The challenges of applying Artificial Intelligence

The growing gap in Artificial Intelligence adoption

The unstoppable advance of artificial intelligence (AI) is creating an increasingly wider gap between companies that have adopted it and those that are just starting. Even those who have started with a proof of concept that did not have the immediate great results expected may experience it as a small frustration and take a "wait and see" strategy.

The concept of the J Curve by Professor Erik Brynjolfsson (Stanford) perfectly defines this initial phase:

"These investments and changes often take several years and, during this period, do not produce tangible results. During this phase, companies are creating 'intangible assets.' For example, they could be training their workforce to use these new technologies. They could be redesigning their factories or equipping them with new sensor technologies to take advantage of machine learning models. They may need to renew their data infrastructure and create data lakes where they can train and run machine learning models. These efforts can cost millions of dollars (or billions in the case of large corporations) and not generate changes in the company's production in the short term."

All of this makes the gap increasingly wider between the convinced and the waiters. As indicated by a recent research by the McKinsey Global Institute, there is a real and growing gap between leaders and laggards in the application of AI both across sectors and within them.

Lack of data and specialized personnel

Precisely, among the most common challenges that companies face when starting AI projects is the lack of data or the lack of specialized personnel, and in some cases, both, requiring large initial investments.

Most current AI models are trained through "supervised learning". This means that humans must label and categorize the data, which can be a considerable task. Unsupervised approaches reduce the need for large labeled datasets, but the reality is that in many use cases we cannot apply them. The use of supervised or unsupervised models is intrinsically linked to the use case (see Machine learning explained for more information).

Furthermore, the most advanced machine learning techniques such as deep learning require training datasets that are not only labeled, but also large and comprehensive. Massive datasets can be difficult to obtain or create. Both obtaining massive data and preparing and labeling it can represent a significant investment.

Without forgetting the challenge of talent. In this case, we can address short-term deficiencies through outsourcing. However, externally delegating all of AI can be a colossal mistake for companies.

Business leaders who hope to narrow the gap must be able to address AI in an informed manner. That is, they must be able to understand for themselves where AI can lead to revenue growth or capture efficiencies. They must also know how to distinguish where AI does not provide value.

Furthermore, we have already mentioned that they (and not the technical profiles) are responsible for understanding and solving the challenge of the "last mile" of incorporating AI into products and processes.

They are challenges that outline a roadmap of several years for companies. That path we will hardly be able to avoid. But we can start using and experimenting quickly with machine learning tools, datasets, and trained models for standard applications, which are widely available.

This is ready-to-use artificial intelligence or AI off-the-shelf, which includes, for example, models for natural language processing and computer vision. Sometimes they come in open source code and in other cases through application programming interfaces (APIs) created by pioneering companies like OpenAI or major public cloud providers like AWS, Microsoft, or Google.

Artificial Intelligence ready to use from AWS

Next, we show the main use cases offered by AWS's off-the-shelf AI technology.

There are already companies that are implementing off-the-shelf AI solutions, either using natural language models, such as the student support chatbots we developed for the Generalitat de Cataluña called PauBot; or computer vision solutions, such as the identity validation system for online assessments that we developed for a major online university, which has helped reduce fraud in non-presential exams.

In practice, we need to be able to combine both approaches and have the necessary capabilities to design the ideal solution in each case. For standardized use cases, we have off-the-shelf AI models that we can implement with cloud architecture and data architecture capabilities, working closely with domain experts.

For non-standardized cases, we will also need data science capabilities to assist us in the decision-making and creation of the artificial intelligence model.

The promise of AI is immense and the technologies that are supposed to make it a reality are still in development.

If you know that now is the time to think about the long-term survival of the business and position yourself in the new era of data-driven companies, contact us and we will try to help you. Without forgetting the challenges of implementing more customized and ambitious artificial intelligence with data science, we can rely on off-the-shelf AI to solve various use cases in a truly fast and effective way.

More information

If you want to read the previous articles in this series, remember that we have explained the data challenge, listed the 4 types of data to apply AI, and explained the importance of the 3 knowledge roles among team members.