An essay inspired by “Power and Prediction” by A. Agrawal, J. Gans and A. Goldfarb.
By Joel Lim, Lead Creative Technologist.
RULES RULE
A lack of information leads to a creation and reliance of rules to mitigate the negative effects of uncertainty.
Here are some examples of rules:
- We don’t know how often a gym member will use the facilities, so the rule is we offer a best-guess set of fixed membership options.
- We can’t be sure if someone has Covid or not, so the rule is we don’t socialise, don’t go to work, don’t take public transport, etc.
- We don’t have data on a child’s interests or level of competency, so the rule is children are admitted to schools in batches by year of birth.
AI has the potential to replace rules by using data to make predictions that lead to informed decisions. We will come back to this later.
RULES ARE GLUED TOGETHER IN A SYSTEM
In a system, rules are “glued” to other rules as well as connected to upstream and downstream interdependencies and procedures. Generally speaking, the bigger the system, the more rules there are.
This makes systemic change difficult, even if change will lead to a valuable outcome.
Replacing or swapping a rule with another one, or eliminating a rule altogether, can have a dramatic effect on whatever’s connected to it. And this effect can be perceived as negative.
This is because “Institutions will try to preserve the problem to which they are the solution,” observes tech writer and thinker, Clay Shirky. Removing a problem, for our context, read: rule, will equal removing a solution, and organisations with silos that specialise in those solutions, will struggle with that.
Change is difficult, but not impossible. To embrace change, an organisation must first be sold on the value they stand to gain from it.
REPLACING RULES WITH DECISIONS
Instead of having a rule, you can have a decision based on a prediction. We call this a point solution.
Gym System
- [Rule] We don’t know how often a gym member will use the facilities, so the rule is we offer a best-guess set of fixed membership options.
- [Point Solution] AI could predict how often, when and how long a person would use a gym. The fitness centre, having this predictive capability, could offer bespoke memberships and do away with limited, cookie-cutter options.
- [Value] This will mean not over-paying to use a gym.
Pandemic Response System
- [Rule] We can’t be sure if someone has Covid or not, so the rule is we don’t socialise, don’t go to work, don’t take public transport, etc.
- [Point Solution] If we could provide a decision engine with real-time information on the likelihood of an employee contracting a virus, it could make daily decisions on whether an employee can head to the office or not.
- [Value] We won’t need to shut down the economy whenever a pandemic strikes.
Education System
- [Rule] We don’t have data on a child’s interests or level of competency, so the rule is children are admitted to schools in batches by year of birth.
- [Point Solution] If a student could be evaluated on subjects they show an interest in and an aptitude for, we could let AI align teachers and courses based on this information.
- [Value] Students will be able to learn what they love and progress at a pace that suits them.
THE GLUE, AGAIN!
Replacing one or a few rules with point solutions does not mean a system is immediately functionally better than before. A point solution may solve an issue but it’s still linked or glued to other parts in a system.
A connected process may not be able to cope with the efficiency of the solution — there might not be sufficient resources in place to handle what should happen next.
For instance, a gym with a point solution for bespoke memberships might not have enough personal trainers or exercise machines to accommodate the surge of sign-ups.
If AI decided who should and shouldn’t come in to work during a pandemic, won’t this have consequences for workers on a daily rate? What new employment contracts will need to be written?
Schools with a fluid, student-centred education model will need to prepare teachers for a new way of working. They won’t have teach in classes anymore. They will also need to determine how exams are created and when they will take place.
You get the picture: A legacy system can bring a well-oiled decision engine to a grinding halt.
REDESIGNING THE SYSTEM TO ACCOMMODATE DECISIONS & HOW TO PUT IT ALL TOGETHER
For a point solution to work, we must zoom out and see all relevant connected components. And we will need to create new processes to unlock and realise the power of one or more point solutions.
To bring about the change we want at a quantity and rate we can cope with, especially during the early stages of deployment, this is what we will need to do:
1. Locate the rules we want to potentially replace with a point solution
What are the most pressing objectives of the organisation? Where are the gaps in information or data that enforce rules in the system? How might decision engines replace these rules?
2. Select point solutions to be developed and prepare connected components
A point solution must not be developed in isolation. All connected components and interdependencies — data, governance, legal, procedures, resourcing, etc — must be looked into and redesigned to work with the point solution. Expect to define and activate components or procedures you currently do not have in place.
3. Progressively develop, deploy and test
We recommend building point solutions in stages, as well as deploying and testing them in the system in increments. This is a methodology widely known as agile development. By progressively releasing the features and power of the decision engines, we give the system and organisation enough time to adjust to and handle the new processes that flow out of each point solution.
A SYSTEMS APPROACH TO AI
Moving from a system of embedded rules to one with AI-enabled decision making will help organisations achieve efficiencies never thought possible. It will let businesses launch novel products and services and offer intuitive customer care.
When we take a holistic approach to harnessing the capabilities of AI, AI will not merely be a very clever or useful “add-on”. It will be the electricity that powers the entire operation to offer more value to users and customers.
