Augmentation of Human Capacity

On Friday I did the opening keynote for the Mindshop Australia conference. The title was “Bringing the Future into your Advisory Practice”. The focus was on ways of creating more value for the clients of advisors in the network. After the session there was much discussion from participants on the nature of work and the sorts of jobs that they should encourage their children to be aiming for.

My response to those questions was to use examples to highlight principles rather than recommend specific jobs because jobs will change. I used the example of the health sector and new AI developments in my presentation as well as in the discussions afterwards. For example:

Self-taught artificial intelligence beats doctors at predicting heart attacks

stylised heart image from sciencemag

On the weekend I was then reading Stowe Boyd’s  10 work skills for the postnormal era and I was struck by the statement on “Freestyling” from Tyler Cohen:

“When humans team up with computers to play chess, the humans who do best are not necessarily the strongest players. They’re the ones who are modest, and who know when to listen to the computer. Often, what the human adds is knowledge of when the computer needs to look more deeply”

This married up with the response I was giving to participants at the conference. The use of AI systems to augment the capacities of humans  does not augment everyone equally. In the world of medical specialists it is a commonly held view among patients that they will put up with specialists with poor social skills or high prices because of the knowledge they hold (putting aside the issues of the professions restricting supply to keep prices high).

If that knowledge moves largely to the realm of artificial intelligence then this re-weights the value of the medical specialist. If the machine can do things the individual or team cannot possibly do by being able to access more knowledge and make more connections in that knowledge than is humanly possible then it changes the system. Knowledge becomes less important and skills such as the capacity to work with the AI, patient empathy and general social skills become more important.

Augmentation  of human cognitive capacities will do that across sectors and industries.

 

The Future Competitiveness of Corporate v Individual Medical and Veterinary Practice Models. Is AI the key?

Before Christmas I did some work on the future of veterinary surgeons and the education and regulatory changes that might have to occur to move with those changes. One of the things that occupied my mind with that work was the issue of artificial intelligence systems on the competitive position of veterinary practices. My belief is that artificial intelligence augmentation of medical/veterinary capabilities is coming quite quickly and a large scale corporate practice model has significant advantages in this space over the individual practice.

Increasingly we have seen a more large scale corporate model in the veterinary practice market in Australia and around the world. Examples in Australia include Green Cross for major city practices and Apiam Animal Health in rural practice.  This has followed similar trends in medical practices where you have organisations like Medical One and Tristar which started out in rural Australia but has expanded into the cities as well.

The basic business model and value proposition of the corporate model is:

  • Presenting a single branded product almost like McDonalds so a patient or client can be confident of going to the practice no  matter where they are.
  • Increasing purchasing power of all of the back end parts of the business from pathology services down to supplies of bandages, etc. This is more important in the veterinary model where practitioners are able to prescribe and supply vaccines and S4 drugs and make a profit from them.
  • Taking over the administrative and compliance parts of the business to allow practitioners to focus on patients. This can include the standardisation and delivery of training requirements, building management, OH&S requirements etc. It can also extend to the supply of consulting hours as a service package.
  • Networking and support for practitioners in smaller practices.
  • The spread of investment and risk over a larger geographic range and customer base.
  • A much more secure retirement/part time/business sale option for owners and practitioners. On a personal note my previous long time GP semi-retired by moving into a Medical One practice and then progressively handing over his clients to the other doctors in a very caring and professional manner.

These models have expanded quite significantly over the last decade which speaks to the financial viability of the model and while there has been the odd flare up and accusation of over-servicing in general the models seem to have worked. I originally started my working life as a general practice veterinarian and the corporatisation of veterinary practices is a hot topic at the reunions I have attended.

The competitive position of the individual practice has been built around personalised service and attempting to be portrayed as caring more for the animals they serve than the big corporate competitor. many of these practices are still doing quite well but the trend is towards the corporate practice.

Which brings us to future competitive positions. I believe that we are rapidly heading towards a future where augmentation of medical and veterinary skills via artificial intelligence is going to be  a ticket to play in the game. This is going to be a narrow focused intelligence rather than a general intelligence. In the medical space we are seeing story after story emerge of new models where AI systems are getting as good as human doctors or have an edge over human doctors:

This AI Can Diagnose a Rare Eye Condition as Well as a Human Doctor

eye image from motherboard vice artificial intelligence

Predicting non-small cell lung cancer prognosis by fully automated microscopic pathology image features

AI is nearly as good as humans in detecting breast cancer

Self-taught artificial intelligence beats doctors at predicting heart attacks

All of the examples above are based on variants of machine learning and one of the defining characteristic of machine learning artificial intelligence systems is they need large data sets. In the heart attack example above the artificial intelligence trained itself on almost 300,000 case records. As we currently understand artificial intelligence, that system is not transferable to cancer diagnosis, it remains a specilaised cardiac application. The non-small cancer system above was trained  on almost 3,000 images and subsequent patient follow up records.

There are two ways to get a large data set to train on in the medical/veterinary field. The first is to work on aggregated image and case records as in the examples above. That will certainly be a major part of the market. Large capital expenditures will be required to assemble the required images/case files and process them in a way that improves patient diagnosis and outcomes when used in conjunction with human doctors. So we will see services offered for specialised areas such as heart attack prediction, organ by organ cancer diagnosis, etc. As always the areas that have the largest and most affluent markets will be the services that are first offered.

However we are also moving in to a world where artificial intelligence systems can be harnessed by smaller players at much lower cost. Take this example of How a Japanese cucumber farmer is using deep learning and TensorFlow:

cucumber-farmer-14 tensor flow artificial intelligence

The son of a Japanese cucumber farmer (who admittedly had some very good tech skills) built an artificial intelligence based cucumber sorting system for his parents. The problem was training the system required a lot of images and the process took a lot of computing power. More and more in the future you will be able to plug into a cloud based machine learning platform that will enable you to harness much more computing power that is specialised to do this sort of a job for you.

So a large network of medical or veterinary practices could offer services that are not offered in the general market by collecting all the data on their patients across all their practices and using machine learning platforms to train a system that is specific to their patient database.

This will be impossible for the individual practice to match. Is this the killer application for the corporate model to almost completely squeeze out the individual practice?

If so I will be watching for models emerging in the veterinary space because the regulatory hurdles and insurance requirements are much lower. If they are successful then I see those sorts of models flowing into medical practices.