AI

Are you ready for this?

You don’t know what you don’t know

Many leaders believe they need AI. Their evidence is based less on what they know about AI and more about:

• What they see in the media, most often variations on a Skynet theme.

  
  

• What they learn in business school, where it is presented as a rushed, bolt-on module to each of their current, crumbling executive programmes. The implication being that AI can be compartmentalised and so once you have acquired a Chief AI Officer all will be good.

  
  

• What they see other leaders do – They are using it, so we simply need to buy a ‘bigger version’. Here, AI is more about an arms race than deployment.

Uninformed leaders are the equivalent of winged birds from an AI vendor perspective. The vendors are in no rush to edit this braided narrative. Such incompetence will have business and societal consequences. Governments, don’t look away.

What it is?

AI can be considered a collection of tools designed to mimic cognition. Unlike traditional software, where the logic is embedded in the tool, AI systems can learn and respond to situations that were not programmed in at the time of their creation.

Over the decades, AI has evolved from rules-based logic to autonomous processing.

• Rules-based: User - Here are my symptoms, what’s wrong with me?

• Autonomous: AI - Tell me your goal and let me decide how it is achieved.

Autonomous AI is better known as agentic AI.

What it is not

Whilst AI stands for artificial intelligence, it is not artificially replicating intelligence. Until neuroscientists really understand how the human mind, attention and intention work, it will be impossible for computer scientists to mimic intelligence.

What we have today is a maths (statistics) parlour trick, aka machine learning. It is very impressive, not least because it plays to my insecurities. The underlying maths is strikingly good in that plausible and often correct responses are presented so eloquently. And all based on guessing what the next word or partword will be.

So what we have really is ‘artificial artificial intelligence (AAI)’. I can’t see many tech marketers rewriting their copy around this new clunky phrase.

It is not your friend, but it can mimic one. It can detect key words that give it a clue as to how you want it to respond.

It doesn’t really understand context, as shown by the riddle - "I want to wash my car. The car wash is 50 meters away. Should I walk or drive?". For a period of time, generative AI tools would apply an energy efficiency argument and recommend walking.

But it is improving in respect of functional context, the car needs washing, so the car needs to be at the car wash. But not the human context, “the last time I used the car wash, it broke my antenna. I’ll walk over and establish if this is likely to happen again”.

Context is everything organisationally. AI is not there, yet. 

Considerations

Data

Unless your organisation has a pristine data lake, your AI will behave like a user-accessible sewer. Very few organisations of any size have their data act together.

Response quality

It we were to think of AI less as a lifeless statistical word guesser and pretend that it is a living system that can actual learn, hallucinate or decide, then we are looking at a species that has little regard for what it eats. Increasingly this will include content it has generated.

Even rabbits, which eat their own excrement, only do it once. So over time the quality of the AI output is going to degrade. We may well be currently enjoying peak generative AI.

A risky bet

Almost half the global equity value is in organisations that need AI to be successful. There is no question AI will be part of our lives whatever happens in respect of investor confidence. But an AI correction is inevitable and so there is a risk in overinvesting this side of the correction.

Tokens

Tokens are the currency by which you acquire AI services. Your prompt will cost several tokens. The underlying processing could consume many processes. Depending on what you asked for, the number of tokens used could be very high.

You might well have a cap on the cost of your AI licences. If not, then you will eventually discover that the cost of using AI does not stack up against the organisational benefits. It probably doesn’t even stack up for those selling the tokens, but this can perhaps be attributed to the cost of marketing.

Reputation

Examples of AI systems getting it wrong are plentiful. Examples include:

• Unfair student exam gradings.

• Racially biased prison sentencing.

• Families wrongly considered as fraud risks.

• Individuals wrongly accused of benefits fraud.

Be careful on unleashing AI on business-critical processes.

Organisational design

Organisations that were built with hierarchy, process-centricity and efficiency at their core, ie industrial organisations, may well struggle to embrace AI. Imagine giving an elderly raver speed. You might well see an initial increase in performance. But what will soon follow is a permanent loss of structural integrity. AI isn’t just another initiative/ bolt on, it requires a structure that can handle its power.

Governance

Because AI behaves like a black box, organisations do not know what caused it to deliver the outputs that led to the organisation finding itself in court. Trusting AI to do the right thing will not be considered an acceptable defence.

What next?

The aim of this piece is not to put you off AI, but to be more circumspect in how you incorporate it into your organisation. Experiment with AI. Build new AI revenue streams. Have fun with it. But for it to deliver at an enterprise scale, you need to ensure that both your organisation’s data is of a high standard and that your organisational design can take the associated pressure.

In summary, sprinkling an old school model with tech pixie dust is not going to end well.