What are the pitfalls for an AI project in a company?

AI is trendy, it’s a fact, and in front of all the communications made on the subject, many business leaders have launched initiatives on the subject. However, the different studies show that many projects remain at the prototyping stage for the moment, due to different pitfalls and/or constraints that companies need to address before moving on to larger deployments. Here are the 4 pitfalls for your enterprise AI project.

#1 Pitfall: Social

The expected development of AI is often compared to the industrialization by which many manual jobs have been replaced by machines. AI will do the same for customer relations or administrative jobs …

Companies are therefore faced with the complexity of wanting to implement a solution that will allow them to save on the number of employees without having short-term solutions to position these people on other activities.

Although a PWC study predicts “a disappearance of 7 million jobs, but also the appearance of 7.2 million others thanks to AI”, the social brake is real and companies remain wary.

#2: Lack of Data

The awareness of the value and importance of data in the company is recent.

Every company has terras of diverse and varied information, stored on servers or databases. However, the effort to transform this unstructured data is often prohibitive.

Indeed, algorithms are currently not sufficiently advanced to process unstructured data qualitatively without a prior human action.

This obstacle should gradually be lifted in the coming years thanks to the efforts of companies to structure their data management and to the improvement of algorithms.

#Challenge 3: The opacity of the model

The principle of a Deep Learning algorithm is to produce an output result (e.g., prediction) following the analysis of input data. In classical software, it is possible to follow the evolution of the reasoning at each step of the algorithm. This is not the case with Deep Learning. Some experts speak of a black box operation.

One must then be satisfied with the result obtained without having access to the “explicability of the latter”.

This opacity is both a frustration and a real constraint for scientific teams.  It can also prove to be an obstacle in some cases of use such as medicine where the notion of reasoning is essential and where the consequences of a decision can be vital.

Many scientific initiatives are launched to pierce the mystery of this “black box”, but no expert has announced any convincing result for the moment.

#Challenge n° 4: the Wahou effect

This is undoubtedly the first obstacle… Because behind the Wahou effect sometimes lies disappointment. Faced with the various communications that shake the world of AI, the uninitiated tend to imagine uses closer to science fiction than real scientific developments.

The results obtained following the implementation of an AI project are sometimes judged as disappointing by managers, even though they are quite relevant, but meet certain limitations.

This divergence of views must be anticipated, and education must be part of all the communications that are carried out around a project.

In conclusion

AI is present in our daily life and will be more and more. We are only at the beginning; it is a certainty! Many new uses will develop to assist us. Some of them, like the autonomous car, will undoubtedly be major changes in our habits.

No one can say today how fast it will develop in the next few years, or even what its limits will be (if there are any limits…).

However, I think that strong AI, in the sense of artificial intelligence with real self-awareness, will remain in the realm of the imaginary.

Would you like to get more information about AI for your business? Contact us today.

Add comment

By gluca

Recent Posts


Get in touch