Artificial Intelligence — what CTOs and co need to knowAs part of Information Age's Artificial Intelligence Month, we summarise everything you need to know about the technology on everyone's minds
The trouble with the word artificial intelligence is the word intelligence. It misleads — people conjure up images of thinking machines, Stephen Spielberg; Arnold Schwarzenegger coming back from the future and saying: “I’ll be back.” Who knows what the future may bring, but for now, and for all intents and purposes, people may be confusing intelligence and sentience. Neither are well defined — we are supposedly sentient, machines are not, and for all we know, may never be.
Intelligence means “the ability to acquire and apply knowledge and skills“. These days machines can do that, machines learn, they deep learn, they can even learn by applying neural networks — that does not make them like people, maybe they possess one subset, of a myriad of sets, that make us who we are.
It is never easy to find a beginning — every event in history had a cause that had occurred before; keep going back and you get to the Big Bang. So when did artificial intelligence begin? Did it begin when Alan Turing defined what is now called the Turing Test? Back in 1951, in a paper with the title Computer Machinery and Intelligence, he envisioned three rooms. In one sits a man, in another a women and in a third a judge. The man and women are asked a series of questions, the judge has to identify from the answers in which room sits a man and in which room sits a women. One of the two people is then replaced by a computer. Will the judge draw a different conclusion?
Artificial intelligence: Transforming the insurance industry
Maybe we can go back further to Ancient Greece and the myth of Talos, a giant mechanical humanoid — perhaps the first robot equipped with what we might now call artificial intelligence.
But to find the first time this phrase was actually used, we need to forward wind the clock to 1956, at Dartmouth College, New Hampshire. This was the location of the Dartmouth Workshop, organised by John McCarthy, a famous American computer and cognitive scientist. It was at this eight-week long event, McCarthy coined the phrase: Artificial intelligence.
Intriguingly, McCarthy initially lost his list of attendees, but the list that finally emerged read like a who’s who of the great and good in the story of AI: Marvin Minsky, Allen Newell and Herbert A Simon: together these men are known as the ‘founding fathers of AI’. Incidentally, John Nash, the mathematician famous for his work in game theory, a winner of the Nobel Memorial Prize in economics, and subject of the film: Beautiful Mind, starring Russell Crowe, was also at the conference.
McCarthy later said: “Our ultimate objective is to make programs that learn from their experience as effectively as humans do.”
These days, many argue that artificial intelligence and machine learning are interchangeable — that’s not strictly accurate, it may be more accurate to say that machine learning is a subset of artificial intelligence, as its super smart sibling, deep learning.
Machine and deep learning
Earlier versions of AI, such as IBM’s Deep Blue, which, to a fanfare of publicity, defeated the world’s top chess player, Gary Kasparov in a series of matches, culminating in decisive victory in 1997, was different. Deep Blue was mainly about muscle — the computer equivalent of horsepower, throwing processing power around, analysing each move and all the potential permutations and combinations. Scientific American interviewed Murray Campbell, who, as an employee of IBM, worked on Deep Blue.
He said: “The 1997 version of Deep Blue searched between 100 million and 200 million positions per second, depending on the type of position. The system could search to a depth of between six and eight pairs of moves—one white, one black—to a maximum of 20 or even more pairs in some situations.”
Taking the next step with AI adoption — overcoming the problem of data
Deep Blue was also developed with a lot of human involvement, chess grandmasters, for example, to help develop the program.
By contrast, what machine learning and deep learning have in common is that they do what their name suggests — they learn. The difference largely relates to the amount of human involvement, with human programmers and designers taking a more proactive role in defining parameters in machine learning. With deep learning, computers often learn from multiple data sources, extrapolating data from quite unrelated areas.
Learn like we do
A good analogy between machine learning and older forms of AI might relate to sport. If you play a sport, such as tennis or squash, your brain does not calculate the trajectory a ball will travel in using advanced mathematical formulae, applying the rules of geometry. It learns that if you hit the ball in a certain way it then reacts in a certain way. If you hit it in slightly different way, it reacts differently. It can then extrapolate from this how the ball will approximately react if you hit the ball in a way that is somewhere between these two practised shots. In short, we learn. If a computer was to apply mathematics to calculating the trajectory of a ball depending on the angle it was hit at, the speed of shot, texture of surface to calculate bounce and a myriad of other different variables, the necessary computing power would be enormous. If, instead, it learnt from studying previous shots (and built a predictive model from data), then the processing power required would be a lot less.
Investing in AI: What businesses need to know
Jeopardy to Go
Many years after Deep Blue defeated Gary Kasparov, IBM Watson defeated the best players in the world at Jeopardy, the US quiz based game. That was in 2011. What was impressive about this victory is that questions contestants are required to ask as part of the game are often quite ambiguous. It was a famous victory for Watson (which was named after a IBM’s first CEO, Thomas Watson, and not, Sherlock Holmes’ sidekick.) But clearly, Watson did not understand its answers, there was no common sense, for example answering ‘Dorothy Parker’ instead of ‘The Elements of Style,’ to one question. In fact, ‘The Elements of Style’ is a book describing writing guidelines, Dorothy Parker, an American poet once recommended it.
Watson’s victory in Jeopardy was impressive — but a computer that could win the Chinese game of Go, an abstract board game with 2×10170 permutations — which is significantly greater than the number of atoms in the known universe — seemed to make winning Jeopardy more like child’s play. Such was the achievement of AlphaGo, a computer program created by Alphabet subsidiary DeepMind.
Whilst Watson applied machine learning, AlphaGo applied deep learning. In March 2016, AlphaGo defeated Lee Sedol, 18-time world champion at Go over a series of five games.
In achieving its victory, AlphaGo evaluated roughly a thousand fold less positions than Deep Blue did in its victory over Kasparov. It secured its victory by evaluating game play.
Impressive though that was, in 2017, AlphaGo Zero, managed a far more impressive feat. This time, the program didn’t learn the game by evaluating the game play from human competitors, rather it learned the game from scratch — playing against itself. All it needed was the rules. Initial gameplay was selected at random, by trial and error and via selecting the fittest gameplay, within three days it could surpass AlphaGo.Within 21 days it was at the level of AlphaGo Master, a later version which had defeated 60 professionals and the world champion online, and within 40 days it had arguably become the best Go player in the world, or so claim DeepMind.
Increasing the adoption of ethics in artificial intelligence
Deep learning and machine learning both apply neural networks.
Neural networks have been going in and out of fashion. Warren McCullough and Walter Pitts, from the University of Chicago, first proposed the concept in 1944. They came into vogue in the 1980s, went out of fashion in the noughties, but are now back.
It may be that until recently, computing power was insufficient to do neural networks justice, it may be that we just didn’t know enough about how they could work.
Like the human brain, neural networks consist of thousands, maybe millions of processors forming nodes — although, the human brain has over 100 billion nodes. The nodes in a neural network tend to be organised in layers, such that each layer of nodes may be given a specific task.
AI can have applications in many guises, including autonomous cars, voice assistance and voice recognition, image/face recognition, personalised health monitoring, advertising and online shopping — such as identifying products that may be of interest to customers based on analysis of data, search, finance trading and in the war against cyber security.
AI can have applications in:
• AI analytics
• AI business processes
• AI data management.
Artificial intelligence: Data will be the differentiator in the marketplace
According to a recent report from McKinsey, AI has the potential to deliver additional global economic activity of around $13 trillion by 2030, or about 16% higher cumulative GDP compared with today.
And science fiction
The realm of science fiction and AI appear to meet in the imagination of the media. But do we need to worry about the more-scary predictions of science fiction — AI gaining awareness? Is AI a threat to humanity as the late Stephen Hawking and Elon Musk, suggest?
Speak to most experts in AI, and they laugh at the idea. But maybe, in asking whether AI could one day become sentient and more intelligent than humans, we ask the wrong question. Instead we should ask, could AI become more intelligent and sentient than an amoeba, because remember, it was from an organism not that dissimilar from an amoeba that all complex life, including humanity evolved. Okay, it took several billion years, but then in a digital environment, evolution could work several order of magnitudes faster.