I listened to a fascinating podcast on using AI in human genomics today where Steve Hsu and Azeem Azhar discuss the intersection of genetic engineering and machine learning. Hsu using polygenic risk scoring, which determines disease predisposition based on analysis from a number of genes. I would commend it to you.
He highlighted that the human genome has 3 billion base pairs but only 1 million base pairs significantly influence the differences that we see in people. This may be complex traits, such as hair colour, or height through to the risks of specific illnesses.
Most complicated predictors, such as height prediction use 10s of thousands of gene pairs. They can predict height +/- 1” from a DNA sample with 60% confidence.
Type 2 diabetes uses date from thousands of pairs of genes. Breast cancer uses 500-1000. If we know that someone has a 50% chance of getting breast cancer we start to test for it much earlier than we normally would, we become alert to its potential emergence. Depending on our tolerance to risk a patient may also decide to have a mastectomy, thereby practically eliminate the risk.
So what does this mean for project delivery? I hear time and again that every project is different so AI will never work. But if we can do this with humans, then surely we should be able to derive insights out of project data?
The polygenic predictions have only been possible because of the availability of data, sourced from a UK database. If we are to do the same in project management the starting point for developing an AI capability is to create the equivalent of a DNA database. A data trust enables organisations to securely pool data and provides the volume of data for researchers and innovators to derive insights. I am delighted to be working with the Oil and Gas Authority and Oil and Gas Technology Centre to develop a data trust for the oil and gas sector. We’ll also be replicating this into construction in the coming months.
The types of data range from risks through to environmental factors. There is also an organisational component, i.e. some organisations are more predisposed to variance against functional activity or components of a work breakdown structure than others.
As we build out the dataset we begin to understand a project’s predisposition to variance. This isn’t a static assessment; the assessment evolves as the project’s circumstances develop. Risks can be retired when we understand that the window of influence has passed.
I also turn to examples such as Crossrail where heroic project management and conditioning to bias, was at the root cause of delay. By pooling data we begin to understand which gene pairs have an influence of project delivery. We can then become increasingly sensitised to looking out for them.
So in summary, if we can achieve all of these insights from human DNA, surely if must be possible to develop increased predictions on project variance. I agree that its not deterministic, but we will be able to drill down into the statistical data and micro narrative to enable us to deliver projects with greater confidence in the future. Public confidence in our professional ability to deliver projects within the approved envelope is under immense pressure. We must try something different.
We are entering a new era and there is so much that we can leverage from other sectors, particularly from healthcare.