Dhanada K Mishra
“If you are not interested in Artificial Intelligence, then you are not interested in interesting things,” says Prof Patrick Winston of MIT, a researcher of AI. If you used your smartphone to type a message and found the autocorrect feature altering your spelling or grammar, it is thanks to AI. Suppose you see your e-commerce platform Amazon suggesting products that you may want or need, then it is using the recommendation engine powered by AI. If Google photos can pull all the images that match your face from all over the internet, it is using AI-powered face recognition technology. In other words, AI is becoming ubiquitous in our everyday life whether we like it or not.
The history of AI goes back to 1956 when John McCarthy first coined the term. According to the Encyclopedia Britannica, AI is “the ability of a digital computer or computer-controlled robot to perform tasks commonly associated with intelligent beings.” One of the fundamental concepts of AI, and perhaps the best known, is ‘machine learning,’ owed to Arthur Samuel. In particular, he was famous for his Checkers-playing programme created in 1959, one of the first that ended up beating the creator in the game he created. The subsequent conquest of AI in various games such as IBM’s Deep Blue in Chess, and Google’s Deep Mind in the board game ‘go’ is legendary.
The easiest way to understand AI is to know that it is a paradigm shift in our approach to problem-solving. The conventional approach relies on a rule or mathematical formula that processes input data to answer. Whereas machine learning, a core concept in AI, comes up with the rule given enough input data with answers are known a priori. For example, one of the fascinating early applications of AI was for computers to read handwritten numerals – for example, a postcode on letters. The AI programme can be trained to recognise any new unlabelled sample using a set of 70,000 labelled handwritten image samples of numbers 0-9. This is the equivalent of the ‘hello world’ programme of machine learning in computer vision.
There are three main categories, or paradigms of machine learning: supervised learning, where defined (so-called ‘labelled’) data and examples are submitted to carry out the task as above; unsupervised learning, where the task is carried out autonomously and usually allows for the identification of structures and trends; and finally, reinforcement learning, where actions are learned through the experiences themselves, as the programme is told when it goes wrong as in the case of the game-playing AI.
Similarly, an increasingly popular subset of AI is ‘deep learning,’ a concept introduced in 1986 by Rina Dechter. Deep learning is a specific class of machine learning methodologies based on artificial ‘neural networks’, systems inspired by the neural logic of the brain. Going further than ‘basic’ machine learning, deep learning uses several layers of so-called ‘neurons’ or a set of logic gates to extract more complex properties from raw data. These methods, particularly deep learning, are used for various end applications such as facial recognition systems or automatic speech recognition. Finally, there are two types of AI. First is the narrow AI, or ‘weak AI.’ It is the AI that is being actively studied and used today – an AI is confined to one or even a few specific tasks with a controlled environment and data. Later, it is expected to evolve into artificial general intelligence (AGI), or ‘strong AI.’ Google recently unveiled its ‘Pathways’ project, which promises to provide a general-purpose AI tool for solving heterogeneous problems that process mixed-mode data such as visual, audio or electro-magnetic in a more efficient manner.
AI can be a double-edged sword when it comes to the biggest question facing humanity – climate change and sustainability. With the explosion of AI applications running on vast amounts of data and massive computing power, the technology’s energy consumption and carbon footprint are of grave concern. It is projected to grow exponentially over the next decade at a CAGR of nearly 44% globally through 2025. At the same time, AI is being increasingly deployed to address questions like the energy consumption of buildings that account for almost 40% of all emissions. For example, Google’s Deep Mind division has developed AI that teaches itself to minimise the use of energy to cool Google’s data centres that have a very significant carbon footprint. As a result, Google reduced its data centre energy requirements by 35%.
AI can be a net positive contributor to environmental sustainability in many industries. In agriculture, AI can help reduce both fertilizer and water use, all while improving crop yields. AI can use deep predictive capabilities and intelligent grid systems to manage the demand and supply of renewable energy. In transportation, AI can help reduce traffic congestion, improve cargo transport, and eventually help with the ‘last mile’ delivery problems. In facilities management, AI can help recycle heat within buildings and maximise heating and cooling efficiency. It can help optimise energy use in buildings by tracking the number of people in a room or predicting the availability of renewable energy. ONLY TIME CAN TELL whether AI will be an energy-guzzling monster that will make the climate crisis worse or help save humanity from the impending disaster.
Prof Marcus du Sautoy, a mathematician from Oxford University, in his book ‘Thinking Better – the Art of the Shortcut in Math and Life,’ makes an interesting observation. He believes that the very reasons that make AI and machine learning so powerful in solving intricate problems are what will ultimately save humanity. Humans discover all the intelligent shortcuts while confronted with complex issues that result in most innovations, including AI. Being endowed with great computing power and data, AI would never look for shortcuts, unlike humans. That may be the saving grace that ultimately helps humanity sustain or prosper irrespective of the intelligent machines!
The writer is a civil engineer, academician and technologist with a strong interest in the sustainability of the built environment. He is currently based in Hong Kong.