The Real AI: Augmented Intelligence
We explain why Augmented Intelligence, is the right AI for 2018.
Last year QuantumBlack joined a panel as part of the World Economic Forum, in Davos, and spoke about a term that was relatively unheard of back then: Augmented Intelligence.
Over the last 12 months momentum has built around Artificial Intelligence, which is what everyone means when they say AI. But we think that the other AI, Augmented Intelligence, is the right AI for 2018. The term has crept more and more into the media and into our vocabulary and that of the tech leaders we spend time with. There have also been signals in the market that have shown the importance of how machines, humans, and processes are coming together to form something very powerful indeed.
This isn’t a new idea. Ada Lovelace in 1842 envisioned a calculating machine and proposed that the machine could be made to determine “that which human brains find it difficult or impossible to work out unerringly” — which is another way of saying the machine would augment the human.
There’s a lot of concern about AI in the public discourse today, and we think we have to address those concerns head on in order to earn (or rebuild) trust. As the World Economic Forum kicks off again in Davos today, here are some of our thoughts on why and how we need to get Augmented Intelligence right; right now. And why we should never forget about the human aspect.
What is Augmented Intelligence?
As Nick Ismail writes in Information Age, “The concept of augmented intelligence is not to replace humans, but rather to capitalise on the combination of algorithms, machine learning, and data science to inform human decision-making abilities.”
Technology is becoming more and more advanced but cannot prosper on its own, the human brain and the experience that humans have is not easily taught, from removing bias to introducing emotional intelligence.
We should be using technology in the right way, to make our roles more efficient and enhancing the human capabilities.
For example, one of our recent projects involved risk-scoring clinical trials for a global Pharma company. We sucked in all sorts of structured and unstructured datasets — finance, HR, emails, the site visit reports (Word documents backed up as PDFs) and so on. Over 100 data sources. Then our data scientists coded machine learning algorithms that used hundreds of different features to predict where patient safety issues might occur. Some of these were obviously a big deal, like previous safety issues; some were only flagged by the machine, as humans had never thought to look at them, like the time lag between an event occurring and it being entered on the system of record. The resulting risk score allowed patient safety teams to be allocated to the highest risk sites, making them 4–5X more effective than being deployed at random as they had been before.
We didn’t set out to replace humans with machines; we set out to make those humans the best they could be — to give them super powers. That’s our goal in all our work.
How do we embrace Augmented Intelligence?
In order to successfully implement Augmented Intelligence, we see five building blocks that organisations need to put in place.
Data is the raw material and the strategic asset, that can help gain competitive advantage. (As a side note, in a world of non-physical asset businesses, we would argue data deserves a place on the balance sheet).
Most of our work involves thinking about non-obvious datasets. If we want to augment humans, we need data on those humans — what they do, who they meet, when they work, even in some cases how much they’ve slept recently.
In one recent piece of work, we helped a rail company reduce injuries by using all the data on their people to predict who was most likely to get injured in a given shift, allowing them to take pre-emptive measures. In some cases — like in elite sports — we have even worked with data on the inner workings of the human body! But often timesheet, calendar and email data is enough to get a good picture of what people are doing.
2. Machine Learning Analytics
Using new techniques and science, organisations are able to interpret and exploit their data in new ways and at a scale and pace traditional methods just can’t. Not only can analytics offer insight into cost savings and efficiencies, they can identify research and development opportunities.
And machine learning algorithms can deal with dimensions of complexity that the human brain just can’t. For example, in spotting fraudulent transactions (without those false alarms that drive us all mad), our algorithm looked at over 1,500 features, like the frequency of transactions in an account, their amount, whether the transaction had been made before, what time of day it was and so on. A human can hold maybe 10–20 of these features in their mind. A machine has no limit. Back to superpowers again — the highest risk transactions are then flagged to a human team to investigate.
Data and analytics offer a way to understand and identify patterns of behaviour and opportunities, whether that be to improve an existing service or product or introduce something totally new to the market. However, just because a pattern is identified it doesn’t guarantee it will be ethically or morally right. As reported in the MIT Technology Review, human judgement is needed by data scientists and engineers to identify and remove bias from data.
It’s imperative that domain expertise, critical thinking and curation come into play, to allow decisions (by humans) to be made constantly, supporting the machine intelligence. In 2017, we started to hear a lot more about ethical AI and this theme is only going to get bigger this year, as data scientists and engineers play the role of orchestrators of our future.
When we bring together design, engineering, and data science it becomes a very powerful mix, bringing together a perfect working group to create a strategy for innovation.
Once the three building blocks above are happening, someone’s gotta do somethin’ about it. Otherwise it’s just pretty lines on a page. We’re big fans of the “OODA Loop” to describe how to take action and keep iterating all the time.
The OODA loop is designed to help make better decisions, the four steps are:
By using this approach and repeating this loop, you can deal with data as it is changing all the time. It’s good enough for air pilots in dogfight situations; it’s good enough for us! It offers truth and allows you to interpret at scale and pace, before taking action.
As processes advance and learn, we’ll also see Reinforcement Learning start to appear, which use continuous learning to “collect tools that run and analyse massive numbers of simulations involving complex computation graphs.”
The final aspect of Augmented Intelligence is establishing the right culture. One that breeds leadership, decision-making and an organisational culture that is radically different to others you’ve seen before. People need to empowered and have more citizen data scientists, so that organisations are able to make data-driven decisions at all levels.
We’ll also start to see a changing workforce, thanks to Software 2.0, whereby the “programmers of tomorrow do not maintain complex software repositories, write intricate programs, or analyze their running times. They collect, clean, manipulate, label, analyze and visualize data that feeds neural networks.”
This year, we think we’ll see more businesses explore the machine + human + process approach as humans augment the technology that isn’t yet available, or ready to exist independently. We hope so, anyway!