Raging against the (learning) machine
The dream team shaping our lives
Last week I went with a few friends to watch a screening of 2001: A space odyssey in a 70mm projection at the Prince Charles cinema in London. This is a film that I have watched many times, and as with most great movies, new things come to the surface with every new viewing. I was particularly captivated by the following scene:
For those of you not as fanatical about the movie, a bit of context. Five astronauts are sent on a mission to Jupiter to follow signals discovered on the moon in 2001 emitting from a black monolith towards Jupiter’s moons. The spaceship is completely controlled by HAL 9000, an intelligent computer which has never had a fault. In the scene depicted above, the astronauts lock themselves in a capsule to discuss HAL’s behaviour (which at this point has become rogue) and their options for disconnecting it. They believe they are safe as HAL cannot hear them…
As I was leaving the cinema, I couldn’t stop thinking about how well a movie in 1968 represents the fear, uncertainty and doubt we feel about Artificial Intelligence (AI) and Machine Learning fifty years later. With depressing news every day about machines taking over people’s jobs, it’s no surprise we feel like staff at the supermarket self-checkout assisting shoppers with the automated machines as we accelerate our own path to unemployment. Feels like the design of a cruel Greek deity but it’s just disruptive innovation at work. The same it has been through history.
Artificial Intelligence has had a lot of bad press so perhaps starting with a definition is necessary.
AI is the designing and building of intelligent agents that receive precepts from the environment and take actions that affect that environment.
This definition from Russell and Norvig is considered the gold standard as it covers a large spectrum of activity ranging from computer vision, speech processing, natural language understanding all the way to robotics. Based on this definition, I don’t see the inevitable link with machines taking over control of the world but as technology continues to improve at its current pace, more activities which used to be done by humans are taken over by machines. Let’s use a recent example:
This year in Wimbledon, all the highlights packages were automatically produced using IBM AI technology to analyse the video, determine the best plays, edit the content and finalise it ready for sharing online or providing it to media organisations. The highlights process was reduced from 3-4 hours to a few minutes allowing the Wimbledon media team to focus on higher-value engagement activities.
The fact that Wimbledon media staff can focus on activities which generate more engagement is key in this example as it illustrates the subtle difference between world domination by evil technology and what I think is really happening. Let’s look at another example which makes this even more evident:
In August this year, Deepmind in collaboration with Moorfields Eye Hospital developed an AI-enhanced process to detect eye disease which ‘can correctly recommend how patients should be referred for treatment for over 50 sight-threatening eye diseases as accurately as world-leading expert doctors.’ The challenge being tackled here is related to the amount of complex 3-D images doctors need to review (1,000 / day in Moorfields Eye Hospital) and the level of expertise required from doctors to interpret these images. Since delays in treatment can cause patients to lose their sight, any improvement in the speed of eye disease detection can have a major impact.
The two examples above can be interpreted as proof that machines will be doing the job of doctors and TV producers, but we need to look at this differently: AI augmenting human intelligence. This view challenges the pessimism and fatalism with which AI is being looked at today and a few thinkers in this space are not using the term AI anymore but flipping it to IA which stands for Intelligence Augmentation. Under this interpretation, the potential of the technology becomes clearer and practical solutions become more realistic and achievable as machines don’t need to compete with humans but work together. Every time an article talks about the demise of a current job because of AI, it is important to focus on which part of the job is being augmented by technology and what opportunity this creates.
To truly appreciate the power of Intelligence Augmentation, it is worth talking about Fan Hui, a 3-time European Go champion.
Go is an ancient board game of strategy that has frustrated computer scientists and researchers for years. In Go there are more moves possible than there are atoms in the universe, so it has always been considered the greatest test for artificial intelligence. Fan Hui was selected by the London-based AI research company DeepMind (now owned by Google) to train the computer system which ultimately went on to beat the world champion Lee Sedol over 5 games. The training was done by Fan Hui playing against the computer and allowing it to learn and evolve its own strategies. After a few months, the machine started beating Fan Hui regularly and the team developing the programme decided that it would be more effective for the system to play against itself rather than continue playing Fan Hui. This should be utterly depressing but what happened when Fan Hui went back to play against other humans is absolutely amazing. Thanks to the interaction with the AI system his skills had improved significantly rocketing him up the rankings from 600 to 300 in three months.
We’ve applied some of this thinking to a new project where HouseMark is partnering with Lewisham Homes and Field Dynamics to look at how these modern data techniques can be applied to the specific problems in Housing and more specifically, to those of asset management. We have used 7 years of repairs data from Lewisham Homes to try and predict the repair costs for properties in the portfolio. This does not aim to remove the need for experienced asset managers, but these modern techniques can help Housing organisations manage their major work programmes based on predictive spend and allow them to regularly reallocate which properties should be in or out of the programme. This is another tool which augments the intelligence of the organisation and gives a new perspective to base decisions on. It is early days, but the initial results are very encouraging. We are looking to get a group of organisations to work together in a collaborative project to build this system as a product which can be used across the sector. We are particularly interested in working with asset management practitioners to understand how these techniques will augment their skills and potentially transform their jobs going forward.
If artificial intelligence viewed as intelligence augmentation can transform the work of doctors, free up TV producers and exponentially improve the skills of a Go player. What can it do for those of us who work in Housing? The answers will not come from raging against the (learning) machine.
This article was published in the first issue of the DIN Bulletin