Giving algorithms a sense of uncertainty could make them more ethical

The algorithm could handle this uncertainty by computing multiple solutions and then giving humans a menu of options with their associated trade-offs. Say the AI system was meant to help make medical decisions. Instead of recommending one treatment over another, it could present three possible options: one for maximizing patient life span, another for minimizing patient suffering, and a third for minimizing cost. “Have the system be explicitly unsure and hand the dilemma back to the humans.”

Hao, K. (2019). Giving algorithms a sense of uncertainty could make them more ethical. MIT Technology Review.

I think about clinical reasoning like this; it’s what we call the kind of probabilistic thinking where we take a bunch of – sometimes contradictory – data and try to make a decision that can have varying levels of confidence. For example, “If A, then probably D. But if A and B, then unlikely to be D. If C, then definitely not D”. Algorithms (and novice clinicians) are quite poor at this kind of reasoning, which is why they’ve traditionally not been used for clinical decision-making and ethical reasoning (and why novice clinicians tend not to handle clinical uncertainty very well). But if it turns out that machine learning algorithms are able to manage conditions of uncertainty and provide a range of options that humans can act on, given a wide variety of preferences and contexts, it may be that machines will be one step closer to doing our reasoning for us.

Algorithmic de-skilling of clinical decision-makers

What will we do when we don’t drive most of the time but have a car that hands control to us during an extreme event?

Agrawal, A., Gans, J. & Goldfarb, A. (2018). Prediction Machines: The Simple Economics of Artificial Intelligence.

Before I get to the takehome message, I need to set this up a bit. The way that machine intelligence currently works is that you train an algorithm to recognise patterns in large data sets, often with the help of people who annotate the data in advance. This is known as supervised learning. Sometimes the algorithm can be given data sets that have no annotation (i.e. no supervision), and the output is judged against some criterion and determined to be more or less accurate. This is known as reinforcement learning.

In both cases, the algorithm isn’t trained in the wild but is rather developed within a constrained environment that simulates something of interest in the real world. For example, an algorithm may be trained to deal with uncertainty by playing Starcraft, which mimics the imperfect information state of real-world decision-making. This kind of probabilistic thinking defines many professional decision-making contexts where we have to make a choice but may only be 70% confident that we’re making the right choice.

Eventually, you need to take the algorithm out of the simulated training environment and run it in the real world because this is the only way to find out if it will do what you want it to. In the context of self-driving cars, this represents a high-stakes tradeoff between the benefits of early implementation (more real-world data gathering, more accurate predictions, better autonomous driving capability), and the risks of making the wrong decision (people might die).

Even in a scenario where the algorithm has been trained to very high levels in simulation and then introduced at precisely the right time so as to maximise the learning potential while also minimising risk, it will still hardly ever have been exposed to rare events. We will be in the situation where cars will have autonomy in almost all driving contexts, except those where there is a real risk of someone being hurt or killed. At that moment, because of the limitations of its training, it will hand control of the vehicle back to the driver. And there is the problem. How long will it take for drivers to lose the skills that are necessary for them to make the right choice in that rare event?

Which brings me to my point. Will we see the same loss of skills in the clinical context? Over time, algorithms will take over more and more of our clinical decision-making in much the same way that they’ll take over the responsibilities of a driver. And in almost all situations they’ll make more accurate predictions than a person. However, in some rare cases, the confidence level of the prediction will drop enough to lead to control being handed back to the clinician. Unfortunately, at this point, the clinician likely hasn’t been involved in clinical decision-making for an extended period and so, just when human judgement is determined to be most important, it may also be at it’s most limited.

How will clinicians maintain their clinical decision-making skills at the levels required to take over in rare events, when they are no longer involved in the day-to-day decision-making that hones that same skill?


18 March 2019 Update: The Digital Doctor: Will surgeons lose their skills in the age of automation? AI Med.

Proposal abstract: Case-based learning in undergraduate physiotherapy education

Abstract for a project I submitted earlier this week for ethics clearance. During 2012 – 2014 we converted one of our modules that runs in the 2nd, 3rd and 4th year levels from a lecture-based format to a case-based learning format. We are now hoping to have a closer look at whether or not the CBL approach led to any changes in teaching and learning behaviours in staff and students.

Case-based learning (CBL) is a teaching method that makes use of clinical narratives to create an authentic learning activity in which students navigate their way through complex patient scenarios. The use of CBL in a health professions undergraduate curriculum attempts to convey a multidimensional representation of the context, participants and reality of a clinical situation, allowing students to explore these concepts in the classroom. While the implementation of CBL has a sound theoretical basis, as well as a strong evidence base for use in health professions education, there are challenges in its effective use that are not easily resolved. However, if it can be shown that the approach leads to changes in teaching and learning practice, which enhance student learning, providing additional resources to resolve the challenges can be more strongly justified. This project therefore aims to determine staff members’ and students’ perceptions of CBL as a teaching method, and to find out how it influenced their teaching and learning behaviours.

This study will make use of a mixed method research design in which the experiences and perceptions of student and staff members are used to determine whether or not there was a change in their teaching and learning practice. Qualitative and quantitative data will be gathered using a survey of all students in the population, focus group discussions of students and in-depth interviews of all staff in the department. The survey will determine if the design of the CBL approach led to a change in what the students did. The focus group discussions will gather data on the nature of the changes and the underlying rationale for those changes. The interviews with lecturers will be conducted in order to delve more deeply into whether or not lecturers’ teaching behaviours changed, and again, to explore the underlying rationale of those changes.

The survey will make use of a self-developed questionnaire that will gather quantitative data using Likert scales and other closed-ended questions. The survey will be sent to all 3rd and 4th year students in the 2015 academic year. The same students will be invited to participate in the focus groups, and the researchers will make use of purposive sampling to allocate volunteers into two focus groups in each year level. All lecturers in the department (n=10) will be invited to participate in the in-depth interviews, including those who were not directly involved in the implementation of CBL. In addition, we will also invite ex-staff members who were involved in the process, as well as postgraduate students who assisted with student facilitation.

Qualitative data will be gathered during the focus groups and interviews. This data will be interpreted via the theoretical frameworks used in the design of the CBL cases. The focus group discussions and interviews will be conducted in English and recorded using a digital audio recorder. The audio files will be sent for verbatim transcription and the anonymised, transcribed documents will then be sent to participants for verification. The transcripts will be analysed thematically, coding the data into categories of emerging themes. Trustworthiness of the analysis will be determined through member checking and peer debriefing and participants will be given the opportunity to comment on whether or not the data was interpreted according to what they meant. The transcribed verbatim draft will be given to colleagues who were not involved in the study for comment.

Design principles for clinical reasoning

graphic_design smallerClinical reasoning is hard to do, and even harder to facilitate in novice practitioners who lack the experience and patterns of thinking that enable them to establish conceptual relationships that are often non-trivial. Experienced clinicians have developed, over many years and many patients, a set of thinking patterns that influence the clinical decisions they make, and which they are often unaware of. The development of tacit knowledge and its application in the clinical context is largely done unconsciously, which is why experienced clinicians often feel like they “just know” what to do.

Developing clinical reasoning is included as part of clinical education, yet it is usually implicit. Students are expected to “do” clinical reasoning, yet we find it difficult to explain just what we mean by that. How do you model a way of thinking?

One of the starting points is to ask what we mean when we talk about clinical education. Traditionally, clinical education describes the teaching and learning experiences that happen in a clinical context, maybe a hospital, outpatient or clinic setting. However, if we redefine “clinical education” to mean activities that stimulate the patterns of thinking needed to think and behave in the real world, then “clinical education” is something that can happen anywhere, at any time.

My PhD was about exploring the possibilities for change that are made available through the integration of technology into clinical education. The main outcome of the project was the development of a set of draft design principles that emerged through a series of research projects that included students, clinicians and clinical educators. These principles can be used to design online and physical learning spaces that create opportunities for students to develop critical thinking as part of clinical reasoning. Each top-level principle is associated with a number of “facets” that further describe the application of the principles.

Here are the draft design principles (note that the supporting evidence and additional discussion are not included here):

1. Facilitate interaction through enhanced communication

  • Interaction can be between people and content
  • Communication is iterative and aims to improve understanding through structured dialogue that is conducted over time
  • Digital content is not inert, and can transform interactions by responding and changing over time
  • Content is a framework around which a process of interaction can take place – it is a means to an end, not an end in itself
  • When content is distributed over networks, the “learning environment” becomes all possible spaces where learning can happen
  • Interaction is possible in a range of contexts, and not exclusively during scheduled times

2. Require articulation

  • Articulation gives form and substance to abstract ideas, thereby exposing understanding
  • Articulation is about committing to a statement based on personal experience, that is supported by evidence
  • Articulation is public, making students accountable for what they believe
  • Articulation allows students’ thinking to be challenged or reinforced
  • Incomplete understanding is not a point of failure, but a normal part of moving towards understanding

3. Build relationships

  • Knowledge can be developed through the interaction between people, content and objects, through networks
  • Relationships can be built around collaborative activity where the responsibility for learning is shared
  • Facilitators are part of the process, and students are partners in teaching and learning
  • Facilitators are not gatekeepers – they are locksmiths
  • Create a safe space where “not knowing” is as important as “knowing”
  • Teaching and learning is a dynamic, symbiotic relationship between people
  • Building relationships takes into account both personal and professional development
  • Building relationships means balancing out power so that students also have a say in when and how learning happens

4. Embrace complexity

  • Develop learning spaces that are more, not less, complex
  • Change variables within the learning space, to replicate the dynamic context of the real world
  • Create problems that have poorly defined boundaries and which defy simple solutions

5. Encourage creativity

  • Students must identify gaps in their own understanding, and engage in a process of knowledge creation to fill those gaps
  • These products of learning are created through an iterative activity that includes interaction through discussion and feedback
  • Learning materials created should be shared with others throughout the process, to enable interaction around both process and product
  • Processes of content development should be structured according to the ability of the students

6. Stimulate reflection

  • Learning activities should have reflection built in
  • Completing the reflection should have a real consequence for the student
  • Reflection should be modelled for students
  • Reflections should be shared with others
  • Feedback on reflections should be provided as soon after the experience as possible
  • Students need to determine the value of reflection for themselves, it cannot be told to them

7. Acknowledge emotion

  • Create a safe, non­judgemental space for students to share their personal experiences and thoughts, as well as their emotional responses to those experiences
  • Facilitators should validate students’ emotional responses
  • These shared experiences can inform valuable teaching moments
  • Facilitators are encouraged to share personal values and their own emotional responses to clinical encounters, normalising and scaffolding the process
  • Sensitive topics should be covered in face­to­face sessions
  • Facilitators’ emotional responses to teaching and learning should be acknowledged, as well their emotional responses to the clinical context

8. Flexibility

  • The learning environment should be flexible enough to adapt to the changing needs of students, but structured enough to scaffold their progress
  • The components of the curriculum (i.e. the teaching strategies, assessment tasks and content) should be flexible and should change when necessary
  • Facilitators should be flexible, changing schedules and approaches to better serve students’ learning

9. Immersion

  • Tasks and activities should be “cognitively real”, enabling students to immerse themselves to the extent that they think and behave as they would be expected to in the real world
  • Tasks and activities should use the “tools” of the profession to expose students to the culture of the profession
  • Technology should be transparent, adding to, and not distracting from the immersive experience

We have implemented these draft design principles as part of a blended module that made significant use of technology to fundamentally change teaching and learning practices in our physiotherapy department. We’re currently seeing very positive changes in students’ learning behaviours, and clinical reasoning while on placements, although the real benefits of this approach will only really emerge in the next year or so. I will continue to update these principles as I continue my research.

Note: The thesis is still under examination, and these design principles are still very much in draft. They have not been tested in any context other than in our department and will be undergoing refinement as I continue doing postdoctoral work in this area.

I enjoyed reading (December)

reading outsideI’m going to try something new on this blog. At the end of every month I’ll write a short post highlighting the things I particularly enjoyed reading. I found that simply pushing them into a Twitter or Google+ feed would tend to obfuscate them among all of the other things that I wanted to point out to people. I guess this post is a way to say, “Of all the things I read this month, these are the ones I enjoyed the most”. I’m not trying to summarise everything I read, just present a small sampling. I’ll try it out for a few months and see if I like the process.

 

The web we lost (Anil Dash). A look back over the past 5-10 years of social media and how things have changed, usually not for the better. In many instances, we’re actually worse off now than we were before the rise of the new social platforms. He talks about how we’re progressively losing control of our online identities, of the content we create and share (and which makes those platforms as powerful as they are), and lost sight of the values that actually led to the development of the web in the first place. Here’s a quote from the end of the article:

I know that Facebook and Twitter and Pinterest and LinkedIn and the rest are great sites, and they give their users a lot of value. They’re amazing achievements, from a pure software perspective. But they’re based on a few assumptions that aren’t necessarily correct. The primary fallacy that underpins many of their mistakes is that user flexibility and control necessarily lead to a user experience complexity that hurts growth. And the second, more grave fallacy, is the thinking that exerting extreme control over users is the best way to maximize the profitability and sustainability of their networks.

The first step to disabusing them of this notion is for the people creating the next generation of social applications to learn a little bit of history, to know your shit, whether that’s about Twitter’s business model or Google’s social features or anything else. We have to know what’s been tried and failed, what good ideas were simply ahead of their time, and what opportunities have been lost in the current generation of dominant social networks.

Update: Here’s a follow up post from Anil on Rebuilding the web we lost.

 

Mobile Learning, Non-Linearity, Meaning-Making (Michael Sean Gallagher). What I liked most about this post is the suggestion, presented below, that the true power of “mobile” is that it transforms every space into a potential learning space.

They refer to the ‘habi­tus’, the sit­u­at­ed locale of the indi­vid­ual. Yet the locale doesn’t define the learn­ing per se as the process of mobile learn­ing trans­forms the habi­tus into a learn­ing space. Tools, con­tent, and com­mu­ni­ty are recon­struct­ed to allow for meaning-making. Turn­ing the envi­ron­ment in which we hap­pen to find our­selves into an envi­ron­ment for learn­ing. Mobile tech­nol­o­gy assists in bring­ing these ele­ments into con­junc­tion, an orga­niz­ing agent in this process. But it is real­ly about the trans­for­ma­tion. From space to learn­ing space. From noise to mean­ing.

 

Arm Teachers? (Tom Whitby). When I first read about the suggestions to arm teachers, in the wake of the Newtown shooting, I dismissed it as ridiculous without even considering it. What I liked about this post from Tom is that instead of just dismissing the suggestion out of hand, he follows it through to some logical conclusions. I realised that his approach does far more to systematically dismantle the argument than simply rejecting it.

 

The demon-haunted world: Science as a candle in the dark (Carl Sagan). Carl Sagan is one of my heroes. Few people have done as much as he did to bring a sense of wonder about the world, to the public. This book is an exploration of scientific thinking over the past few centuries, highlighting the many areas where a lack of this critical approach to the world has led to a stumbling of our species. Think of the hysteria of witch-burning, UFO abductions, racism and all the other instances where a lack of critical thought has brought so much suffering and misunderstanding about the world. This book should be required reading for everyone.

 

The robot teachers (Stephen Downes). Stephen argues against the idea of universities and higher education in general as a system designed to maintain division between a cultural elite and everyone else. He suggests that the solution is not to open up those institutions (i.e. MIT, Harvard, etc.) but to build a better system outside of them.

We must develop the educational system outside the traditional system because the traditional system is designed to support the position of the wealthy and powerful. Everything about it – from the limitation of access, to the employment of financial barriers, to the creation of exclusive institutions and private clubs, to the system of measuring impact and performance according to economic criteria, serves to support that model. Reforming the educational system isn’t about opening the doors of Harvard or MIT or Cambridge to everyone – it’s about making access to these institutions irrelevant. About making them an anachronism, like a symphony orchestra, or a gentleman’s club, or a whites only golf course, and replaced with something we own and build for everyone, like punk music, a skateboard park, or the public park.

Twitter Weekly Updates for 2011-10-17

  • @Suhaifa it’s an easy walk, easier than lions head, don’t stress 🙂 #
  • Daily Papert http://t.co/vJQhrNh4. We can’t solve the world’s problems with the same thinking that created them #
  • Critical Thinker Explains Skepticism vs. Cynicism http://t.co/Zmxh81m9 via @zite #
  • RT @engadget: MobiUS smartphone ultrasound hits the market two years too late for relevancy http://t.co/DaWRQqXo #
  • Stephen’s Web – Free learning: essays on open educational resources and copyright http://t.co/b8d7fDXK via @zite #
  • The atomic method of creating a Powerpoint presentation http://t.co/1ikf4gBO via @zite #
  • The Complexity Of Learning http://t.co/YSJfJwkq via @zite #
  • @USMCShrink Focus in education is that tech is good 2 get more content 2 more students in less time 4 less money, which misses the point #
  • A Tablet for the Blind? – Technology Review http://t.co/jBqyDK0R. Elegant and clever solution #
  • What’s Behind the Culture of Academic Dishonesty http://t.co/UuTvlENW. Cheating doesn’t help if learning matters more than grades #
  • #Zite now my favourite news reading app on the iPad. Flipboard not iterating fast enough #
  • Ask the Students: Their Wise Wishes for Improving Education http://t.co/KUDtKP6p. I keep saying that health education needs more art #
  • @USMCShrink its about not making the assumption that technology in education is automatically a good thing #
  • @USMCShrink I just highlighted a quote, so it was out of context. Did you read the rest of the post? #
  • “technology will be used…for the profit of corporations rather than…the benefit of children” http://t.co/RLAjJI5u #
  • If you are a clinician who supervises or teaches healthcare students, consider completing my survey http://t.co/x1MXf3AJ. Please RT #

Problem based learning: transitioning to an online / hybrid learning environment

A few weeks ago I attended a short presentation by Prof. Meena Iyer from Missouri University. Prof. Iyer spoke about how she moved her PBL module from using a traditional, mainly face-to-face approach, to an online / hybrid approach. Here are my notes.

—————————-

“All life is problem solving” – Karl Popper

How do we get students to think like professionals in the field?
How do we foster group interaction in online spaces?
How do I assess learning in online spaces?

PBL addresses the content issue, as well as enhancing critical thinking through the collaborative solving of authentic, real-world problems

Mismatch:

  • PBL → solving problems is the tool, learning is the goal
  • Traditional → content is the tool, problem solving is the goal

PBL is all unstructured (but it can be scaffolded), and there’s not necessarily a right/wrong answer

Six steps to problem solving (IDEALS):

  • Identify the problem (What is the real question we are facing?)
  • Define the context (What are the facts that frame this problem?)
  • Enumerate the choices (What are the plausible actions?)
  • Analyse the options (What is the best course of action?)
  • List reasons explicitly (why is this the best course of action?)
  • Self-correct (What did we miss?)

The problem should be authentic and appealing (a mystery to solve)
Clearly outline expectations for each step of the process

Why move from face-to-face to online?

  • In F2F, you can only move forward at the speed of the slowest learner
  • Significant time requirements for F2F
  • Identify…can be anonymous online → fewer preconceived biases among students

Challenges:

  • How do you transition F2F to online
  • What tools are appropriate / feasible / viable / affordable?
  • How do you do collaborative work when everyone is online at different times?

Format:

  • Cases are presented in multiple formats / media
  • Introductory week to familiarise students with online environment. In addition to learning the content and critical thinking, students also have to learn about PBL
  • Scenarios are released in 2 stages over a 2 week period
  • Scenarios are accompanied by a set of probing questions to stimulate discussion
  • Teacher provides support during the discussions
  • Students must also design their own case
  • Assessment is based on content and depth
  • Wiki used for question / answer. Each student must answer each of the questions, each answer must be different i.e. must add to what has already been added (this means that the question can’t just be a knowledge question)
  • Discussion boards are used for students to dissect the cases (All and Group)
  • Each group assesses their own knowledge base, and define what the gaps are, and therefore what they need to find out (who provides the links to the resources, or can students use any resources?)
  • At least 3 posts per student, including: Summarise and question one citation; Answer another students’ question; Follow up any discussion on their own posts
  • Reading assignment: written, critial appraisal of a published article relevant to the case study. This summary must be posted online.

Important for students to learn how to share information in supportive environments

Assessment:

  • What parts of the process need to be assessed?
  • What parts can be graded as a group?
  • What needs to be submitted for individual assessment?
  • What are the time constraints for the grading?
  • How do you balance grading workload with the need to externally motivate student performance?
  • There is also a syllabus quiz to ensure the students actually know the content

Design:

  • Make the problem compelling
  • Outline expectations
  • The problem analysis should relate to the professoinal field
  • As student proficiency develops, withdraw support
  • Use learning issues to encourage EBP
  • Ensure that solution development is based on critical appraisal

Resources

  • Barrows, HS (1996). Problem based learning in medicine and beyond: a brief overview. New directions for teaching and learning
  • Barrows HS & Tamblyn, RM (1980). Problem based learning: an approach to medicla education. New York, Springer Pub. Co.
  • Hmelo-Silver, C (2004). Problem based learning: what and how do students learn? Educational Psychology Review, 16(3)
  • www.criticalthinking.org

 

Twitter Weekly Updates for 2011-04-18

Posted to Diigo 04/12/2011

    • humans are incapable of imagining something they have never actually experienced
    • this is one of the most important reasons that it is so hard for the teaching of thinking skills to take hold in education
    • Teachers and curriculum designers who were never asked to think in their own educations cannot imagine how to include it in their own teaching. More importantly, they have no idea how to assess it.
    • Differing ideas bring different levels of value, but they all bring value. Even an observation that leads to a dead end is not wrong — it is just an observation that turns out not to be productive, and the process of finding the dead end is helpful in itself. Students are not only learning to think, they are learning the collaborative process by which modern work teams complete their projects.
    • To use this technique, the teacher must be comfortable facilitating a process that may go in a totally unpredictable direction. Students may notice things the teacher hadn’t already known. Students might be puzzled by and ask questions about something the teacher doesn’t know. The skilled teacher will not be bothered by this because the act of working this out with the students is an excellent lesson in itself. A teacher who cannot say “Hmmm. I have never thought of this before. Let’s think about it” will not be happy with such an approach.
    • A course that focuses instruction on thinking skills needs to focus assessment on thinking skills as well
    • Assessment is actually the key. If students are assessed by how they use thinking skills to analyze a new work of literature, devise a scientific experiment, draw a conclusion about historical data, or apply appropriate mathematical processes to solve a problem, then they will need to be taught how to do it in the first place
    • We talk hard about life-long learning, but I do not believe that it is figuring in to the procedures, policies, and pedagogies of formal education nearly as much as it should
    • Today, with everything changing so fast, the ability and proclivity to learn is as critical as the basic literacies were in my time
    • Imagine education focusing less on what’s been taught, and much much more on skilled, curious, resourceful, and habitual learning