Home / Dinners / AI & Health

AI & Health

AI & Health

Eighty Arc Fusion leaders joined us for a wonderful evening of talks, dance, film, and a dinner produced by Dosa, the famed South Indian restaurant in San Francisco. Our group broke up into eight separate tables to hold “Jeffersonian Dinners” to discuss the question “What are your greatest hopes and fears about A.I. And A.I. and health?”

Arc Fusion Dinner
San Francisco
May 18, 2016

AI & Health

Eighty Arc Fusion leaders joined us for a wonderful evening of talks, dance, film, and a dinner produced by Dosa, the famed South Indian restaurant in San Francisco. Our group broke up into eight separate tables to hold “Jeffersonian Dinners” to discuss the question “What are your greatest hopes and fears about A.I. And A.I. and health?”

What people said

You go through three phases in addressing revolutionary, exponentially growing information technologies. One is ecstasy at the potential to solve these age-old problems, and then despair at the potential destructive power of these technologies. And ultimately the third phase… is understanding that it is the nature of being human to make progress and overcome problems and risks.

Ray Kurzweil
Google, Author

The types of areas that A.I. should learn include: how do we notice patterns like reading CAT scans in lung cancer so we spot people in phase one [stage of cancer], two and three versus four? How do we find the patterns of women who could take easy preventable actions that could prevent cervical cancer…?

Kevin Noble

Computers have learned to do things which previously were just sci-fi. So for example computers have learned how to recommend replies for you, and if you use inbox by Google, you will see this feature where the email will actually suggest possible replies to your email after reading it automatically.

Jeremy Howard

We think there's things that people do terrific that can't be replicated or mimicked by any machine: self directed goals, common sense, value judgment. And there's things that machines do terrifically that humans could never touch upon on a massive scale, things like large scale math pattern discovery and statistical reasoning. So together we think there's a power of the collaboration, an ability to elevate decision making to a new level.

Lori Fiber

AI talent and proprietary data sets will be the new oil and gold that wars are fought over, and we're already seeing this balkanization, sort of this acronym, this FAGMA if you will. Facebook, Apple, Google, Microsoft, Apple, these are the new superpowers battling it out for these. They're sucking up every AI talent out there. They're sucking up all the proprietary data sets because they are the new superpowers.

Tim Chang



Arc + IDEO “Framework” on AI & Health

Table Discussion and Ideas

Our dinner guests split up into eight tables where everyone answered a table talk question guided by a table talk leader:

What are your greatest hopes and fears about A.I. And A.I. and Health?

To answer this question yourself, and participate in the evenings' polls, click here.
Thanks to our table talk scribes who recorded the conversations at their table each in their own unique style.

Table 1

Jane Metcalfe
Table Talk Leader

IDEO Facilitator: Andrew Lovett-Barron

During our discussion, here are a a few things that popped out at a high level.

  • The idea of repping patient needs. Could ML systems service to identify and represent patience perspectives/advocate for patients?
  • The idea that things are just starting, but everyone is underestimating the difficulty of building this stuff. Healthcare infrastructure around data is going to seriously get in the way.
  • Technology can suck the profit out of unoptimized industries. The question is whether that will be reinvested in getting to good outcomes. Worthwhile question to explore.
  • We know a lot less than we claim to know, and the combination of people NOT understanding AI and NOT understanding wicked healthcare problems from cancer to nutrition is going to be a problem.

Hopes Most of our hopes were centered on patient care and outcomes.

  • AI would finally put the patient squarely at the center of the equation, unifying all records and capturing all relevant data so that care providers are always working with complete information
  • AI could lead to a democratization of healthcare, surfacing the best science and treatments from around the world and making them available for all
  • Perhaps there would evolve non financial incentives for treatment
  • Serve as physician support, not replacement
  • Generate medical and scientific breakthroughs by identifying patterns not see before
  • Avoid mistakes, increase efficiency, lower costs
  • Evolve into predictive models so patients don’t end up sick in the first place, and preventative


  • Organizations at the table are insurance companies, hospitals, doctors associations and pharma – patients are not at the table – entrenched interests solidify their positions
  • Crap software, bad data lead to bad or unreliable outcomes
  • Lack of qualified personnel
  • Only the rich have access to AI medicine, aggravating the gap between rich and poor
  • Robots replace doctors
  • Dominated by short term/financial goals more than patient outcomes or public health


  • Software competence
  • Sufficiently educated workforce
  • Nutrition – we know NOTHING about nutrition and the role it plays in disease (or as Ethan put it, we know SHIT SHIT SHIT about nutrition!)
  • Microbiome – also unknown the role it plays, but AI can help us learn
  • Bad data – how to we make sure we’re capturing good data to begin with
  • HPPA interferes with our ability to collect good data

Table 2

Rodrigo Martinez
Table Talk Leader

IDEO Facilitator: Farzad Azimpour

Rodrigo Martinez – Veritas:

  • Healthcare is manageable
  • Research not Development
  • How to engage people
  • Wish I could have genetics and microbiome data integrated into my care

Comments by others at the table:


  • When we talk humans we assume we’re talking about the best humans
  • Human intelligence has not been leveraged
  • You need to have humans! We can’t have isolated conversations.
  • Mental health would be better served with AI since humans just don’t “get it”.

IDEO Designer:

  • Hope that AI gets people off the plate so docs can focus on care
  • Augmented intelligence
  • Enhance visibilities provide uninhibited value


  • Make the “hassle map” visible
  • Fear that anti-bodies will kill the system of AI.


  • Won’t think of ramifications of all that data
  • Start small and roll it out. But people will get in the way of effective roll outs.


  • I want to see a system that makes people care about their health so they can engage and prevent health problems.

Table 3

Susan Feeny
Table Talk Leader

IDEO Facilitator: Grace Hwang

Notes on the dinner conversation below:

  • Starts with physical exam – who does this? How does AI make this better?
  • Applications driven by incentives and structure
  • Need Education, has not remade public system
  • Interdependent, Rigid > Modular, Felexible
  • Health incentive now: reward cost not value
  • Need data reform
  • Firms can’t make money by treating value – More efficient providers are paid less – 3 cents on dollar wasted
  • Opportunity for AI
  • Race to the bottom, need to create incentives

Table 4

Paul Saffo
Table Talk Leader

IDEO Facilitator: Peter MacDonald

Key Takeaways:

  • Optimistic Forecast – Super intelligence will treat us like pets
  • Pessimistic Forecast – Super intelligence will treat us like food
  • Will AI evolve to being alive?
  • AI will develop an immune system, will eliminate privacy and aberrant behavior
  • Hope: AI augments intelligence and gets good data so humans can solve bigger problems likfe climate and health policy
    • AI augments intuition
  • Fear: AI will make us dumber – we will not our brains or develop new skills
    • In the last 30 years, bigger gap between the knows and not knows, less math and science
  • Hope: Find a cure for cancer, fund new drugs
  • Fear: Bigger gap between rich and poor – loss of jobs, more homelessness (20 years from now?)
  • Hope: Will make us more human, more connected – evolve exponentially to solve problems faster – eliminate disease
  • Fears: Ethics of people building AI will cause the machines to lack ethics – the future is about ethics and about human compassion
  • Hope: We will quickly solve the world’s problems – it will give me my time back to focus on important things
  • Fear: Arms race develops… need to get people together to agree on safety
  • Transparency and connectiveness are key
  • Human mind and body are one
  • Hope: All rooted in nature/body
  • Fear: Becoming abstracted and disconnected
  • Hope: Sharing experience recipe based on own genomics to inform your actions
  • Fear: Losing cultural norms
  • Hope: Solve distribution problems in society – humans can evolve more spiritually and philosophically – beyond ancient traditions
  • Fear: Evil robots
  • Hopes: Life extension, access to healthcare and justice – replacing lawyers now
  • Fears: Massive social upheavel


  • What time frame are we talking about?
    • In our lifetime
  • We need to intentionalize the future
  • Make it personal for me and for people I care about
  • Hope and pursuit of intelligence – and don’t forget spiritual
  • Can AI learn intuition? Can this be taught?
  • Worry – A bunch of young boys developing this
  • Need for wisdom – for old and young to
  • What is the constitution and the Bill of

Table 5

Atul Butte
Table Talk Leader

IDEO Facilitator: Grant Wedner

Key Takeaways:

  • Hope: AI spots the critical moments and alerts us
    • AI brings a tipping point for extending life expectancy
  • Fear: The existential risk
  • Hope: AI eliminates the medical training gap
  • Fear: Mob psychology fears AI
  • Hope: AI augments education and stimulates creativity
    • Will AI have a gut?
  • Fear: We talk to too many people like ourselves
  • Hope: AI for education
  • Fear: Disruption to economy, loss of jobs – takes away purpose
  • You don’t need an astronaut to fly a crop duster
  • How much can AI fix if we don’t want to share the data?

Table 6

Christine Mason
Table Talk Leader

IDEO Facilitator: Kevin Ho

Key Takeaways:

  • Do people want to share data? 90% of big data security is not the issue – authentication is the issue.
  • Cybercrime – Target data stolen, not consequential
  • Personal health record – this should be done by UCSF and other providers.
  • Physician as God – should be coach, AI will allow patients to own their own data, patient will have knowledge
  • But… lots of patients don’t want to take ownership of their data or their health
  • AI needs to present data and interact with patients in a way that matches personality and desire
  • AI can help with predictions on behavioral adherence –
  • Information – Filter (based on values and personality) – Risk – Tech to solve for policy, regulation
  • Multimodel system – Fall off to healthcare? Lower value for providers?
  • Challenge: In whose hands are the AI? And what is the motivation of whomever controls the AI? Hospitals? Academics? Payers? Pharma? IT Companies?
  • For AI to be effective, patient community should have control – if it doesn’t, then AI can become insidious
  • AI needs to act like an organism, absorbing info and reacting to it with flexibility and acting with the input and output systems like a cell
  • We can use what’s been learned in behavioral economics – can AI be trusted to incorporate behavioral economics?
  • How can you create digital empathy? This could become a conduit for doctors, could anticipate people’s anxieties.
  • Make digital values and morals
  • Cannot train for culture
  • Converging with AI pattern recognition – does deep thought get limited by distributed cognition?
  • Emotion recognition – will this evolve?
  • Empirical loop – fast
    • Ethical loop – slow
  • Gap in ethics and technology – we need “computational ethics”
  • Short term – anxiety about AI
    • Long term – optimism
  • Could AI move too fast?
  • No shutdown mode
  • Small countries can do genetic engineering
  • Mother nature will have something to say about the evolution of humans and AIU- environmental changes will create unforeseen impact

Table 7

Bruce MacGregor
Table Talk Leader

IDEO Facilitator: Remy Fisher

Key Takeaways:


  • Opportunity for AI to do what we can’t do well.
  • Inventing machines for what we’ve not good at (like calculations)
  • Fears: We are the weak links, 2 things:
    • Predicting non-linear events
    • How can we adapt for that both positive and negative? A natural learning curve, can we keep up?

Biotech Marketer:

  • Hope: Ability to recognize patterns that we can’t see – identify them earlier – exponential complications
  • Fear: Human tendencies to silo info, share info to confound what AI can do – it is only as good as we develop it

Neuropsychologist and diagnostician:

  • Hope: AI will be able to offer all providers a way to intake patient history in a complete way – capture everything I need to know to give the best medical advice – make more time for the human
  • Fear: Decision-making relied on too soon – when there will be errors


  • Fear: Augmented existence – loss of intuition/relationships/connections – this has already happened with Google Maps
    • Loss of social structures and culture
    • Autopilot where pilot forgets how to fly


  • Hopes: Go on medical journeys with AI – health lens is through humanity
    • AI can help navigate complex systems
    • When dying, look into the eyes of a person, not a thing
    • Can AI be machines of loving grace?
    • AI can help us understand intricate connections – microbiome
  • Fear: Disconnect and reliance on AI – racist chat bots, reflecting on what is said on Twitter and social media – isolations


  • Only has fear – EMR’s have been optimized around billing, not care – automated tools are bandaids – AI = bandaids – a screwed up system is being amplified


  • Ironically the idea of EMRs is pushing out innovation
  • Optimist – Internet has brought vast amounts of information directly to us
  • Fear: human genome becomes a commodity
  • Political and sociological constraints will prevent AI from really working
  • How much do you trust a machine?
  • How can we make AI totally transparent?
  • Microsoft bots have no values
  • I fear AI will amplify dysfunctionality

Table 8

Kevin Stone
Table Talk Leader

IDEO Facilitator: Hanah Williams

Key Takeaways:

  • Fear: intention in creation – we need neutral AI, but we live in a patriarchal system, will AI be able/need to Delta the way we think of gender? What gender/who determines how AI thinks
  • Hope that AI will help find pathways independent of institutions.
  • Fear it will take the joy of discovery away from us, hope it will
  • Mindful – augmented vs. replacement, human brain has so much capability that we don’t use  - how to tap into cosmic consciousness and enhance – vital we keep our physicality and emotions – hope it can deepen our human experience
  • Will need data improvement –
  • Humans are AI think we will develop more on ourselves don’t understand why others are scared – use tech to be more psychic
  • Predict best health mgmt., bring clinicians and data scientists together –
  • Hope: Augment and scale came to deliver care – Fear: giving away responsibility of thinking – does not absolve you of responsibility but augments you to your fullest potential.
  • Hope: that we can scale
  • Hope: how to think what it means to be human, who is teaching right from wrong – Fear: It may influence us more than we influence it – we don’t overoptimize to the err of losing our human experience
  • NEED: Human side, has to make ourselves better, how to make a doctor/decision better
  • Fear: most studies in med are pretty poor, how to analyze the science when the database is so bad
  • Augmentation you’re giving is very bad
  • If electronic records become ubiquitous then has the quality improved?
  • NEED: Real time aggregation to show what you have
  • NEED: Higher quality data
  • Do we need another level of scrutiny
  • What sources of information will influence your immortal self
  • Curated version from social media
  • Let your AI watch you rather than teach it
  • Even if you have all of the inputs, will it ever be more than a mimic of you vs. an embodiment of its own
  • How to replicate: Creativity, empathy, ourselves (as we evolve), consciousness/spirituality
  • What is the Fitbit for your consciousness? Took for unfiltered messaging of your soul, can it be implantable rather than gathering
  • What is the authentic mirror of the self?
  • concerned about hubris of humans
  • Our thinking side is what makes us unique
  • to be more connected, empathetic, human new communications
  • Empower us to be more; how to have AI make us more human
  • Make things we contribute more powerful

Arc Fusion Survey on AI & Health

Arc Fusion Photo Gallery