Don’t ask AI ‘how to get started with AI ?’

If you ask Copilot ‘how to get started with AI?’, you’ll get a number of suggestions. ChatGPT has a similar result.

I’d say the exact opposite of CoPilot recommendation #1 – since learning a programming language is a significant task by itself. Understanding AI concepts and how it can be applied are much more important than the mechanics of doing it.

Sure- if you already know Python, are familiar with libraries such as numpy or pandas and you understand data structures like dictionaries, lists and arrays- then jumping into Python might work for you.

ChatGPT suggests study of mathematics, probability, and statistics. Depending on your mathematical inclination and appetite, some of the basics such as linear algebra, matrices and vectors would be familiar to someone with high-school maths, but it quickly gets into unfamiliar territory such as tensors and eigenvectors. I’d say it’s not essential to understand any of this, at least initially.

Likewise with exploring Tensorflow or Pytorch– this is way too complicated for many people.

Some of the other CoPilot/ChatGPT recommendations do make sense, like understanding the basics and applying to a real problem.

Challenges

The challenges in getting started with Artificial Intelligence are:

• hype and misunderstanding, with sentiments ranging from wild optimism to deep suspicion
• a bewildering array of academic & technical information, and few simple explanations of these complex things
• moving from fun & interesting AI –> practical, useful & profitable AI

In the current environment, many organisations are compelled to ‘just do AI’, a little like a musician searching for an audience. Any AI has to be mind-blowingly wonderful and impressive.

Joining the dots between a potential application of AI and tangible value can be hard. AI discussions often gravitate to highly technical issues, and the fundamental drivers of saving time & money or producing better outcomes are overlooked.

Gimmickry vs genuine value

Generating enthusiasm by the use of novel technology is essential and a ‘good news’ story goes a remarkably long way. For example, it is amazing how much interest a (more-or-less useless) gizmo like a robotic dog will produce. However, it is a balance between a generating interest versus perceptions of gimmickry as well as producing some tangible value

Useful litmus tests on longer-term viability of an AI initiative beyond the initial burst of interest include:

• Will it make life of end-users easier, not harder ?
• Will customers pay for and actively use it ?

Start off small & simple

I’d see small and simple ‘baby steps’ as a key aspect of any AI strategy, particularly for an organisation that is just starting this journey.

I’ve found the AI tools in the Microsoft Power suite and Azure AI Studio are good starting points. These contain several AI tools of varying sophistication covering a wide range of purposes. For example, AI Studio contains sub-studios such as Language, Document Intelligence, Vision and Speech.

Most people would be able to make the connection between these tools and potential uses in their business or role.

The ‘AI Builder‘ tools in Power Automate/Power Apps are the simplest and an ideal starting point. In my experience, many organisations are already paying for/have access to this as part of an enterprise agreement, but might not know it…..

There are simple services for document or image processing that would save a lot of time and tedious work for many businesses. For example, the Document Studio contains pre-built tools for processing certain types of documents, or you can build a custom model to suit a particular need. I covered the use of the Azure Language Studio in this post.

These services are quite intuitive and well explained and involve no-code or minimal-code. You don’t need to be a data scientist or a developer to understand the concepts. Similar things could be achieved with open-source Python libraries or other commercial offerings, but often come with a steep learning curve or are point solutions for a particular purpose.

Getting more sophisticated

Within Azure, there are more sophisticated options such as Azure OpenAI or Azure Machine Learning.

It would be an easy progression to start off with some of the basic ‘Power’ tools and move into more complex & customisable Azure ones, such as classification, regression, forecasting, computer vision, & natural language processing.

The model catalog contains a huge number of options from Microsoft, OpenAI, Nvidia, Meta & others for a myriad of purposes across generative and traditional AI.

The Speech tools are a bit of fun, such as creating an alarmingly good fake version of your own voice just by recording a few sentences. I probably wouldn’t use it much in my own work, although it might be useful for my otherwise flub-prone video narration.

Summary

In summary: for any business interested in introducing AI, these Microsoft tools are certainly worth a look and offer an easy and progressive pathway.