Cognitive Collaboration—Why Data Science NeedsHuman-Centered Design

Jim GuszczaUS Chief Data Scientist Deloitte Analytics

‘AI IS THE NEW ELECTRICITY’

“Just as electricity transformedalmost everything 100 years ago,today I actually have a hard timethinking of an industry that I don’tthink AI will transform in the nextseveral years.”Andrew NgFormer chief scientist at Baidu, co-founder at Coursera

The AI “master narrative:” our new computer overlords

“Every aspect of learning or any other feature of intelligence can, in principle, be so precisely described that a machine can be made to simulate it” -- 1956 Dartmouth Conference

“About 47% of total US employment is at risk[of computerization]”-- Oxford U study

“Google’s AlphaGo is demonstrating for the first time that machines can truly learn and think in a human way”-- New York Times March, 2016

“Before the prospect of an intelligence explosion we humans are like small children playing with a bomb”

-- Nick Bostrom

The development of computers that are able to do things normally done by people …

… any program can be considered AI if it does something that we would normally think of as intelligent in humans.

How the program does it is not the issue, just that is able to do it at all.

That is, it is AI if it is smart, but it doesn’t have to be smart like us.

— Kris Hammond, Narrative Science

What is AI, really?

The problem of “artificial stupidity”

Smart technologies are unlikely to engender smart outcomesunless they are designed to promote smart adoption

on the part of human end users.

Smart technologies are unlikely to engender smart outcomesunless they are designed to promote smart adoption

on the part of human end users.

Effective and Ethical AI needs human-centered design

The problem with the designs of most engineers is that they are too logical.

We have to accept human behavior the way it is, not the way we would wish it to be.

— Don Norman, The Design of Everyday Things

The AI revolution needs a design revolution

Human-centricity: understanding the user

AI and other data products will yield better outcomes if they are designed to go with the grain of human psychology.

Data science teams must think like designers …

… not just “engineers.”

Slow, then Fast

Thinking slow

AI as “automation”

“… about 47% of total US employment is at risk.”-- Frey/Osborne (Oxford U)

Can underwriting be “computerized?”

Can underwriting be “computerized?”

Can underwriting be “computerized?”

Actuarial vs. clinical prediction – the motion picture

Human judges are not merely worse than optimal regression equations; they are worse than almost any regression equation.

— Richard Nisbett and Lee Ross

Time for a sequel to Moneyball?bias

Maybe it’s time we break for lunch

http://economix.blogs.nytimes.com/2011/04/14/time-and-judgment/

noise

A false comparison

Models are a form of “artificial intelligence” that augment (but do not replace) human expertise.

Equations > experts(Equations + experts) > experts

nn XXXY *...** 22110 ββββ ++++=

Algorithms as “eyeglasses for the mind’s eye”

Equations > experts(Equations + experts) > experts

nn XXXY *...** 22110 ββββ ++++=

Marvin [Minsky] was advocating what’s called “commonsense reasoning.” Machines have shown essentially no examples of doing that.Therefore, they are complements to people.People are actually not so bad at that.However, they are somewhat lousy at tuning things and keeping exact accounts of stuff. Machines are good at that.That gives the idea that there could be a human-machine partnership …

— Sandy Pentland, Deloitte Review 2017

AI = Augmented Intelligence

Copyright © 2017 Deloitte Development LLC. All rights reserved.

AI = Augmented Intelligence

Copyright © 2017 Deloitte Development LLC. All rights reserved.

AI != Human Intelligence

AI != Human Intelligence

The prequel to Jeopardy …

Human-computer collective intelligence

Human-computer collective intelligence

Their skill at manipulating and “coaching” their computers to look very deeply into positions effectively counteracted the superior chess understanding of their grandmaster opponents and the greater computational power of other participants.Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.

— Garry Kasparov

Human-computer collective intelligence

Their skill at manipulating and “coaching” their computers to look very deeply into positions effectively counteracted the superior chess understanding of their grandmaster opponents and the greater computational power of other participants.Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.

— Garry Kasparov

Creating such processes goes beyond data science – psychology and design thinkingare needed.

Human-computer collective intelligence

The future of work is “Freestyle x”

The problems that we face with technology are fundamental … We need a calmer, more reliable, more humane approach.

We need augmentation, not automation.

– Don Norman

consistent de-biased informed meaningful

data + human judgment / empathy decisions that are…

Thinking fastHow data science and behavioral science can work together

The “last mile problem” of data science

MODEL

Algorithms can point us in the right direction, but are not a complete solution.

The “last mile problem” of data science

MODEL

Algorithms can point us in the right direction, but are not a complete solution.

They must be followed by the right judgments, decisions, or behavior change.

Nudge is design thinkingDesign considerations• Smart defaults• Present bias• Loss aversion• Social proof, “Social Physics”• Framing effects• Intuitive language / infoVis• Status quo bias• Mental accounting• Cognitive load / “Scarcity”• Pre-commitment • Lotteries (overweighing small probabilities)

• Unit bias (“mindless eating”)• Removing bottlenecks

“While Cass and I were capable of recognizing good nudges when we came across them, we were still missing an organizing principle for how to devise effective nudges..

We had a breakthrough … when I reread Don Norman’s classic book The Design of Everyday Things.”

– Richard Thaler, Misbehaving

Example #1: “push the worst, nudge the rest”

Data sciencePredictive algorithms to deploy building inspectors to the highest-risk buildings.

Behavioral scienceBehavioral nudge tactics could be employed to ameliorate lesser risks that don’t merit immediate physical inspections.

… and similarly with • health / safety inspections• workers comp premium audits• program integrity audits …

Example #2: the power of positive psychology

Data sciencePredictive algorithms can estimate severity/ duration of workers comp claims.

Behavioral science• Mild cases: use Opower-style peer

comparisons to prompt timely return to work

• Severe cases: Cognitive behavior therapy [CBT] reduces time away by ≈40%

Example #3: keeping ourselves honest

Data scienceAlgorithms can identify cases relatively likely to be collecting benefits improperly.(Problem: false positives!)

Behavioral scienceUse nudge messages to prompt accurate disclosure of income.

(A/B test messages in online environment)

Sometimes “Nudge” can enable ethical AI

Naïve view “Nudge” view

Taking on board behavioral insights and choice architecture (“Nudge”) enables new ways to operationalize predictive algorithms algorithms.

“Successful use of wearable health devices will depend more on the design of engagement strategies than the technology itself.”

Data: Gathered by wearablesDigital: Smart screen interfacesDesign: Commitment devices, loss aversion,

temptation bundling, gamification, …

Wearable Devices as Facilitators, Not Drivers of Health Behavior Change

Journal of the American Medical Association, Feb, 2015Mitesh S. Patel, David A. Asch, Kevin G. Volpp

Additional Resources

“The Last Mile Problem: How data science and behavioral science can work together” Deloitte Review, January 2015http://dupress.com/articles/behavioral-economics-predictive-analytics/

“The Importance of Misbehaving: A conversation with Richard Thaler” Deloitte Review, January 2016https://dupress.deloitte.com/dup-us-en/deloitte-review/issue-18/behavioral-economics-richard-thaler-interview.html

“Cognitive collaboration: Why humans and computers think better together” Deloitte Review, January 2017https://dupress.deloitte.com/dup-us-en/deloitte-review/issue-20/augmented-intelligence-human-computer-collaboration.html

“Smarter together: Why artificial intelligence needs human-centered design” Deloitte Review, January 2018https://www2.deloitte.com/insights/us/en/deloitte-review/issue-22/artificial-intelligence-human-centric-design.html

“Superminds: How humans and machines can work together”(Interview with Thomas Malone, MIT Sloan School of Management Deloitte Review, January 2019https://www2.deloitte.com/insights/us/en/focus/technology-and-the-future-of-work/human-and-machine-collaboration.html