Why most Artificial Intelligence projects fail, and how to change this scenario

This question may seem simple, if you consider that the answer is that most projects are based on hypes approaching Artificial Intelligence (AI) solutions, mainly regarding the cognitive capabilities for solving complex or general problems.

However, if we ignore this possibility, and consider that many organizations and teams do know the limits of AI very well, especially in the field of narrow intelligence, then the answer starts to get more complex.

Actually, we have entered a field of infinite complexity, since the threats and opportunities of applying AI in the market are increasingly greater and more necessary, especially if we take into account the synergy with several relevant fields, such as Robotics, Computer Vision, NLP, Big Data, IoT, etc.

Therefore, I will try to present an answer based on my enterprise architecture vision and market experience, which I consider just two cents for reflection.

In this context, the first point that I consider fundamental to understand is that the major advance in the field of machine learning, mainly due to the subfield of deep learning, i.e. AI/ML/DL, is related to the personalized modeling of information representation at scale. I think this advance is so big and significant, especially if we compare it with the AI of the past — mainly based on expert systems — that it generates a false expectation of autonomous cognitive capacities for most organizations. I would even dare to say that the AI we live in is much more AR, i.e. artificial representation, than exact intelligence, as it depends a lot on people’s effort, and often programmatic, to really transform AR into AI.

The second point, and directly related to the first one, is that, in addition to the limitation of specific intelligence, AI, even in the state of the art, still depends on expertise and algorithms created programmatically, similarly, at a certain point, to the specialist systems of the past, in order to really be a competitive differential that impacts in terms of solution, and that really justifies the investment.

In practice, we must tackle problems where AI can solve or help to solve by correctly distributing the tasks of people and machines, and using models and algorithms. Undoubtedly, this is not a simple or easy choice, and it also requires a human knowledge of the application of each capacity.

I think that this complex logic justifies the success of solutions like Google’s Deepmind AlphaZero in the gaming field and — at the same time — its difficulty in applying the same technology in other fields of knowledge.

In other words, any investment or movement in this direction, using Artificial Intelligence to solve or even just help to solve problems, must consider a complete vision of the capabilities addressed by the technologies and the needs of people to complement those capabilities.

Obviously, having this vision, which is much more holistic in terms of systems and ecosystems, and therefore — to my understanding — more complete, is a challenge for any organization, especially when it needs to address highly complex problems, so prevalent today. But a designation of information technology is that we are unlikely to find simple solutions to these types of problems, and correctly addressing AI technologies is a beautiful example of how, when properly architected, machine intelligence can become a fundamental strategic tool in the modern days.

___
By Rogerio Figurelli at 12/30/2020
Senior IT Architect & Solutions Consultant
https://www.linkedin.com/in/figurelli/