Maslow's Hierarchy of AI Needs

Intelligent bots have needs too! 🤖

The psychologist Abraham Maslow has proposed almost a century ago a model of rising human needs. We must all address immediate physiological needs like air and water, with physical safety coming right after, before we can consider emotional needs of love an belonging. Only as we work through the rising layers can we reach the pinnacle of self-actualisation.

Turns out, AI is about the same 😜🤖

At least, the working with and building of AI-powered applications. This post will consider how to implement these rising layers in a sensible way, and how to map where you are and what you need — whether you are working on an “AI startup” with a public product or a BigCo trying to play with AI to see what it can achieve.

Pyramid steps

Let’s explore the steps that build up to AI-Maturity for an organisation.

Tools and Tech

At the base of the pyramid are all the tools, techniques, and general tech of AI. When just starting at this space, people and organisations are exploring and learning about all the options: from not calling Midjourney an LLM, or why RAG is a counter to hallucinations, to the difference between generative AI and vector search (and why both have been around a while).

People also start testing the various models, from using the current “brand names” to produce some bad poetry and shonky images, dream about reducing costs across everything, and end up in court making silly arguments trying to claim that their chatbot is a legal entity.

As you settle on your chosen tech stack, bear in mind that “brand name” isn’t the only dimension to choose a provider. Consider what your 3rd party provider (your supply chain) says about your ethics, privacy, security, and other dimensions that will reflect on your public face.

Specific Knowledge

You can expect to start seeing real value when you mix in your own organisational knowledge, over that of a generic model. It’s like having an enterprise search over your organisation’s systems when trying to find specific documents, rather than using an external search engine like Google. You want to add your secret sauce (data, knowledge, practices) into the model to ensure that your AI-powered application, to deliver your own unique value to the customers you service.

There are techniques like RAG (Retrieval-Augmented Generation) which try to limit the model to the knowledge articles you provide it with. That’s how you build an answer engine, and reduce the model from hallucinating and pulling in info it found Reddit. Other ways include data you’ve collected about your customers and users to build preferences and recommendation, using it to process large volumes of documents, and generally injecting some AI boost in either understanding or manipulating data to aid your humans.

That example above about Air Canada shows how it can go wrong, so do invest time in some good old fashioned solid software development practices, like testing to your level of organisational risk appetite.

Personalisation

Speaking to customers, so many are taking about wanting / having a mandate to start delivering hyper-personalised experience. It makes an alluring sense, trying to individualise what you deliver to each of your customers, tailored specifically to their needs for a spectacular experience to keep them happy and loyal.

At which point I need to stop and ask them a few questions, like what do you hope to achieve? or do you have enough data to support this? Since I work on enterprise software and customers are usually government agencies, utilities, or other service organisations (like legal or financial firms), it often comes up as “not sure” and “err… no”, respectively.

Building personalised experiences is a journey in and of itself. It’s a topic that larger than this blog post, but consider the stages, needs, and available information as you move from basic segmentation, through targeted content and experiences, to true individualisation. Not everyone needs — or has the data to support — something like the Spotify recommendation engine. Sometimes differentiating between business owners versus private individual enough, and differentiating dairy from wheat farmers is about as fancy as you need.

Image from Ninetailed’s Personalisation Maturity Model article

Waves

The above image shows the layers like a cake — and anyone who’s been on the internet long enough knows the cake is a lie. Even when considering Maslow’s human needs, addressing them isn’t linear progression up but rather waves of rising and decreasing intensity.

This is better explained with another image:

As you begin the journey, you may have personalisation on your mind, but you’ll spend most of the time playing with and learning about the various models and technologies.

As you begin to settle on your tech-stack you’ll want to introduce your specific organisational knowledge. You’ll find that the rule of of GIGO still holds, and good content is king with AI as with everything else.

Once you have those streamlined and under control, you’ll also mature in tailoring the experiences in your AI powered app to segments and individuals, and spend more time in that area.

Sense-making

Whether it’s your executive that had a drive-by “Add AI!” directive or a new idea that new GenAI genuinely unlocks, you’ll go through the above stages. First, you should put on your traditional product manager hat and think about why and for whom you’re doing this.

Then, when it comes to the actual innovation and design of product and services, you’ll likely find yourself thinking about the above dimensions. You can probably recognise where your organisation is, so hopefully this article helps you plan how much effort you need to invest in thinking and doing as you build up your capabilities. Expect a lot of reading and experimentation by yourself, by the dev team, and with your users — this is an emergent field, after all, so it makes sense to probe-sense-respond. You’ll soon know what data and resources you have, what are the right problems to work on next, and where are the gaps you need to fill, before you can move to the next problem.

And, as a personal plea, please consider the Ethics of AI. The field is currently less regulated than the Wild West, and while the rich are getting richer it’s up to us to build products the right products right way.