Navigating The Sea of Data — Machine Learning for Human Beings
The rise of the humans
Drowning, not waving
We hoped that data might provide the answers. The more we know, the more we can do, the more targeted we can be, the less waste we will create. But information without context not only baffles and wastes time, it can also feel really disempowering. Over 70% of CEOs questioned in a recent survey for Forbes said they had disregarded data-driven insights in favour of more instinctive or intuitive decisions.
There’s real resistance to both the deluge of data and the automation of processes, for good reason — they often stifle agency, rather than enhance it. No-one wants to feel like they can’t move for information, or that the crucial moves are being made without them.
Look at Gmail’s recent roll-out of automated reply options. If you’re not a Gmail user, here’s how it works — the text of your incoming email is read by Gmail’s algorithm and analysed, then a selection of likely responses from you are presented before you start replying.
Got an email that talks about meeting at the cafe tomorrow?
With one click, you can now automatically reply with ‘Great!’, ‘Sounds good!’ or ‘See you there!’.
You can see the logic. A huge percentage of emails are quick, everyday communications. ‘See you at Dave’s later.’ ‘Thanks for your help.’ No problem.’ The faster you can get through those, the happier you’ll be, right?
Wrong. Within weeks of the feature being launched, users started to complain, for a number of reasons. Firstly, there was resentment that Gmail was assuming a tone and style of writing on their behalf. Secondly, it instantly created a social division — does your correspondent care enough to write you a reply in person, or are you only worth the click of an automated response? It’s the digital equivalent of a direct mail letter with your name ‘handwritten’ by the printer. Instead of making things better, it’s created an extra layer of consideration.
Plotting a course
No degree of data capture, or machine learning, or AI, is going to be of much use unless it’s helping human beings do something better. Instead of grabbing every possible byte first and sifting through it for meaning later, we should be learning to target specific issues where that data and automation can solve issues we can’t tackle alone.
Here’s an example of cutting-edge technology fixing a human problem in a very specific way.
Rembrandt’s famous 17th century artwork is the pride of Amsterdam’s Rijksmuseum. But fame takes its toll. Permanent display means exposure to light, air and at least two knife-wielding attackers.
In order to restore it, museum experts need to know as much as possible about the condition of the painting. The original oil pigments, the pigments used in previous restorations, the level and timeframe of how their degradation, how air changes them, everything.
Over 56 separate 24-hr periods, the canvas will be scanned using macro x-ray fluorescence and the team will take over 12,500 still images. The process is going to generate vast quantities of data on chemical composition, colour, depth, texture and so on.
By using AI to interpret that data, restorers are going to get to the answers they need — how to recreate original oil paints that match, how deeply to apply them, how to negate the effects of sunlight etc. — faster than they ever could unassisted. And with any luck, the Night Watch can be restored in such a way that we’ll all be able to see it as the artist himself once did.
In other words, this is the opposite of grabbing any and all data first and asking questions later. It’s a specific, tailored and entirely people-driven project. It’s using the power of machine learning and artificial intelligence to enhance a human being’s capacity to exercise considered and expert judgement, and it’s been designed with a specific goal in mind.
Start at the end
So if you’re drowning in data, or you want to use the power of machine learning and AI to make things better, then don’t start with the technology. Start at the end — by taking the time to identify what your ideal outcome would be.
Maybe you’d like to get a better permanent rate for all your global employees’ business travel instead of constantly switching suppliers.
Sure — you could immediately start building something that can calculate total number of miles travelled, method of transport, length of stay, average costs, destination frequency — every variable you can think of.
But the overall goal isn’t to know every detail of every trip for its own sake — it’s to reduce the waste of time and expense that excess corporate travel generates.
So how about analysing employee emails for travel and meeting references, cross-referenced for locations, length of meetings and expense claims, then offering text prompts in email that incentivise employees, with rewards for replacing physical travel with video conferencing?
Ask the right questions
Asking the right questions, from the right people, and understanding the context of those human responses, is a skill. It takes time, it takes expertise and it takes objectivity.
But wouldn’t you rather start your next voyage toward progress with the best map than the biggest boat?