Tuesday, August 22, 2017

The Trouble with AI: No Single Wringable Neck

Pretty exciting news in that as of June 2017 I am now certified in 3 things: Open Water Scuba Diving, Internal Medicine and Scrum Agile Product Ownership.  (The first was the most fun.)

Anyhow, one of the concepts brought up in Scrum Agile is the idea that the Product owner is the single wringable neck.  The one person on the team that the customer/management/anybody can blame when everything goes wrong.  Sounds like a fun job, right?

So why do you need this? Because, something always goes wrong. (and thankfully for that!)

What exactly does this have to do with AI?  As a doctor in technology I am constantly bombarded with really exciting applications of artificial intelligence that are going to make me irrelevant.  These are usually put forward by very well intentioned, highly intelligent and well-funded individuals that haven't the faintest idea of what a primary doctor actually does (shame on us).  In general I have adapted a wait and see attitude, although there have been some promising breakthroughs in using AI to accomplish what needs machines the most-- the dreaded paperwork of medicine.

Even outside of healthcare we see robots that can climb stairs, robots that can fold clothing, and most obviously, self-driving cars.  And we hear the same story-- once these get good enough, they are going to take over.

And in some sense, I agree.  However, the individuals working on these technologies are product people: scientists, engineers, product managers, designers, etc. They are all motivated to make the best/coolest thing possible.  And that is exactly what they are doing.

But this approach alone will fail.

What exactly is the quality threshold where we will trust AI/robots/computer driven cars?  We will likely see long stretches of accident free days like we never have before.  But perfection today is no guarantee of perfection tomorrow.  Somewhere, at some point during this utopia of perfect driving, one of these cars will kill someone.  Or it won't, but we won't be able to shake the idea that it could happen.  They already have far fewer accidents than humans.  But they lack something humans have. A wringable neck.

If you get hit by another driver, we have an elaborate series of mechanisms to punish that person and make you whole (ish) again.  We have someone to blame and a way to blame them.

How do we hold the robots accountable?

I propose that this is the hurdle we must overcome in order to expand robots, AI and other autonomous devices in our lives.  It is a legal one, a regulatory one, a cultural one, not a technical one.  No amount of perfection will rid us of the human desire for retribution when harmed.  So to all of the makers out there-- please keep improving the quality, but know that you must offer a neck a some point if you want others to adopt your technology.


Friday, July 7, 2017

What I've been reading lately

Following the advice of Charlie Munger, I've been broadening my reading list recently to include additional disciplines.  Specifically, leadership, in industries entirely outside of medicine: submarining, manufacturing, DevOps and software testing.  Here is a brief list of what I have been reading/am currently reading.  What I can tell you is that mixing in lessons from disciplines has rapidly accelerated my thinking in my own field-- Primary Care innovation.  So what in doubt, go read something else!

Also, if you don't have a library card-- get one!

  1. Turn the Ship Around, by L. David Marquet
  2. The Goal: A Process of Ongoing Improvement, by Eliyahu Goldratt
  3. The Phoenix Project, by Gene Kim et al
  4. Sense & Respond, by Gothelf & Seiden
  5. Scrum: The Art of Doing Twice the Work in Half the Time, by Sutherland and Sutherland
And of course, some of the more useful books over the past few years:
  1. Antifragile, by Nassim Nicholas Taleb 
  2. Tribal Leadership, by Logan et al
  3. Leaderhip & Self-Deception, by The Arbinger Institute (Institutes can write now?)
  4. Drive, by Daniel Pink
  5. Give & Take, by Adam Grant
  6. Rework, by Fried & DHH
  7. How Stella Saved the Farm by Vijay Govindarajan and Chris Trimble
Enjoy!  Happy to start reviewing any of these if there is interest.  Until then, read some amazingly in depth book reviews by John Norman over at 7fff

Tuesday, April 4, 2017

Review of Systems Podcast

Excited to share a recent appearance on the Review of Systems Podcast, talking about my work at Iora Health.  Created by Audrey Provenzano, MD, Review of Systems is a great collection of interviews with folks pushing the boundaries of healthcare innovation.  Check it out!

Tuesday, March 7, 2017

On the Reduction of Meetings

A common complaint in many modern work environments is the sheer number of “meetings.”  Collections of individuals that inevitably interrupt creative “maker” work.  In general, there are good faith attempts have at the leadership and team level to indiscriminately “reduce” the number of meetings as well as “improve meeting hygiene.”  On reflection, the issue may not be so much a meeting issue, as it is a communication problem within the organization.  That is to say, meetings have a tendency to be one-sided--the person calling the meeting needs something from the attendees--and therefore feel interruptive and unimportant by some subset of the attendees.  By exploring a taxonomy of internal communication needs, we may be able reduce meetings through improved communication options rather than gross reduction and/or shaving minutes through hygiene.  External meetings are not yet included.

Taxonomy of Meetings:
  1. Approval
  2. Decision-making
  3. Problem-solving/Advice
  4. Working session
  5. Collaborative Creation session
  6. Informational/status update

  • Who calls: Decision-maker with responsibility for a decision but not authority
  • Who attends: One or more approving managers
  • Agenda: Clearly disclaim that this is an approval meeting, send the decision up front if possible to avoid the meeting.
  • Length: 15 minutes
  • Benefits accrue to: Generally no one, meetings of this nature are often avoidable and reflect poor decision making process.
  • Opportunities to reduce: Delegation of authority by role (staffing plans, budgeting, etc.)  
The Approval meeting exists whenever we have set up managers with responsibility to solve a problem but without the authority to do so.  These meetings are often the result of insufficient up-front planning (including budgeting) that would allow pre-delegation of authority to a trusted manager.  In general, these meetings should be reserved for novel situations and should be generally reduced through appropriate work planning and clear decision-making authority granting.

  • Who calls: Decision-maker with responsibility and partial authority
  • Who attends: Peer decision-makers, usually of different functional areas with the remaining authority
  • Agenda: Inform the participants ahead of time that a decision is to be made by the end of the meeting.  Send point of decision and any initial options ahead of time, but do not present initial options as constrained choices.
  • Length: 30 minutes
  • Benefits accrue to: All parties as long as this decision-making pattern exists
  • Opportunities to reduce: Not clearly a positive.
The Decision-making meeting is different from the approval meeting in that peers with partial authority for decision-making come together to discuss multiple options. One member is not necessarily trying to get buy-in from others, but rather using the diverse functional experience of peers to choose together between a series of options.  As long as our business remains complicated and novel, these meetings will persist.  Reducing these meetings should not necessarily be a goal as we believe unilateral decision-making is less likely to produce superior decisions.

  • Who calls: Decision-maker with responsibility and authority, but insufficient information
  • Who attends: Functional experts from different disciplines
  • Agenda: Inform the participants ahead of time that a you are seeking input on a decision that you will make. Try and determine if you need factual input or judgement input. Send point of decision and any initial questions ahead of time
  • Length: 30-90 minutes depending on topic.  
  • Benefits accrue to: Decision-maker & judgement giver (to help organize their own functions)
  • Opportunities to reduce: Functional experts can maintain knowledge bases to reduce the factual input-style meetings.  
Problem-solving/Advice meetings are similar to Decision-making except the attendees are there to advise rather than decide.  The meeting caller can accept or reject any input, and should make this distinction clear so as to not confuse or offend the attendees whose input is not incorporated in a final decision.  Fact-seeking type meetings may be avoidable if functional areas are able to invest in transparent knowledge bases.  This leaves the bulk of these meetings to be judgement related which are incredibly beneficial for both the caller and the attendees.  The caller gets advice to make a better decision.  The attendees have an opportunity to verbalize and potentially re-organize their thoughts about their area of expertise, which can often lead to breakthroughs elsewhere.  

Working Session
  • Who calls: Decision-maker with responsibility and authority, but insufficient skill or labor
  • Who attends: functional experts, skill experts, labor from own team
  • Agenda: Inform the participants ahead of time that a you are trying to complete a fairly well-described task together, which will likely be complete at the end of the meeting. Send a great description of the problem at hand and initial broadly sketched solutions and/or previous solutions.  Time and task management is crucial.
  • Length: 60-240 minutes (with breaks)
  • Benefits accrue to: Company & Customers
  • Opportunities to reduce: Build functions for commonly occurring tasks/problems.
The working session is when you have to get things done, often early in a small company's life cycle.  A great example is board-book preparation, or customer mailers.  Any time that a project is important or complex enough that more labor and/or special skill are required in parallel, call a Working session to get things done. Regular working sessions to solve the same problems repeatedly may be call for creation of a function with new positions to handle new regular work.

Collaborative Creation Session
  • Who calls: Decision-maker with responsibility and authority, but insufficient skill or labor
  • Who attends: functional experts, skill experts, labor from own team
  • Agenda: Inform the participants ahead of time that a you are trying to complete a complex, poorly-described task together, which will likely NOT be complete at the end of the meeting. Send a great description of the problem at hand and initial broadly sketched solutions and/or previous solutions.
  • Length: 90-240 minutes (with breaks)
  • Benefits accrue to: Company & Customers
  • Opportunities to reduce: Not a goal.
The Collaborative Creation session is used when multiple disciplines are required to solve a new, complex problem. Our Cross-functional teams are an example. Unlike the working session, your goal is not to “get things done” but rather build something new and different.  This is generally hard, messy, takes more than one meeting, and is exactly why many of us come to work every day.  We have plenty of internal skill, specifically on the Operations Development, Talent and Culture and Product teams at how to facilitate creative meetings so I will defer to them. Managing these meetings over time is often a necessary skill as the work will spill from meeting to meeting and participants will not always be available.

Informational/status update
  • Who calls: Functional area/Team leader, Someone who wants to know something
  • Who attends: functional experts from one or many teams
  • Agenda: Clear agenda of topics to be covered, either ad hoc or regular agenda. Inform the participants ahead of time of what is to be covered so they can prepare their updates.  Make it clear that the goal here is to inform and not necessarily advise or decide.  Allow for Q&A (to distinguish this meeting from what could have been an e-mail). Time management key to cover all areas.  Will likely spawn additional other types of meetings.
  • Length: 10-60 minutes (with breaks)
  • Benefits accrue to: The need-to-be-informed. This is generally a cost for the informers unless they learn as well.
  • Opportunities to reduce:
    • Any unilateral information dump of non-contentious nature could be converted to a written communication, e-mail, knowledge base, snippets, tweets, etc.  
    • Smaller sized meetings also more likely to be universally beneficial.
    • Ask yourself if there is another way to learn the information that takes less of your participants time
    • Find ways to bring additional content that is useful to the attendees 
The status meeting.  On the positive side, these meetings allow very sensitive or nuanced information to be discussed, access to senior leadership decision-making thought process rather than facts.  They help keep everyone synchronized and working together and on the right things.  Done well, these meetings are crucial to a healthy organization.  
Perhaps the greatest paradox is that we do not have any time to do work, read notes/updates, because we are always in informational or status meetings.  These are the biggest potential offenders for wasted, interruptive time.  Because they tend to be large, they are often scheduled at a time that is optimal for no one.  Also, it can be hard to make these meetings sufficiently beneficial to everyone as the benefits often accrue to the uninformed, who call the meetings, and then continue to call them causing a downward spiral.  These meetings tend to be regularly occurring without clear need or agenda.

Thursday, February 23, 2017

HIMSS 2017: Buzzword Roundup

Greetings all!

Glad to be back from a dizzying couple of days at HIMSS 2017, the annual see and be seen conference in the Health tech world, where prestige is measured in convention floor square footage and carpet depth (at least one vendor area felt awfully reminiscent of a bouncy house).  Having not been in about 10 years, HIMSS has become a rather dizzying affair.

This year the buzzwords seems to coalesce into 4 related concepts, neatly placed on nearly everyone's marketing materials:

  1. Value-based care
  2. Population Health
  3. Big data
  4. Machine learning/AI
One gentleman I met on the floor suggested that all of the presenting companies could have agreed on one booth with one pattern of signage given the convergence.

Anyhow, if you look deeply into those four buzzwords (buzz phrases?), it would seem that the direction of clinical care is artificial intelligence supporting groups paid to take care of populations.  Individual clinicians and patients were nowhere to be found. Not entirely surprising for a tech conference, but as another colleague pointed out-- this isn't really tech, these are tech enabled services.

Hold that thought in one hand for a moment while we try and square it with the rise of personalized, relationship-based, individualized medicine as all the rage elsewhere in the healthcare world.  

So then, are the two ideas: care for populations driven by machine intelligence and care for individuals driven by personal genomics in conflict, competition or concert?

I'll add a third one into the mix-- behind the high gloss of hi tech (computational or biological), the problems of health care are actually extremely messy, mundane, and human. Accurate lists of names, getting clinicians and patients to do the things we already know work, organizing emotional narratives from which to drive decisions; these are the problems of today that each of us ignore at our own peril. 

So maybe I'll put forth a possible litmus test for evaluating any new health technology:

Does said technology (medicine, reminder app, clinical decision support tool, diagnostic coding suggester, etc., etc.) require conscious human effort and therefore adds complexity to a chaotic system, or does it remove conscious effort and reduces complexity in a chaotic system?