I was honoured to be one of the speakers at the recent AI Summit for business in San Francisco. Here’s a summary of thoughts and learnings that I took away after two days of amazing talks and demos from some of the most innovative companies in the world.
AI is synonymous with machine learning
Connectionist, sub-symbolic approaches to mimicking human cognition are enabling the fourth industrial revolution by providing the technical solution for managing data of colossal volume and complexity. The digital revolution of social media and unstructured data was merely the precursor to the data cataclysm of the 21st century. More, and more complex, data means that we humans will be simply unable to process and comprehend without the help of AI. Meanwhile, our lives become more complex and interconnected as the benefits of personalisation and data optimisation require accessing multiple systems. We are seeing the last days of “apps”, as well as of mobile devices and laptop computers as the de facto machines that manipulate information. The more complex the data the less pervasive the hardware will have to become, till one day it will surely become invisible. Information is fast evolving to a utility. The information infrastructure, like the electrical generation and distribution infrastructures, will be removed from our everyday experience. Artificial Intelligence will provide the communication layer between humans and information. The mediators between us and the machines will be a network of machine intelligences capable of applying machine learning to a wide spectrum of data. Collectively, these networked intelligences will be able to deliver whatever information is necessary, at the right time and place. Such multiple, back-end intelligences, may be “unified” at the front-end by a personal virtual assistant, the “AI mediator”.
They will speak our language
Natural Language Understanding has gone into high gear. The need for machines that we can talk to and they can understand us is profound. Till now machines have been clumsy with language. Siri is like Dorie the fish; she forgets everything, who you are or what you are looking for, after a single exchange. You cannot possibly have a conversation with something like that. But the day is approaching where your personal digital assistant will not only be able to converse fluently and remember what you and “it” (”him”? “her”?) are talking about – it will also change its tone of voice by sensing your emotions. The assistant and you will be able to forge a relationship, it will be a trusted friend. Which poses a series of interesting question: for example, what happens to the intimate knowledge about you that the machine holds? Who owns this knowledge? How is it used, and by whom? And what happens when this knowledge is hacked? The more we move from carbon-based intelligence to silicon-based intelligence the risk of cyber hacking rises.
Big Data is just data
Industries across the whole spectrum discover the joys of machine learning. A lady from a big oil company described how machine learning is used to save millions of dollars during oil exploration by predicting catastrophic failures and giving time for preventive actions. Another lady from a major agricultural solutions company illustrated how machine learning is revolutionizing harvest yields. But the more one moves into the world of human interaction data and machine-learning manipulation of data starts to become less objective. Human bias enters the equation. Take for example a system that opines on the creditworthiness of an individual based on demographics such as zip code. Humans confuse easily correlation with causation and this bias enters the process of supervised learning and replicates human biases in the machine. We often forget that machines can only correlate data, and that you need a scientific, controlled, experiment to establish causation. And yet this nuanced but profound point may go amiss in a world where automation delivers highly-valued efficiencies.
The industrial internet
The Internet of Things is fast becoming a reality. Manufacturing and industrial companies are entering the fray of machine learning, furnishing everything with intelligence and the ability to communicate. A fully automated factory is a far-fetched idea anymore. And neither is the idea of having machines maintaining and fixing themselves without human intervention. Automating what have hitherto been human-driven processes suggests a further deterioration of mid-level, mid-skilled human jobs in manufacturing and industrial plants.
Impact on work: Trains, trucks, ships, and airplanes
The medium-term impact of AI on work is gradually becoming apparent. The transportation industry will probably be the first “victim of cognitive automation”. The impact may be dramatic, and similar to what happened when the media and entertainment industries which went digital ten years ago. Traditional business models in transportation will crumble, and millions of people will become jobless. There are approximately 3.5 million professional truck drivers in the US. If one adds jobs along the whole supply chain more than 8.7 million jobs will be affected. Driverless cars will come on line very much later, when the adoption and social acceptability curves have smoothed out. Fully-automated ships and trains are a lot easier to implement.
But what about airplanes? It is most certainly possible to have a self-piloted airplane, we already have drones that are self-piloting, or piloted by remote. But would you fly on a plane that does not have a pilot? Who knows? Maybe in the future when airplanes will be able to self-maintain and self-heal, flying without a pilot will be seen as perfectly alright…