Making AI work in the Real World

Welcome to the first of my AI posts, where I will be sharing my insights into the evolving world of Artificial Intelligence. As an architect and engineer of enterprise data systems, and more recently AI solutions, I want to make complex AI concepts accessible, and to demystify how AI can be implemented to transform businesses. I will be covering both technical and business considerations with a strong focus how to make AI work in real world business situations. So regardless of your role, business or technology focus or AI knowledge, you should find something to help you navigate the challenges and opportunities presented by AI.

In this post we cover the key types of AI that are of interest to businesses from a commercial or operational perspective, looking at their capabilities and some of the practical applications they are used for.

Understanding the Different Flavours of Artificial Intelligence

There is a lot of buzz about AI and, while much of this has been triggered by significant recent advances in Chatbot technologies such as ChatGPT, this is only a subset of what can be considered as AI. So, what are the others? ?How important are they? And to what extent are they already utilised by businesses?

While there are many different areas of AI capability, the four most relevant types to businesses are probably:

Machine Learning (ML)

This involves training models (a type of computer program) on data to enable them to make predictions, classifications, or decisions without being explicitly programmed with rules to do this. For example, a machine learning model could be trained on past loan repayment data to predict the likelihood that a new applicant will repay a loan, or a model trained on past consumer behaviour that can make recommendations to ‘similar’ consumers.

This type of AI has been around in organisations for a while under the guise of data science or advanced analytics. Where sufficiently good data is available, ML can offer great solutions to well defined decisions for enhancement, automation, or optimisation of business processes.

Natural Language Processing (NLP)

This is a subfield of machine learning and AI that specifically deals with processing, understanding, and generating of human language data by computers. This can cover speech as well as text, enabling capabilities like conversational agents, text summarization, translation, and sentiment analysis. Business applications are wide with use cases such as analysis of medical or legal document to identify key information, summarisation and analysis of contact centre interactions, and personalisation of marketing or customer communications.

NLP is well established in businesses and often embedded within the tools and applications used by them.

Computer Vision

This focuses on enabling computers and systems to derive meaningful information from digital images, videos, and other visual inputs using specialised algorithms and models to mimic human visual abilities. Applications include object detection and recognition, facial recognition, autonomous vehicles, and employee / customer movement detection. Again, there are wide use cases in optimisation based on movement analysis (e.g. handling in a terminal, warehouse etc.).

Other use cases include security, quality management and behavioural analysis, with 3-D and advanced visualisation utilised in areas such as engineering, healthcare, and entertainment. Another key use case in increasingly automated work environments is using visual image processing to detect and mitigate health & safety risks to the human workforce.

Currently while visual processing capabilities are prevalent in a lot of industries, they are most often found as pre-integrated capabilities in specialise tools or applications. However, as the more accessible visual capabilities of generative AI models (see below) evolve, organisation are likely to develop their own solutions.

Generative AI

These are artificial intelligence systems and models that are capable of generating new, original content such as text, images, audio, or other data. A key component of these are Large Language Model (LLMs) that are pre-trained on vast sets of textual (and other) data to learn patterns and relationships, and then to use this knowledge to create new outputs that resemble the training data but are unique and novel.

This is the AI that has being taking the world by storm since OpenAI suddenly made it more readily accessible by the release of their ChatGPT model in late 2022. While these may sound similar to the above types of AI, key differences in GenAI models are:

• They can display creativity and imagination that go beyond simply regurgitating memorised information.
• They can do this in an open-ended way that is not constrained by a predefined set of possibilities (whereas ML models or Rules based systems are).
• They can understand and incorporate context, allowing them to generate content that is relevant, coherent, and tailored to specific situations or prompts.

These differences unlock great potential to create tools that can massively increase human productivity. There are an increasing number of chatbot interfaces and co-pilot assistants that are appearing as embedded components in the standard business applications we regularly use. Similarly, they have the ability transform business processes by creating applications that can act, and interact in a more human like manner, superfast and with reference to significant learned or contextual knowledge.

The health warning is these are still evolving and can behave erratically and generate inaccurate responses (often referred to as ‘hallucinations’). Hence applications often keep a ‘human in the loop’ to mitigate this. Other challenges are the proprietary nature of some of these models, their costs, and potential data loss risks when sharing business information as context to interactions with models.

So why does knowing this matter ?

A key point to note is that, while there is both convergence and overlap in the capabilities of these different AI models, the application of different models to the same business problem can yield results of highly variable quality, performance, and cost. Hence, picking the right type of AI for a business use case is key factor in its successful implementation and realising its benefits.

Don’t worry we will be coming back to GenAI a lot is these posts. Its significance and the impact it will have on businesses cannot be overstated.

In the next post we will step back to compare ML based “Quantitative” AI and Generative AI, both of which are driving a lot of innovation in businesses. We'll look at some real-world examples and business use cases for each type, exploring their benefits, metrics for success, and potential risks and I will share a simple framework for evaluating your own business’s AI use cases.

Till then, why not jot down what you think might be some good use cases in your business, so you can assess your ideas next time?

?

If you read this, I would love to get your feedback, so please feel free to leave any comments, questions, or suggestions.

?