It can also be a source of data!
You might even consider obtaining more general data, such as the local daily weather report. Data relating to hours of sunshine, precipitation, wind, temperature, etc., will likely be useful in determining energy use. Your team will have the opportunity to verify this later on.
Clean Your Data
You and your team have collected a lot of data. Before you can use it, however, you must ensure that it is reliable.
You can check reliability in a couple of different ways:
Check to see whether any data is missing. Most likely, your data is not exhaustive.
For example, following a computer malfunction, some sensors may have failed to record energy consumption.
Make sure there are no outliers.
For example, looking at the temperature data, you notice a day on which implausibly high temperatures were recorded that can’t occur. It could be just a random computer error!
In both these cases, the data concerned is unsatisfactory. However, your team’s data scientist can replace the missing or erroneous data using statistical tools known as statistical imputation.
You have made it to the end of this stage. You now have high-quality data that’s ready to use. Let’s move on to the next step!
Explore Your Data
You are now ready to examine the unique characteristics of your data. This examination is called data exploration or data mining.
Your team’s data scientist will work with experts in your industry and data sources to develop a clearer picture in this stage.
Your goal is to understand the plant’s energy use. The data will help answer questions such as:
How much energy is consumed on average, every day, week, and month?
When does the plant consume the most energy?
Exploring the data will enable you to validate hypotheses or intuitions. For example, the team may believe that when more employees work on a Monday, energy consumption is higher all week. This hypothesis can be validated or invalidated by cross-tabulating the data.
Model Your Data
And now you are finally going to be able to use some AI tools!
Your goal is to predict the plant’s future electricity consumption as accurately as possible.
To do this, you will apply machine learning, which you will learn more about in the next chapter. In concrete terms, you will model electricity consumption based on all of the variables at your disposal.
Let’s use an example to illustrate what this means. You hypothesize that temperature has an impact on energy use. When it’s cold outside, people tend to use more energy to heat buildings. Conversely, when it’s warm out, electricity consumption goes down.
You can reduce all of this to mathematical formulas enabling your model to anticipate future energy use based on local weather forecasts!
Data modeling involves two phases:
Phase one is learning. This phase will train your model on examples, giving the system meteorological data and data concerning electricity use during previous periods.
Phase two is prediction. Your system is ready, and you can use it to predict future energy consumption.
Evaluate and Interpret Your Data
You have just developed an initial model that allows you to predict future energy consumption.
But how do I know this model is reliable?
You have to evaluate the model; that is, confirm that it is useful and provides reliable predictions.
To do this, you will test it by predicting electricity consumption for a period on which your AI system has not been trained, such as the previous month.
Deploy Your Model
Your artificial intelligence system is ready for use in the real world! It will enable the plant to control and use energy as efficiently as possible.
To ensure your system meets this goal, your project team will evaluate its performance after a pilot period. Is the AI system relevant? In particular, does it add business value?
Let's Recap!
Artificial intelligence projects involve many different team members, including AI specialists and non-specialists (administrative, production, support). Each of them contributes their expertise to develop a positive, effective AI solution.
AI projects have different stages: data analysis, collection, cleaning, exploration, and modeling, model evaluation and interpretation, system production and maintenance.
You now know what goes into an artificial intelligence project. Let’s look at how the model learns by taking a deep dive into AI's sub-discipline: machine learning!
