AI future and continuous development: what is the impact on Businesses of AI as a continuous MVP?

In a conversation between Ezra Klein and Gary Marcus ( podcast link), the discussion was based on the future of AI around several topics, such as the relevance of an AI responsible development, risks mitigation of ethics, biases, and alignment of AI development with human and society values. It also went through the limitations of Machine Learning, Deep Learning, the design of a hybrid learning model, and a recommendation about Artificial General Intelligence (AGI) the further development of AI, (the point when AI can be able to understand and execute a huge number of tasks similar to humans)

The already increasing proliferation of new business models should be aligned with human and societal values, guaranteeing a responsible and effective deployment of AI including and prioritizing safety, and privacy.

What emerges from the discussion is that the development of AI as technology will not guarantee a definitive Final Product, but a progressive Minimum Viable Product (MVP).

What does all that matter from the perspective of Businesses, and customers for AI adoption? A Minimum Viable Product is known as “an early, basic version of a product that meets the minimum necessary requirements for use but can be adapted and improved in the future, especially after customer feedback.[1]Considering the continuous technical development expected for AI, and the limitations plus concerns for society and people, it is legit to match the future of AI with a special MVP characteristic.

From the Start-ups world, an MVP can be of interest to customers and requires a minimum set of functionalities that will be improved and added. Those features can satisfy minimum requirements.

So, is the concept of an MVP also relevant to artificial intelligence? YES.

I think that the following high-level phases can identify an AI MVP (Minimum Viable Product) cycle. Feature’s introduction/Appsà Users additional tests while working with ità Improvement features/remove issuesà New apps.

Will AI always be an MVP despite how AI creators and developers will create some maturity levels? YES.

The question from the Business is now: what am I going to do with AI?

Businesses need to acknowledge that AI products have unique requirements that must be met. Ability to collect and process high-quality data, incorporating human expertise to improve accuracy, and providing value (money) to customers since day 1.

Requirements?

· Data: Data are fundamental while supplying them to ML (machine learning); training ML is another step. There is a phase when input data is required to start, and the second phase is to structure data flows from diverse sources in real time.

· Hybrid Models? AI products cannot be created using machine learning alone, even with clean, proprietary data. However, a priori knowledge modeling and logical reasoning rules may be used in early products. A hybrid model that requires less data is less complex and provides enhanced transparency but requires exploratory research about the ways of building it.

· Domain knowledge: Never forget that an MVP needs to solve a particular business or client’s problem, so it is essential to have domain knowledge of that problem. And I add this cannot be left only to the AI technical engineers. Does it ring any bell?

· Minimum Viable Performance (MVPF): Many of you, I bet, have already experienced what performance is in using the basics of AI functionalities. Performances are essential to creating and improving AI. The solutions should deliver value to customers and improve upon the status quo.

From an Executive point of view: how can I evaluate the status of the AI’s MVP for my requirements?

At the moment it looks like Gartner has the answer which is based on rating the major vendors’ solutions including features. But, to my knowledge, considering the continuous development of technology a sort of Solution Maturity Level for AI cannot be found. It could facilitate the requirements gathering and a more company-focused evaluation.

In the next article, we will try to answer a question from an executive point of view: How can I be sure that what is in place can allow an AI implementation?

Stay Tuned

[1] Oxford Languages