Co-founder and CEO of Windward (Lon: WNWD), AI entrepreneur.
Getty
Machine learning (ML) and artificial intelligence (AI) have almost become buzzwords at this point, but what do they really mean? On the simplest level, they are basically software that can learn from previous data and then offer predictions on the future.
Vertical software is designed for a specific industry or domain. In the maritime industry, for example, vertical software would fuse shipping and AI expertise to optimize shipping routes and make informed decisions. This is called maritime AI.
How accurate are machine learning and artificial intelligence when it comes to vertical software? The data shows that if the data pool is deep enough and the models are being trained (retrained, actually) often, prediction accuracy significantly increases when compared with the predictions of ML and AI solutions that are not meeting those requirements.
The Quality Of The Input Determines The Outcome
An AI model, whether it uses machine or deep learning, is initially only as good as the data used for training during the early stages. The model will look at a huge amount of data and learn different patterns and behaviors to predict future behavior.
A model will not be able to predict events that never happened or a completely new behavior that cannot be derived from past examples.
In a world of constant change, an active model that is not retrained and enriched with new data is not up to date, and its predictions will suffer as a result.
Just imagine what would happen if you asked a person from the year 1965 how life would look in 2023. They could take some wild guesses based on current knowledge but would be nowhere near as accurate as someone who’s asked the same question in 2022.
Model Retraining
The maritime landscape is constantly changing due to military conflicts and resulting sanctions, port congestion, new environmental regulations, emerging deceptive shipping practices and much more.
Without consistent training and retraining, models will quickly become less effective at accurately predicting behavior and outcomes. For instance, the issue of attempting to conclude bulk vessels’ ETA to a particular destination port requires deep learning sequence models based on different data sources and training.
This trend emerges in studies outside the maritime industry. Regular retraining of an AI model operating in the healthcare industry led to a significant increase in the accuracy of diagnoses. In contrast, a model that was not regularly retrained experienced a decrease in accuracy.
In the field of credit risk assessment, retraining an AI model on a regular basis resulted in a reduction in the number of false positives, leading to a positive financial impact for a company. In contrast, a model that was not regularly retrained resulted in an increase in false positives and a financial cost for the company.
A Systemic Approach To The Challenge
To ensure the accuracy and effectiveness of AI models in the long term, “MLOps” platforms can be used to manage the retraining process. The name MLOps is a fusion of “machine learning” and “operations.” MLOps is the intersection of machine learning, DevOps and data engineering.
These platforms provide a suite of tools and processes to manage the entire lifecycle of an AI model, including training, deployment and monitoring.
By retraining AI models on a regular basis using these platforms, companies can ensure that their models are able to continue making accurate predictions and decisions even in rapidly changing environments. Leading examples of such platforms include Qwak, Amazon SageMaker and many others.
In contrast, outsourcing AI development as a one-off project will likely fail to provide the necessary tools and processes to maintain the accuracy of models over time. This can lead to a degradation of performance, resulting in costly errors and lost opportunities.
When evaluating vertical software, it is crucial to ask whether the company is using an MLOps platform to manage the retraining of their AI models. This will ensure that the models are able to continue making accurate predictions and decisions in the rapidly changing environment of the maritime industry. Or in other words, AI isn’t something you add like ketchup on french fries—it’s a deeply technical domain that requires repetition, platforms and iteration.
Enterprises that choose to procure vertical software usually make a long-term decision, as the cost of replacing embedded software, or an API, can be high. It is best to ensure that the models your users will rely on keep performing so that your business keeps performing.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
