AI is the new electricity. If you’ve been anywhere near Artificial Intelligence or Machine Learning, you’ve likely heard this phrase being said often, and usually by people who may not know too much about the topic. But still: I personally love that line – just not in quite the same way as most people mean it. So, who’s right? Well, I bet a story might help you see things my way.

The year 2023 saw a slew of problems pile up: war, climate disasters, and the escalating cost of, well, everything. Considering those soaring expenses, my best friend will not have been the only one trying to do as much as possible himself when building his house. And despite him not being a very handy guy, to my surprise, he managed to pick up enough electrical skills in just a few weeks to wire the entire house himself – saving thousands of euros in the process. Now remember: electricity is still a lethal force that most people don’t really understand, and the systems powering almost every device in our homes are fairly complex. So, how did we reach this point where today any random person can easily learn to design, build, and safely use electrical systems?

Well, the phases through which electricity evolved are those of really any broadly useable technological innovation. Initially, the ancient Egyptians and Greeks observed and studied natural electrical phenomena, but it took another 2000 years to really start the age of electricity, marked first by its use in small experiments and parlor tricks. In the 1700’s, one notorious trick involved suspending a child in ropes and then electrifying the child, so it could lift feathers from a distance and shoot sparks from its nose! At fairs and markets people were fascinated to get the thrill of being shocked by electricity, something children today still do with electric fencing. And electric fields and shocks were used to treat every imaginable bodily issue: infertility, tootaches, tuberculosis, baldness, fatness, and even a general lack of intelligence. Nobody really understood electricity, and all these experiments were rather dangerous: Benjamin Franklin once shocked himself senseless while trying to prove that turkeys killed by electricity were more tender then those butchered in other ways, and a contemporary scholar of his named von Richtman died repeating some of Franklin’s experiments with lightning. The early 1800’s saw a next phase where the understanding of the nature of electricity leapt forward, with scientists like Ørsted, Ampère, and Maxwell, evolving into another phase in the late 1800’s with numerous innovations and novel applications, including lighting, telecommunication, electric engines, from inventors like Bell, Edison, and Tesla. And finally, following this very closely was the phase of electrical safety, where good practices and safety techniques were identified, although it wasn’t until the 1950’s before electrocution was well understood, leading for instance to the development of the electrical circuit breaker. It is this journey from observing phenomena, experimenting and playing, then understanding and applying, and finally safeguarding that brought us to where we are today, with knowledge, mass-produced products, and best practices and standards, all together allowing my friend to do very advanced things very easily, and completely on his own.

Now, when people say data and AI are the new electricity, or the new oil or gold, they typically mean that AI is ready to be this omnipresent force, widely supplanting activities within companies, effectively, and efficiently, and hugely valuable. However, those people typically fail to consider the expectations this creates, because electricity has at least 4 key characteristics: first, we know what it is and how it works. Second, we know what we can do with it, and how to make it work for us. Third, we know how to use it safely, and fourth, and very importantly: electricity works in exactly the same way in the office building of one company versus the factory of another versus my friend’s private home. Only by combining all 4 characteristics, electricity becomes an off-the-shelf commodity like oil or gold.

As an AI leader, it is vital to understand that none of the 4 above are exactly true for AI. Firstly, even experts struggle to understand the inner workings of some methods: why are the characteristics of deep learning systems what they are? Some things that should work, don’t, and vice versa. Secondly, we are still finding new applications every single day, and we are literally in the midst of figuring out which of those applications work and are valuable, while poor and stupid use of AI is still rampant. And while there are huge AI successes in the comfort of our homes or in our everyday work, many people are still stuck in that second phase of laughing about parlor tricks like yet another Snapchat filter. On the third element, the field of ‘trusted AI’ is still in development and every day brings new stories about ‘AI gone wrong’, even at the hands of prominent organizations, underscoring that safe use of AI is not yet a matter of simply adhering to standards, no, it requires extreme caution and specialized customization. And finally, and perhaps most important for an AI leader in business: what proves effective for one organization, may not work at all for another: the data is always different, the context differs, and the ability of organizations to reorganize around AI-driven actions differs.

With one type of AI we discussed in the previous episode, the Generative AI, we’ve reached a point where anyone without any expertise can go into Microsoft Azure, and use a foundation model like GPT-4 to build within just a few hours a working AI system that goes way beyond anything that was possible even a year ago. So can we compare this to my friend setting up his own electricity, easily and safely? Well, even something wonderful like GPT-4 for language understanding is still a while away from being a bug-free product that out of the box perfectly supports the full scope of use cases. And when it comes to that other type of AI we identified, addressing those ambiguous decisions, its deeper context-dependence makes it even less standardized and out-of-the-box.

One reason for this present state of AI is clear: for electricity the phases of observation, experimentation and knowledge building, public tomfoolery, application invention, and the establishment of safe use happened somewhat in sequence across hundreds, if not thousands of years, but for AI most of it is taking place all at once in a mere decade, before our very eyes. All this happening in so little time has many consequences, just one being that most people received their education and spent the majority of their career in AI-free environments, and everybody, including the experts, is really just inventing how to behave with AI as they go along. And in contrast to electricity, where the regulations came after the usage had sort of settled, well-intended and promising regulations like the AI Act are coming in the midst of the development of the field of AI. So while these regulations aim to provide a robust framework for safe use, at the very least they risk being found as impractical, or unsuitable for the novel approaches to AI that are still surfacing every day.

So all things considered, saying that AI is the new electricity puts fair emphasis on the wonderful potential AI holds, and may help you achieve breakthroughs in mindset in more conservative organizations. But as an AI leader, you should also avoid inflating expectations with this sensational analogy, so I advise every AI leader to say “Yes, AI is the new electricity, but it really still is that electricity of the mid 1800’s”: not completely understood, not fully settled on what can best be done with it, and not by default safe for use.

Thank you for listening to this third episode, I hope you enjoyed it, or that it at least got your thinking going. Because, this was just one of a 101 stories to cement your AI leadership, and many questions remain. Luckily, we still have a few episodes to go – 98, if the title is anything to go by. Next time, I hope to offer still some more historical context, and explain why you should remind people that AI really puts us in a unique moment in time, the year 0, so to speak.