Prominent business visionaries have touted artificial intelligence (AI) as being a bigger revolution than electricity. And we’ve seen AI diffuse into almost every aspect of our businesses and industries. It is, undoubtedly, effecting change. But is it the kind of change that we want? How can we ensure this AI-fueled change is both meaningful and aligned with our values and objectives?
Designing and implementing AI products is an endeavor that requires harmonizing antagonistic dimensions, like trying to collect astonishing amounts of data while respecting privacy. It also applies to leveraging automation to provide better, more engaging experiences, without alienating consumers when something goes wrong, or rendering the workforce obsolete.
What we have learned, through trials and hardship, is that we can’t treat AI as a standalone technology. Instead, it’s an interdisciplinary domain that requires a diversity of viewpoints and objectives.
Let’s consider the first viewpoint, the coming of age of AI. This means focusing on how technology is transferred from academia into industrial and enterprise settings. When we think of academia, it’s related to constantly challenging the state of the art of what AI models can do. When we turn to the enterprise, it becomes about engineering. As we have learned from software development, frequent improvement is better than potentially cutting edge in the future. There are many practices converging into what is coming to be named MLOps, from agile data science, all-around versioning – data, pipelines, models, training, products, everything in the form of code-, automated monitoring, continuous integration and delivery, and so forth. Availability, feedback, and iterative improvement are all hallmarks of a mature MLOps system. MLOps, in turn, acts as a litmus test to gauge how healthy are the development and operations of AI systems in the company.
Another important perspective is data product management and what constitutes “relevant experiences”. They need to be designed taking into account that, in the end, there will be humans enjoying -or suffering- the outcomes of the AI models we put in place. What is the value that we are adding? What decisions are we making easier for users? How are we anticipating their needs and adapting? We have learned that if the selling point for our product is that it has AI, then it’s missing the point. -AI can be an extremely powerful enabler, but not an end in itself. It opens up the doors to a whole new set of possibilities to innovate new digital experiences that we couldn’t think of before. So it’s not only about replacing a custom component with AI, but rather designing around the capabilities that AI enables.
Finally, the issues of fairness, security, bias, and sustainability represent an ongoing concern, so the purpose and consequences of using AI really need to be evaluated objectively. Using AI in a manner that helps vulnerable populations, augments our working capacity and ensures equal opportunity for all, becomes more important as these bigger and more powerful AI models are developed. We also need to be mindful that training these models is a power-intensive activity, and we need to balance the value in training one more humongous model versus reusing and adapting the existing ones at the product definition stage.
AI is not just a very powerful and exciting technological toy. It can indeed be the biggest revolution of our lifetime. And given that the enterprise is the place where it will come alive and change the world, we must make sure we are playing this balancing act very well, considering an interdisciplinary perspective with modern best practices in a humane way.
This article was originally published on AI Business. Globant will be at the AI Summit in London, from 22-23 September to discuss all the latest developments in the world of AI. We look forward to seeing you there!