Automation or Intelligence

By Dr. Mayank Gupta, Partner, Collaborate and Eight

---

Automation or Intelligence? The Coffee Spill Test

There is a lot of chatter in the news and social media that Artificial Intelligence (AI) has taken over, accompanying them are fancy terms like singularity, human existential threat, etc. When I read such headlines in the past I was instantly scared and felt hopeless. However, after the initial despair fizzled out and my brain estimated the low probabilities for these incidents, I would return back to being normal, i.e. being human. 

As I would take steps to eliminate the threats, it would slowly dawn on me that AI wouldn’t take over in my lifetime - I am forty one years old with a resting heart of 47. Basic reasoning - comparing human genius/innovation with data retrieval from AI -  would lead me to the conclusion that the current version of AI is closer to automation than intelligence. I would then calmly go back to reading a book, putting the clickbait headlines tucked away in the pocket. I shall try and outline the difference between automation and intelligence below. 

What Automation Actually Is. 

Automation is repeating a process and becoming better at it by eliminating inefficient steps. Its final version includes a series of steps or protocols that must be ardently followed, popularly known  as an assembly line. Henry Ford’s intelligent mind - not automation - is widely accredited with developing the first assembly line for Model T. 

What Intelligence Actually Is

Intelligence includes automation and more. In the ‘more’ aspect of intelligence we lean towards non-tangible and non-quantifiable (can be vaguely quantified, but not accurately) aspects of human activity - observation, empathy, reasoning, problem solving, innovative and critical thinking. One could automate these steps of thinking with AI. It would create a rote learning method confined to an initially defined objective or purpose - we call them robots or parrot scholars.

The Cost of Blurring The Line

The non-tangible aspects differentiate us from machines and keep us relevant both at work and in society. But, we are in a rush to commercialise AI by inflating pattern recognition and rote learning to genuine understanding, replacing wisdom with speed, and data retrieval as innovation. These distortions may be harmless for large corporations and trend seekers, but it distorts public policy, misdirects investments by businesses and fosters false expectations in people and businesses. 

This thought process will lead us to self-fulfilling defeatism. But, let’s take a sober look at everyday scenarios that reveal the chasm between what AI does today and what intelligence truly demands. 

The Coffee Spill Test

Let’s assume you are in a work meeting. You reach for your coffee, the cup tilts over and a dark stain spreads across your silk tie. Can AI clean that spill today through its own decision? I believe today no large language model (LLMs - popularly masquerading themselves as AI) can assess the fabric, gauge the pressure needed to dab without smearing, decide between cold water and sparkling water, or negotiate the social grace of excusing yourself without disrupting the speaker’s flow. In the future, it probably will.

However, can the AI differentiate between a probable joke? Can it understand the seriousness of the room or not to interrupt a senior leader in the room? Can it read the room to know if a joke about the spill would ease tension? This is intelligence - emotional intelligence and social skills. Again, these can also be programmed. However, the networks that an LLM will use to access those million data points will create more overhead expenditure in the company’s books and irreversible damage on environment. The trade-off is a human brain making that decision in split seconds at the energy cost of less than 60W light bulb used for a minute. (Sometime soon we will discuss the energy cost and trade-off between LLM usage and human thinking)

Why Your Amazon Recommendations Are Wrong

Let’s step out of business meetings and look at online marketplaces. As we look at the recommendations on Amazon and Facebook marketplace, many have laughed at the recommendations or being surprised at them. But, how many have asked the question - ‘Why does it not suggest us what we really want even though it has so much data on our behaviour, choices and plans?’ 

If AI were truly intelligent, its suggestions would be near infallible, and our shopping carts would be perpetually full. Yet we ignore, dismiss, or laugh at recommendations for a third vacuum cleaner when we own two or discounted flight tickets a day later to the destination we already booked. (Skyscanner, I am looking at you!) Why? 

Can you recall the last time someone close recommended you a restaurant or holiday destination that you didn't try? I happily try the recommendations and have only come back disappointed once. The people close to you have less data on you than companies, but the former have the 'critical' and intelligent data on you. (Yeah, we can start a separate discussion on scalability of data but let’s leave it for some other time)

AI models are good at making us feel good by reducing our time to process thoughts and feeling. It diminishes our ability to sit with feedback and criticism, and process it. And, it is the feedback and criticism that makes our opinions, ideas and knowledge intelligent. However, the current sycophancy at display by AI models increases our bias towards our own parochial outlook, limiting the expansiveness of our own ideas and thinking.

We are decades away from True Artificial intelligence because we lack a theory of consciousness, a model of common sense, and a framework for subjective experience. So, I can assume AI wouldn’t take over in my lifetime - I am forty one years old with a resting heart of 47.