AI in Cancer care and managing the great expectations around it

I woke up on Saturday morning and read this WSJ article IBM has a Watson Dilemma . As always when such articles get published, this was followed by a lot of criticism on twitter, linkedin etc – and I read most of them. And today morning, I saw this article on IBM blog site from Dr John Kelly titled Watson Health : Setting the record state .

I am very hesitant about expressing my personal opinion

Especially when my employer is the one being criticized. I am not an impartial party here at all – I am an executive at IBM ( Not a very senior one by any stretch – there are a couple of levels between me and the CEO) , I hold IBM stock , I am not a company spokes person, and till recently I managed a business of which Watson and Watson Health consulting services were a part of. Also, Almost exactly a year ago, I wrote about my opinions on IBM Watson after an analyst wrote an article that I thought I should weigh in, in a personal capacity.

On the other hand, AI is a topic I have great interest and some expertise in (Again, I am not a hands on ML developer or anything like that today – though that might potentially be one option in future given my passion).

I lost my dear Aunt Geetha to cancer a few years ago – she was a second mom to me. I was at the hospital with her when she fought the deadly disease for her last few days (I was on vacation in Trivandrum) , and I spoke with several folks at the hospital whose dear ones were going through the same battle. So anything that helps fight cancer is a topic I have a deep interest in, and one where I will happily donate my time and money.

So for what little it is worth, here is my take and I am just going to address two specific issues

Is Marketing Hype the big culprit ?

I am not a marketer by trade, though I appreciate high quality professional marketing. When I am in a sales role, I prefer a soft selling approach – and that might be because I am an engineer first and foremost. So, I totally get it when IBM Marketing gets accused of going overboard by someone on social media. I have also been in this industry long enough to know that without massive awareness created by marketing, no young technology gets the air cover it needs to mature. I personally know of no client who has made an enterprise purchase only because they saw awesome vendor marketing. Marketing opens doors no doubt – but clients subject their purchasing decisions to an array or dimensions ( proof of concepts, risk management, analyst reports , references etc)  before someone signs a check.

What I readily agree is that marketing does contribute to setting big expectations for new technologies. And big expectations are good – as long as everyone gets the nuances that go with it. When it comes to finding good solutions for deadly diseases cancer, I doubt it helps to not have bold goals. I always encourage folks to ask good questions – and proceed with eyes open.

Is it bad that Watson agrees with doctors most of the time ?

To state the obvious, it would be terrible if Watson and doctors disagreed all the time. But is it bad if they agreed most of the time ? The WSJ article implies that since human doctors agree with Watson most of the time, they stop using it or at least limit its use.

There are commercial use cases that follow the same pattern as cancer diagnostics. For example, Watson can ingest training manuals of several machines and can have a Q&A with a mechanic or a customer who is faced with a live problem. An experienced mechanic usually agrees with Watson most of the time, and probably does not see much value. But think about the less experienced mechanic, or a customer who is not technical. The solution is of high value to them. The ideal situation is that the experienced mechanic continues to train Watson (via agreeing and correcting when wrong) and Watson helps several lesser experienced mechanics and customers from what it has learned. That is the incentive to have the experienced mechanic continue to use Watson.

Sitting outside the room where my Aunt was struggling with her fight, one thing was abundantly clear to me. In USA, we have several oncologists and specialist hospitals that are the envy of the world. That is not universally true. Even in the hospital I was at with my aunt in India, there were plenty of oncologists – but nowhere close the number that is needed to cover the sheer number of patients. They have very little time to keep up with the latest in cancer care – or to even spend enough time with one of their patients. They deal with patients who flock there (and several of them thankfully don’t have cancer and was sent there because of poor diagnosis where they initially went to ), and even if that process can be streamlined – they can save more lives.

Now think of all the hospitals in a country with a billion people – and several of the people not diagnosed or treated just because of poor access to specialists !

I also vividly remember the line in front of the radiologist’s office there in India – one very tired lady trying her best to read images and make notes while highly stressed out patients and their relatives started shouting around her. I felt terrible for the doctor and the people around her.  With advances in computer vision, this scenario can be improved exponentially.

It it bad that Watson cannot figure out great solutions for rare cases ?

Ideally, I would love for AI to help us solve cases where humans have very limited options. I don’t think tech will solve this in near term. But that is not to say there is nothing tech can do today. It still can find useful information more quickly for a doctor than they can find manually, and WSJ article does talk about that.

There are several obstacles to getting AI to work as we need it to – and getting data organized for Ai to learn from is one big one. Even in areas where we have been at it for decades – like loading legacy data into a new ERP system, it takes a lot of effort . You can only imagine the additional complexity included in getting data in the form that an AI model needs to learn from. It is not an insurmountable problem, and newer approaches keep coming up and at some point it will become mainstream, and easy to estimate the effort.

That is where I see the true potential of AI , including Watson ! It helps take expertise from people and institutions that have it and move it places where no such thing exists. And if we can save one more life, or reduce the pain for one more patient, or reduce the grief of one more family who will lose a dear one- I think it is totally worthwhile. I sure hope we don’t give up on this journey !