Finding sales pitches for various AI products and platforms is easy, but finding a discussion regarding how to relate your business activities to the top level concepts of machine learning and AI, much less so.
In 2019, you’d normally invest into using existing machine learning tools, in order to acquire new key capabilities for your existing lineup, which would have been prohibitively expensive before this boom.
However, if you foresee a difficult competitive landscape over 2020-2025, you might want to go further, and evaluate your access to, or the availability of, valuable data, which can be combined with a more major investment into machine learning mathematics and computer science research, in order to develop major new markets, or solidify your customer relationships and market position.
The basis of ML is showing examples of input (and sometimes output) to your algorithms.
Your examples, your data, is the most significant predictor for your success.
Be too voluminous or complex for laymen to interpret.
Help you actually predict the steps your customer should take, in order to succeed in their mission.
Or you should take, to be perceived to be acting in a way which your customers see as helpful.
Stay available for your exploitation, as it updates and evolves over time.
Not be riddled with BS or lies.
Let’s say that you’ve participated in the business of software, and have a understanding of its realities. So how is this different?
Rather than employing a group of software developers, architects, and project managers, tasking them to learn a problem domain, and to create software products or solutions, you will:
a) show at least tens or hundreds of thousands of examples of the input you might provide, (and sometimes the output you might wish to receive,)
b) to certain machine learning algorithms,
c) running on graphics card installed on server computers.
Someone needs to make sure that one way or another, you get your valuable data, either by harnessing your existing business processes, by developing new data gathering processes, or outright acquiring valuable data sources.
When you have your valuable raw data, your software developers and analysts need to clean it up, combine different sources of data, and create your data pipelines, which produce data which can be used in model training.
Your computer scientists, machine learning engineers and mathematicians will then use existing abstract mathematical tools and concrete model implementations, to try combining various alternative model architectures with the data they have access to, and produce long-running series’ of experiments aiming to solve families of business problems.
Your machine learning engineers will take these experiments, and package them into solutions, which consist of a lineup of model versions and data pipelines.
The solutions, consisting of model versions and data pipelines, are then deployed into the cloud by your DevOps teams and cloud engineers, working closely together with your machine learning engineers.
Finally, your software developers will utilize these solutions, through web APIs, and deliver the business value to your users.
Mathematics is a vast mountain range.
Machine learning is one of its peaks.
2019’s ML tools are flint arrowheads and stone axes.