By Louis Fourie
Every year typically has its defining moments. The year 2020 is no different and contained several paradigm-shifting and world-changing developments. In fact, at the end of 2020 we are still fighting a deadly “pandemic” – the Merriam-Webster’s word of the year 2020.
For many 2020 will be a lost year, remembered for the immense loss of precious human lives due to the pandemic and many unrealised personal, family and career goals. Countless important events, celebrations, and family get togethers were unceremoniously cancelled or forced onto a somewhat impersonal virtual platform.
But 2020 will also be remembered for the swift action of proponents of artificial intelligence (AI) and many remarkable breakthroughs. Two weeks ago I reported in this column about the half-century-old challenge that has been solved by AI. DeepMind (a division within Google doing ground-breaking work in machine learning) then announced that their AI system, AlphaFold, solved the half-a-century old challenge of “protein folding” by computationally predicting a protein’s 3D shape from its amino-acid sequence with incredible speed and precision. This solution is a major advance in biology and makes the development of medical treatments easier for a range of diseases from cancer to the coronavirus.
As the founder of DeepMind, Demis Hassabis, said “algorithms are now becoming mature enough and powerful enough to be applicable to really challenging scientific problems.” This breakthrough is typical of the Fourth Industrial Revolution (4IR) era where intelligent technologies and devices are increasingly permeating our lives. AI is one of the most noteworthy technologies of the 4IR and is changing how we live, work, and communicate. Its influence on government, education, healthcare, and commerce is profound as have been discussed in this column during 2020.
Just the contribution of AI to relieve the crisis caused by the Covid-19 pandemic is remarkable. Damo Academy, the research institute of Alibaba in China, has developed an AI algorithm that can detect the coronavirus from chest CT scans within 20 seconds with 96 percent accuracy. In South Korea, the molecular biotech company Seegene used AI to accelerate the development of Covid-19 testing kits and was thus able to produce a solution within three weeks instead of the normal three months development.
Benevolent AI in the United Kingdom used AI to sift through thousands of research papers and other data for existing drugs that could be used to treat coronavirus patients.
AI was also used in the race to produce a vaccine. Vir Biotechnology and Atomwise in the USA used AI algorithms to isolate a molecule that could assist in the treatment of the virus. Due to the time pressure on researchers and pharmaceutical companies to develop a vaccine against the coronavirus, algorithms were used to analyse the data for potential adverse reactions. This dramatically shortened the testing time of the vaccines.
All over the world and also in South Africa such as the Wits University Covid-19 Dashboard (https://www.covid19sa.org), AI algorithms have been used for epidemiological modelling and forecasts regarding the spreading and second wave of the virus.
Without doubt, the impact of AI in this traumatic year of 2020 has been profound. In fact, AI has made impressive progress in the last twenty years. Intelligent computers can now read medical images and make diagnoses with ease, accurately predict customer behaviour, manage financial portfolios with automated transactions, act as a psychologist, compose music and poetry, and generate art. In many things today AI are even outperforming humans.
The dramatic progress in AI can primarily be attributed to a class of algorithms called artificial neural networks that process large quantities of data and extract statistical patterns from it. To perform tasks, they merely match the input data to the most relevant patterns to compute the result. This computational pattern matching is incredibly powerful and can simulate many functions of human intelligence as was proven when the IBM supercomputer Deep Blue in 1997conquered the world chess champion, Gary Kasparov, in a multi-game chess match with 4 to 2.
In March 2016 a Google-owned computer at DeepMind defeated, to the amazement of many AI experts, the best Go player in the world, Lee Sedol, in the very complex 2500-year-old Go game. To achieve this, researchers created a sophisticated AI programme, AlphaGo, consisting of remarkable 13-layer-deep neural networks and a Monte Carlo Tree Search technique.
OpenAI, a non-profit company, formerly sponsored by Elon Musk, released its latest natural language processing system called GPT-3 in July 2020. GPT-3 displays uncannily high, almost human-like, versatility and is exceptionally good at executing a comprehensive range of language tasks, from writing stories to composing e-mails. It is able to translate between numerous languages, write technical documents, answer common sense and reasoning questions. GPT-3 is so amazing that the output generated, is indistinguishable from human-written content.
When compared to its predecessor GPT-2, GPT-3 is qualitatively similar, but quantitatively much more powerful. GPT-3 has 175 billion parameters versus the 1.5 billion of GPT-2. It thus seems if intelligence is a function of computational complexity. Does this mean that once AI progresses quantitatively to reach the trillions of parameters equal to the trillion of neural synapses in the human brain, that computers will be as intelligent as a human or even become more intelligent?
Indeed, computers can understand verbal commands, translate languages, recognise faces, drive cars, and play games better than humans. They can master strategic thinking, tactical actions and risk analysis, as well as display imagination, creativity and foresight. Moore’s law has over the years shown that since Charles Babbage’s Difference Engine in 1822, computer power and speed are constantly increasing. According to Ray Kurzweil, an American author and Director of Engineering at Google, computers will reach the capacity of the human brain in 2030, in particular due to optical, quantum and DNA computing.
However, AI still has some way to go, since it is often programmed to solve problems based on rules for a specific task, while humans are able to use their intelligence in a wide variety of contexts. Humans are endowed with social skills
and also display an emotional aspect, often referred to as emotional intelligence (EQ). Unlike computers, humans learn by making a personal connection with their sensory organs to develop and interest in what is being learned. When humans learn, all sensory capabilities are channelled towards the learning, including emotions.
But all over the world research institutions and companies are working on a totally new generation of computing technology, such as quantum and neuromorphic computers (mimicking neuro-biological architectures of the human nervous system) that could bring AI to the next level. Applications that would have been dismissed as science fiction not long ago are now reality.
Some scientists are therefore convinced that not only will AI reach the human level of intelligence, but may soon surpass it due to superior memory, multi-tasking ability, and its almost unlimited knowledge base. These advantages could give computers better analytics and in-depth decision-making heuristics than the human brain. While, quantitively, the human brains are limited by biology, computing systems are not subject to such limitations due to the continuous development of new technology - well at least in theory.
But as AI systems constantly become smarter, it is inevitable to wonder if human intelligence is qualitatively different from artificial intelligence (AI) or if the differences are only quantitative. Currently computers mostly emulate human beings by reducing their abilities to recognisable patterns – understanding numbers, language and thinking. But is possible to reduce all human intelligence to mere pattern matching? Or does human intelligence entail something qualitatively that cannot be recreated computationally? Perhaps something not only in our brains, but also our genes?
AI will bring many benefits to the world in the years to come, amongst others solving some of our greatest challenges. But would AI become a threat or could it be that the real opportunity of AI is to eventually unleash the full power of human intelligence and unique human skills? Could AI perhaps make us more human in the future?
In about two weeks, 2020 will come to an end. As it begins to recede into the past, how will we look back on this anomalous year? Will we see it as a year of loss due to the pandemic or the immense growth in technology and AI solutions to rise to the challenges caused by the pandemic?
Prof Louis C H Fourie is a Futurist and Technology Strategist