(Thread) at the #mkt4intel conference in Toronto on A.I. Lots of interesting economic insights on impact of AI. Will try and capture highlights in this thread.
Let’s ignore for a second the catchiness of the conference name of “machine learning and the market for intelligence, a conference by creative destruction lab at the Rotman school of management at university of Toronto” 😱😅😂
Good question from @professor_ajay on what is the impact of the power of medical diagnosis moving from distributed across doctors globally to AI... because eventually the power will consolidate to just one or two AI companies. See libraries-> google. Cabs->uber,lyft
Nice to hear a mention of @AndrewYang’s coverage of the political impact of automation and job loss. Even if @joshgans doesn’t completely agree, good to see #yanggang creating awareness up in Canada
Laura Rosella up now taking about diagnostics and prevention of colon cancer using machine learning... well... that’s surprisingly personally relevant to me. Maybe a little too much for 6.45am PT
@LauraCRosella talking about using machine learning for precision public health strategies. Would have to agree that feels like we could do a lot more to preemptively adapt strategies for eg cancer or diabetes based on data. Would need to solve privacy concerns tho. #mkt4intel 
Inequality widening in (largely preventable) premature deaths
Specific medical applications where machine learning is already at or above performance of a human, shown by @MarzyehGhassemi - some pretty impressive things in this list.
Off topic... have to say it’s refreshing (and just objectively better) to be at a conference where it’s equal or more female speakers than white male speakers. Tech is lagging way behind academia despite efforts. #mkt4intel
@MarzyehGhassemi we did research on mortality including typed doctors notes. “Natural language processing models just die trying to interpret typed doctors notes. Worse than their hand writing” 🤣
(I’m too stupid to be able to synthesize any highlights from @A_Aspuru_Guzik’s talk. But the intersection of chemistry, computer science and ML seems like it would be interesting to work in, if you, unlike me, didn’t suck at chemistry.)
Agenda switches from academics to tech companies after the break. I expect slide design skills to improve significantly, but maybe less interesting unadulterated altruism. #mkt4intel
Ok having talked to enough people in the break to translate it for me, the last talk was about some fairly amazing work. Using ML to study all known materials to be able to predict what other materials *could* exist, to then be able to eg make synethesized battery anodes. 🤯🤯🤯
Enormously entertaining talk by @GaryMarcus who could go do standup comedy if book tour fails. But good points that deep learning is not good solution if outlier detection is important, not good at surmounting bias, & needs augmenting for experimentation. ML failure modes 😂🤣👇
“Until we have deep comprehension as well as deep learning, we won’t have AI that we can trust”
McKinsey analysis - value of AI market $3.5T to $5.8T. Must be reasonable error bars on that prediction... #MKT4INTEL
Solid advice from McKinsey that transformation of business to using AI to solve problems really requires top down education and change at most senior executive mgmt level. The scale of the change in approach required means it’s not going to trickle up from one team.
I’m going to shamelessly steal all of mckinsey’s data for every business case I write at @twilio, and you should too.
If we were able to label the data of audience participation in this room with “took photo of slide” and “has business owner job” there is an enormous spike right now on that intersection during McKinsey talk. Great great data and lessons learned. #MKT4INTEL
“50% of the effort and cost in scaling AI is in operating model change. Don’t underestimate it”
“95% of all human requests to alibaba are handled by a bot” - Michael Evans, President alibaba group 🤯
“We think chatbots save us 160,000 hours a day” - alibaba. Of course at that ROI and scale, alibaba are able to invest huge huge resources into developing / scripting their bots to get to that level of coverage. I doubt many other companies are anywhere close to 95%.
The scale that alibaba operates at is insane. But it also means their ROI calculations on AI efforts are not going to be the same as most companies. Their scale mean essentially anything plausibly feasible w a marginal value would be worth investing at any cost. #mkt4intel
Uber self driving adapts driving based on turn signals of other drivers. But it wasn’t taught that turn signals indicate turning intention. It just learned it (and that sometimes people are just listening to music with their signal left on)
“Toronto has two seasons. Winter and construction season” 😂
Next presentation up... @suzannegildert asking “could *all* work be automated”. She thinks yes, in 1 product. mapped every occupation to abilities skills and knowledge required. Building humanoid robot to take on every primitive task in database. #yanggang #mkt4intel
According to her, classifying automation of *workers* not work, assuming incremental improvement of each primitive task, no jobs replaced until 2024. But by 2027 the wage bill replaced by automation in US is $7.65T. Interesting - but seems optimistic timeline #mkt4intel #yanggang
Not exactly diversity dimension but pleased to hear so many British accents here :)
Jack Clark from OpenAI talking about how scarily good AI is getting at generating entirely credible, well written but totally Bullshit made up content, with GPT-2. Take a look at this image - with given human generated input at top, the entire rest is artificially generated 😱🤯
Various interesting ethics discussions about AI. but I think the real ethical dilemma is how do we still allow 3D-ball style bullet points to be a thing? #mkt4intel
Fascinating discussion from Jim Keller @ intel ranging from electron leakage to efficiency of cat gif searching. All numbers at Intel are mindblowing. He has 10,000 people on his team. But I like this chart.
That’s it. Shout out to @shivon for being a great moderator throughout.
