The AI is above all a computer program whose purpose is to perform functions usually performed by a human intelligence. What distinguishes a simple computer program from an AI program is the data and the learning that is done with it!
In AI, the algorithm needs to be trained on a data set (Machine Learning, Deep Learning) in order to improve itself and to achieve the goals set.
Today we see that the quantity and quality of the data used is ultimately more important than the algorithm itself. This explains why actors such as Facebook, Google and Amazon are becoming established in the field of artificial intelligence.
The purpose of AI is to replace or improve tasks usually performed by humans. An AI project in a company is above all a business project, it is therefore essential to have within the project team people who perfectly master the tasks to be performed. This will monitor the learning of AI and validate or not the proper functioning before a possible production.
The various stages of an AI project
Regardless of the process to be automated, it is important to clearly define the targeted business problem and to ensure the conformity of the results obtained. This mission does not require Data-Scientist skills, so the integration of a business expert in the project is essential.
The latter must be involved from the study phase and will also have a strategic role in the qualification of the data for the apprenticeship phase.
Data Validation
When setting up a new AI model, it is essential to be able to select clean and sufficiently quantitative data to carry out the first trainings of the algorithm. Random quality data can distort a model and simply cause the project to fail.
Choice of the algorithm
With the democratization of AI, the types of algorithms have multiplied. Depending on the objectives set, it is not necessarily necessary to choose a complex algorithm. On the contrary, the more the algorithm will be evolved, the longer the learning phase will be.
Prototyping
An AI project must be seen as an innovation project, for which it will be necessary to proceed by iteration. Indeed, it is impossible to know when the project is launched whether the results obtained will be positive or not. In order not to mobilize a team and significant budgets without guarantee, it is preferable to carry out a prototyping of the project. This will allow testing the model (algorithm + data) on a restricted perimeter and without external constraints. The iterative approach is then launched, and it is only after the validation of the prototype results that the project as a whole is developed.
Learning
The learning and the choice of the data sets used will allow the models to be optimized to approach the expected results. For example, on a sales prediction project for an e-commerce site, the actual sales data will be compared with the data proposed by the algorithm. As long as these data are not close enough, the model will be trained with the real data from the site statistics.
Technical implementation
AI solutions are very RAM and CPU intensive. The complexity of the algorithm, the amount of data processed, but also external parameters such as the need to work in real time, can have a strong impact on the resource requirement.
This point is easier to deal with if you use a Cloud actor with whom you can adapt your infrastructure. If you choose an installation in your data centers, you will have to anticipate the resource requirements of your solution and size the associated server architecture.
Project follow-up
While initial learning is the most significant, it is important that your AI solution evolves over time with a notion of continuous learning. Indeed, if the latter is frozen without using new data, the results obtained will be quickly outdated.
Like any software solution (business + technical) a team must be in charge of its operation to ensure proper functioning with a validation of the output results and the realization of possible adjustments of the model (algorithm, data sets …) according to contextual events.
What resources should I provide in my team?
Several profiles are relevant in order to have the most complete spectrum possible, in particular:
- Data scientist: He is at the heart of the project and is involved in the selection and curation of the data as well as in the construction of the algorithms used. These profiles are in high demand on the market … but it is complicated to do AI without Data scientist …
- Business expert: He is the functional guarantor of the solution. These missions will therefore be to define the objectives to be achieved, and to verify that the results obtained are valid and conform to expectations.
- Developer: He will work in collaboration with the Data-Scientist to develop the AI algorithms and implement the connectors with the company’s databases and applications.
Would you like to get more information about AI project? Contact us today.
