Computers are getting faster, more powerful, and smaller. They are becoming ubiquitous, they are personal and “in-machine”, they are borrowed, stolen, manipulated, pimped, and applied in many different ways. They are in televisions, toasters and theatres. They are in shock absorbers, shops and shared resources like internet cafés. But fast, small, everywhere doesn’t necessarily mean smarter. It’s easier for us to talk wirelessly now, because of mobile phones. It’s easier for us to work from home, because of network improvements. But that’s all stuff we used to do anyway. So where is the “smart”? Where is the “clever”? Where is the “ahaaaa…”?
Maybe it’s enough that stuff gets smaller, faster, cheaper. That means we can do more stuff, and therefore we’re more productive. But smart means that we actually don’t have to do stuff anymore. Like looking things up. Making connections. Applying personality, or relevance to technology. Like search, for example. If we’re looking for stuff, we need to search for it. Sometimes we don’t know what we’re looking for either, until we learn about it – or see an ad. Sometimes, suggestion is all we need…
A few years ago (almost four years ago, in fact) BusinessWeek published a front page article called “Math Will Rock Your World“. Just as the quants had changed finance in the mid 1990s, now they were looking at other things, like advertising, propensity modelling, and other forms of knowing. And it is this knowing that has the potential to transform, well, stuff. What does knowing mean? What does awareness bring? We’re not quite talking about sentience here, but if we combine predictive analytics with context awareness, we can get pretty close. Statistically speaking, really, really close.
It’s a little like deja vu. It’s a little like an aging couple who finish each other’s sentences. Its like your dog knowing you’re going to feed them, or that you’re going to let them out one more time before going to go to bed. When you analyse these things, you figure out that there are hidden triggers in everything we do. We as people learn – and predictive analytics does exactly that, it learns – those triggers, and often we learn those triggers intuitively, in a hidden way. Once you have automated sensory perception and instrumented the data sources, the rest is simply maths. And the maths in many cases is not that complicated. We model behaviours that we wish to track and predict, we instrument and identify data sources, then we train the model and begin to predict, acting on those predictions.
Let’s take a simple example. Online shopping. We know that Anthony likes Bob Dylan, and tends to buy Bob Dylan records with reasonable frequency. The man isn’t quite dead, so he keeps churning out new stuff, and even if it’s re-recordings or remastered back-catalog, there’s stuff there that’s likely to be interesting to Anthony. We know that frequency pretty well – that it’s once every eight weeks he buys a Bob Dylan somethingorother. We know that he buys Music at the weekends – usually – and books during the week. Anthony himself probably doesn’t even know that. We know he browses in the mornings without buying. He buys on Thursday and Fridays and sometimes on Saturdays. We know that at certain times of the year – November, July – he tends to buy more than usual Christmas, wife’s birthday). We know that the stuff he browses for on Mondays he never buys. Put it all together, and we can know propensities that Anthony himself is not conscious of. Now, predicting the future is not simple. Just before the point of purchase, Anthony’s house could get hit by a tornado. The key point, howeever, is that it’s unlikely. It’s more likely that because it’s Friday evening in November, and Anthony hasn’t bought Bob Dylan in nine weeks, and there’s new Bob Dylan stuff out, and he didn’t browse it on Monday, that Anthony will go ahead and buy the new Bob Dylan compilation album at $9.99.
Fine – that’s business for you. And we can sell more stuff, personalise the shopping experience, and tailor the commercial engagement in a way that is likely to make Anthony happy, and make the CFO happy with the conversion rate per visit to the web site. The same logic applies to what you’re going to say next. We are all creatures of habit, we do things – subconsciously – in routines, in an order that we craft and hone and develop through our lives. We know that people often say “good morning” in the morning when they meet. We know that people stumble into the kitchen, trip over the cat, and scream some obscenity at the dog for chewing another leg of the chair. We know each person’s preferred obscenity. Or – rather than knowing – we can learn. And when disorder is introduced to our lives, we become confused and upset. So – simply put – if I usually scream tat the cat “you stupid cat” and my wife hears me, she won’t be concerned. If I roar at the cat “you f%^&ing c$%^ f$%^ of a b£$%rd f$%^er cat”, or something similarly unusual, she will say something to me like “ah he’s just a cat, why are you so upset?” These patterns, routines and expectations are mirrored in every home across the world. Machines – properly instrumented – can learn all of these things. Once the machine learns, it can predict. Taking a sequence of data, including triggers – like a visual sensor seeing the cat streak across the kitchen towards the feet of a sleepy and tired Anthony – the machine can predict that Anthony’s going to say “you stupid cat”, and then traipse over to the kettle. Again to be more accurate, the machine can predict that there is a certain probability that Anthony will say “you stupid cat”, though it’s never certain. A tornado may hit Anthony’s house just as he is about to utter those words, and the conventional predicted behaviour may not happen. Unless we’re monitoring weather patters as well. And – on that point – atmospheric pressure could well be taken into account as it impacts on mood.
I wrote a blog entry in a different blog in another lifetime about artificial intelligence and how through interaction with computers, perhaps people were becoming more like machines rather than machines becoming more like people. In a similar vein, perhaps we should think about how well machines can understand humans, and our behavior, before they begin to act like us, and become like us. And if machines know us, and can predict or behavior, that’s the point at which they can begin, well, to dream!