How to AI

Artificial Intelligence is not a matter of "if" nor "when", but "why" and "how".

While debate is still needed to fine-tune (pun intended) the "if" around governance and ethics - to determine which use-cases require more stringent boundaries - we need to focus on the reasons for using AI and determine the best mechanism to achieve optimum results.

?“AI is the new electricity...I actually have a hard time thinking of an industry that I don’t think AI will transform in the next several years.” — Andrew Ng, co-founder of Coursera and Google Brain

AI at our fingertips

It's highly likely you've encountered AI already, being built into tools and products you use regularly. And if not, it won't be long until you do.

Companies are investing heavily in feature enhancements to increase the capabilities of their offerings. Such as:

• Summarising large amounts of data and text
• Combining a variety of information sources to provide a consolidated view of an entity (customer, product etc)
• Generating content - images, videos, email responses, blog posts....
• Analytics without needing to know the particulars of the system (scripting, languages, queries)
• Automating processes where we can use natural language
• ...and many others

AI for all

Every day there are new products and services created based on AI.

These range from very simple to fairly complex and capable. Usually there is a cost associated, as it will be expensive for these companies to create and/or run the service.

The list grows every day, although it's becoming a crowded marketplace where only the best will survive in the long run.

Build your own AI

There are many ways to create your own designer AI.

The simplest is to take an existing creation (be it a model or service) and customise it.

This customisation could be in the form of an instruction set (aka Prompt Engineering), so you are not creating or modifying the AI process but are guiding it to a specific outcome style. This AI is likely to come from the categories explained above.

Another way is to associate information with the AI model in a way that allows the AI to efficiently generate meaningful results. An example is using a repository of information (procedures, FAQs, website) to feed into the AI model. One key advantage is that this association (using embeddings) allows the intelligent part of AI to better respond even in the absence of key words that a typical "search" would require.

“The playing field is poised to become a lot more competitive, and businesses that don’t deploy AI and data to help them innovate in everything they do will be at a disadvantage.” — Paul Daugherty, chief technology and innovation officer at Accenture

The two most involved and expensive methods are:

• Fine-Tuning - taking an existing model, thus avoiding the huge expense of creating such a marvel - and further training the model specific to your need. An example is object recognition in images where you may have a specific application that a generic model doesn't perform well in. By training with your own data, the model performance improves.
• Custom Model - I suspect very few will venture down this path, but you can create an AI model yourself. It's almost reinventing the wheel however there would be use-cases for this scenario. It requires a lot of good (and bad) data, time and expensive hardware (buy or borrow).

AI Platforms can help

Unless you have a large, well cooled data centre with capable hardware then you'll likely need to make use of some external resources.

There are a range of platform offerings from bare bone compute up to very automated and orchestrated services.

In conclusion

The key to success is knowing the options, understanding your "why" and delivering the "how" as effectively as possible.

New knowledge is required, which can be gained thru informal and formal training with either requiring ongoing "hands-on" time as well.

At the very least, these topics should be on your agenda for strategic discussions.

“AI is not going to replace managers, but managers who use AI will replace the managers who do not.” — Rob Thomas, senior vice president of cloud and data platform at IBM (2020)

#ai #artificialintelligence #governance #itstrategy #DigitalTransformation #FutureofWork #TechTrends