After a short description of my current state of mind with some self-pitying sugarcoat, I briefly describe three main fields I could address in my PhD.
I was looking for the key for years
But the door was always open.âAravind Adiga, The White Tiger1
Here I am, frantically looking for a key while many doors are open. I think this is one of the main burn out causes during a PhD. There is a problem in your mind that follows you all day long and somewhat at night too. The psychological burden is quite heavy. Being in the unknown is a very paralysing and energy-consuming state of mind. You search for a key to unlock an answer for a question you barley understand. But the answers are already there, waiting for you to look at them. Maybe there is no need for a key. Maybe there is no key.
The Prussian general and military theorist Carl von Clausewitz (1780 â 1831) wrote On War, an essay on the psychological, political and strategic principles in a conflict (if I remember correctly). At some point, he describes how, in war times, individuals are surrounded by a âfogâ. Everything is unclear, unpredictable and confusing. According to him, a genius is one who advances nonetheless amidst the fog of uncertainty.
What is this fog? To give some flesh to this essay, here are some of the current deliberations in my head. The initial topic is: âwhat does AI brings to the management of care-path customisation?â Yet, I donât know what to do exactly. Many other subquestions arise instead, which are not totally related, but linked to this problem nonetheless. To put some order: a very intuitive way to address AI in healthcare it to look at three main âstepsâ:
Data collection
Processing
Implementing
/Data collection
A lot of work is needed to collect and manage data, be it in healthcare or other contexts. Many hospitals and other actors are collecting data related to healthcare in order to allow third parties to exploit it. Their work aims at creating the infrastructure. The more this job is done, the more easy it becomes for data processors to use and implement solutions. Theoretically.
Hence, they try to gather as much data as possible from the existing sources (payment records, lab analysis results, satisfaction questionnaires, etc.) and eventually find new sources of data. Itâs a loooot of work.
There are many things to understand: their strategy or vision, their goals, the type of data they are collecting (and thus the data that isnât collected), the governance, their business model (i.e. how they plan to produce value), their role as gatekeepers, etc. As they make the main infrastructure to develop data science applications, they have a key role.
/Processing
Although data represents the âbasisâ for AI tools to work2, in the end, it is processed in order to create value to it. To pastiche economicsâ style:
f(data) = knowledge
meaning there is a function that takes data and outputs knowledge. This knowledge is a prediction of a value or category (spam/not spam) and can lead to a decision (put it on the spam folder/leave it in the mailbox). But how is the function created?
Many questions can be studied besides the technical aspects of this function. For instance: Who are the people doing it? Startups, firms, government, etc.? Understand their vision they have for their applications: what needs do they see and what problems do they want to tackle? How do they work with the data? Do they have biais? What trade-off do they make? How do they evaluate if it works well? Are there ethic committees? Is the algorithm âfairâ and how do they define it?
My initial topic is about the âmanagement of care path customisationâ3. Thus, many actual applications might not âfitâ into it. But analysing peripheral applications can also be worth it.
/Implementing
Even if a solution exists, they are not necessarily used. As mentioned in âIndian Summer for AIâ, many promisses of data processing revolve around the new knowledge it will unlock. However, there is already available knowledge that isnât put into practice. Thus, even if more knowledge is created, if the pipeline for organisational change is jammed, no real output will be made.
Actually, in innovation management studies, the âdiffusionâ (i.e. how an innovation is introduced to society) is a large body of literature. As innovation breaks the statu quo it is badly perceived at first.
Small note: funny enough, entrepreneur or innovators are figures that were very unpopular in Wester culture until recently. Societies used to be more conservative and groups more endogenous. On the other hand, entrepreneurs/innovators usually âtransferâ something that exists instead of creating it from scratch. For instance: you use IT tools build in a sector to another one. Mostly, innovations used to come from the military like planes. As the entrepreneur is somebody outside of the initial group or that brings foreign knowledge, they are perceived as suspicious. The actors in power can also feel threatened by change if they are not controlling it. However, the high value innovations can create nowadays gives to entrepreneurs a better image. But if you look into the details, it is still a figure with prejudice.
Hence, many things can lead to kill a product or a service. Lack of adoption can be due to a bad time-to-market when users are not ready to change. Other times, as buyer and user are not the same, it is difficult to convince the former even if the latter agrees on adopting it. For instance, in France, care is mostly paid by the government and by complementary health insurances which in turn decide what they refund. Many other high barriers exist for innovations in healthcare, like lobbies impeding disruption or red tape.
AI solutions will challenge many things and they will need to push some boundaries for them to be implemented: new ways to capture value, to prove their benefits over the risks, resistance to change form patients or organisation, etc.
Back to square 1
Well, this was a quick glance on many âdoorsâ around me. Itâs not exhaustive but hopefully you could have a peak to the diversity offered. Yet, Iâm still looking for a key. Maybe part of it comes from a psychological bias to inactivity: the human mind easily finds excuses to avoid action.
The excuses I find are of two types: first I donât know where to start. The choice is very wide. Many options seem attractive but I want to find the one with the highest potential, as I wonât be able to tackle them all. Yet, when I opt for a direction and I push my reckonings, I quickly find small blocking points. I start looking at another option. Find flaws in it. Look at another one. Repeat. Repeat again. And Iâm lost.
The other excuse: I have to find something that relates to the initial topic of my PhD. There might be many initiatives in AI x Health but I still donât see the customisation part emerge specifically. Another approach is to look at what other sectors do in AI x Customisation (voice analysis in call centres for instance).
In one of my previous posts (about methodology part 1/many), I claimed it was legit to try to have a theoretical understanding as a first approach. Now, I feel itâs too limiting (at the stage where I am). I need to confront myself with the field, with what happens, with how people think and behave.
There is a time for thinking. Now, itâs time for action!
A very interesting book about the history of a poor Indian who becomes an entrepreneur despite the many hurdles of his county and situation. I read it six years ago and I still remember having a deep chock. Some Indians I disused said it was exaggerated and so on, but I still thing the book is worth it since it tells a more fundamental story than the situation in India.
âFew shots learningâ or âzero shots learningâ techniques use knowledge which is distilled data.
Personal note: every time Iâm having a headache when I try to figure out what it means as it looks like a concept inside a concept.
JMW Turner is my all time favourite painter....
and see how a sky free of tempests and fogs is boring: https://www.tate.org.uk/art/artworks/turner-a-clear-sky-above-a-landscape-d17180
In viticulture, the toughest growing conditions challenge the vines to grow harder, the struggle is necessary to yield beautiful fruits. What makes Sauternes a liquid gold is that it strives to flourish in the afternoons, after the cool morning fog. I hope your fogs will lead you to a beautiful place and that you will find your direction and hold onto it dearly Mr Scientist