Separate names with a comma.
From algorithmic bias in self-driving cars to Google’s ranking of useful and relevant content, AI is a frequently discussed topic on the forums. When DebtCapitalMarkets queried how prepared we should be to adapt to AI’s fast paced changes in a recent thread, it reinforced the popular perception of AI and machine learning – or, as Popular Science put recently, “the pre-emptive myth of the evil robot”.
Our thinking is constrained by how creepy we find the idea of non-humans with adaptive, human intelligence.
But we no longer need to predict the future because it's not just some distant horizon anymore. It's right now. The World Economic Forum has already declared "the Fourth Industrial Revolution" and research into AI is further along than most people realise. “It has made tremendous leaps and bounds in recent years. What we’ve been able to achieve in the past 5-10 years has been substantially more than any other period in the past,” says Professor Parham Aarabi, a professor of engineering at the University of Toronto and entrepreneur involved in AI.
“There’s a lot of room to grow but the systems are certainly very capable. What I find really interesting is that they are now, in many areas, at the same level or slightly higher than that of humans. AI systems are now on par or slightly better than humans.”
In the media, the terms of ‘machine learning’ (ML) and ‘AI’ are often used interchangeably. This isn’t quite right. As Aarabi explains, AI is the overall area and machine learning is one way of achieving AI. “Historically, there have been different, competing methods of achieving AI,” he explains.
Out of these competing strategies, ML has provided the most breakthroughs. “That is, machines that can learn from input. Just like babies that are learning and growing. ML is essentially that same idea – getting machines to learn from data and inputs and be able to achieve intelligence.”
But even ML is now becoming more and more refined. The most popular approach to AI 30 years ago was ‘expert systems’. The idea behind it was creating a machine that learns from expertise. “A system that can learn from doctors or engineers based on the inputs they provide,” explains Aarabi. “The problem with those systems is that they could only achieve a certain level of competency based on what the human expert would provide.”
Expert systems work really well when they’re combined with ML, however. “What we found is that when you took the expert input and provided it to the machine learning component – so, letting computers learn beyond the information that the expert provided – it was able to identify algorithms that humans could not find.”
This is very high-level stuff – but these applications have already started to seep into the market. Aarabi, for instance, has used his research to design a computer vision system used by retailers. If someone, for example, shares a photo and there’s a lipstick or a purse in that photo, Aarabi’s technology is able to understand that image and tell them ‘oh, by the way, if you’d like to buy it – this is the actual product’.
Its wide applicability is thanks to the reams of data the modern individual produces. All of this data – yours and mine – can be modelled mathematically. “Every day, people are producing their own personal algorithms based on the data being acquired right now,” says Professor Massimo Fornassier, chair of numerical analysis at The Technical University of Munich (TUM).
What’s really changed the game now, as Fornassier explains, is the “explosion of new methods to acquire and interpret data”. For the purposes of a business, this means more ready access to a concept previously sequestered to large corporations and university campuses. Acquiring, processing and storing data is now cheaper than ever.
This revolution as we know it can be traced to a shift in thinking within mathematics. “Applied mathematics, in the 20th century, was pretty much devoted to physics,” explains Fornassier. “Physics is something that’s absolute, you can repeat it. Physical forces are universal.
“But now, maths has started to model things that are man-made. The digital revolution has been happening in a tremendously fast way. We’ve learned we can go beyond this security; we can be a little more daring.”
This is key to why the business world is suddenly aflame with AI optimism (and angst). Social dynamics are still unsure; what you learn about social behaviour won’t be valid in five days, for instance. But ML overcomes this by constantly learning and re-learning. The vagaries of a consumer’s behaviour can be catered for by a machine that makes micro-perfections and small-time predictions.
It’s got to a point where we are now talking about ‘neural networks’. “A neural network,” explains Fornassier, “is a mathematical model realised eventually in the form of a computer as a concatenation of units that simulate the function of a neuron.”
It’s hard to convey how significant this is but, to put it bluntly, machines are muscling in on our turf.
The impact these developments will have on commerce is gigantic. So big, in fact, that like the movement of tectonic plates, they’re unnoticeable without explicitly researching and confirming them. No one is exempt; Google, for instance, is now reimagining itself as a ‘machine learning first company’.
“The reason is because they have a lot of data,” says Nick Heller, a former Googler and now founder of the AI startup Fractal Labs. “As data becomes more and more available, the algorithms around ML become stronger and stronger, and better and better.”
According to Heller, this preponderance of data is what's driving AI forward. And what’s exciting, and what Fractal Labs’ own AI financial assistant counts on, is that data is accessible to anyone who's paying attention, not just Fortune 500 companies.
In Fractal Labs' case, financial information for private businesses is now available because accounting data has moved to the cloud. "Or," says Heller, "it's available in some form where we can pull it in the cloud.
“Once you have that data you can apply algorithms to that data and learn from it. For instance, you can take data from private companies and compare them to how public companies perform. We can elicit patterns from that comparison and give that information to a smaller business who’d never be doing this sort of analysis on their own.
“It’s the wisdom of the crowd. With all that data across a number of small businesses is available you can now leverage that to help a single business.”
But there are some misconceptions among business as to how these algorithms work, Heller explains. “The misnomer with AI is that you just need to apply algorithms and they’re just automatically gonna learn,” he says.
Firstly, any data needs to be prepared so that it’s ‘clean’. Secondly, Heller says, “It’s the questions you ask. You have to apply it back to a real customer journey and a real set of questions you need answered. Only when you ask the right questions to set up the right experiments will you actually be able to come up with meaningful results.”
The new AI world will require companies to attain a new kind of data fluency. The companies that most effectively augment human judgement and decision-making will ultimately be the real winners. The human still needs to be there because, as Heller points out, “We don’t know what we don’t know”.
“If we can’t pick up a data point on it, we wouldn’t know it. I wouldn’t know if the owner of a business concluded a handshake deal over lunch. I can’t know that. Which is why the human judgement is still important and still more important than ever. You need to pepper in with all this data, inputs from the user. And, in turn, the AI learns from that, too.”
From Parham Aarabi’s vantage point at the University of Toronto, small businesses have adapted rather well. “We’re seeing that even smaller companies are finding new directions in which they can use this technology and pushing the boundaries of AI with great success.”
The trick will be to, like Heller says, tie these advances back to your customers. Customers who are fleshy, fragile bundles of neuroses. It’s uncertain, to say the least. Colin Angle, the co-founder of the American advanced tech firm iRobot, perhaps said it best:
“It’s going to be interesting to see how society deals with artificial intelligence, but it will definitely be cool.”