🔮 Robots, jobs, education; Comma.AI vs Tesla; rules of self-driving; iPhone 8; death of liberalism; stem cells beat paralysis; eagles, dolphins & Berlin++ #79
Robots, jobs and opportunities in employment and education. Self-driving cars and rules of the road. The death of liberalism. The climate tipping point brings a decline in oil investment and growth in solar. What will the iPhone 8 be made of?
Hope this stimulates some great conversation this week ;)
SUPPORT EV: Donate via PayPal. (You choose the level.)
Very helpful if you 👍🏽 Like us on Facebook | Recommend on Twitter
Dept of the near future
⚖ Which algorithm do you want? Tim O'Reilly: Whose black box can we trust? MUST READ
👷 Goldman Sachs on how to manage automation of employment. “Closing the jobs gap requires a new approach to risk-sharing, one that spreads the burden of investing in human capital more broadly. This risk-sharing approach should include a greater educational focus on social skills, creativity and judgment, not only STEM subjects; …innovative financing structures to support investments in human capital and career transitions; lower barriers to entry into certain professions; increased support for small-business creation; and regulation that supports the growth of the ‘freelance economy.’” Read the summary if you are in a hurry. LONG, EXCELLENT. (Alongside, Forrester Research estimates 6% of US jobs will be lost to robotisation within five years.)
🎓 The $200bn opportunity: Sebastian Thurn, founder of Udacity, estimates the demand for engineering talent for autonomous vehicles is about $10m per head. WOW (Separately, employment for eagles as anti-drone police seems to be on the rise, video at 1min40s)
🔥 Approaching the first climate tipping point: on track for 1.5C by 2035. xkcd’s fantastic cartoon also pictures it. (Elsewhere, oil industry investment will decline 40% in the 2-year period to 2016; alongside an 85% increase in US solar capacity to 2016. The divergence in oil vs solar investment is comforting.) SCORCHING
🗽 On the death of liberalism: “economic liberalism of the 80s … has hit rough water. So too has the socio-cultural liberalism of the 60s, with its belief in universal rights and the meaninglessness of difference.” THOUGHT PROVOKING
💰 16 definitions on the economics of venture capital. VERY GOOD introduction to the funding sources, strategies and management of VC firms by @scottkupor
Ich bin ein Berliner
🇩🇪 I will be in Berlin next month. The Blueyard Capital team and I are hosting a meetup on Humans and Algorithms. October 6th, you’ll find details here. It would be great to see you!
Also, I’m interested in meeting Berlin entrepreneurs building in natural language understanding, machine vision or unsupervised learning. LMK.
Dept of autonomous transport
German transport minister, Alexander Dobrindt, outlines three likely principles for self-driving cars. The translation isn’t great, I am guessing the first principle should read “damage to people takes precedence over damage to property.” He suggests manufacturers will ultimately be liable, an idea we have covered in EV previously. Of course, until the auto industry secured jaywalking as a crime in the 1920s, manufacturers, not just drivers, were often liable for injury.
💡 Well-known hacker GeoHot gives a breathless pitch for Comma.AI. (Video, first five mins is pitch.) Comma makes a $999 add-on to existing cars that gives them autonomous driving capability. Interesting to note GeoHot has already crowdsourced 300,000 miles of vehicle training information. That is still a far-cry from Tesla’s fleet which has generated an estimated one billion miles of data when I last looked.
Geohot argues that because Comma.AI gets video data it has qualitatively better than the data Tesla has. (Tesla updated its autopilot this week and gave some nice details on fleet learning.)
“Annotating real-world data is an expensive operation and the current approaches do not scale up easily.” This is why autonomous vehicles might be taught how to drive in simulators like Grand Theft Auto. Although I am sceptical of this approach because of crowdsourcing techniques discussed above.
What it is like to ride in Uber’s self-driving cabs.
Meanwhile, Google’s self-driving project seems to be stumbling.
Elsewhere in AI
A generative adversarial network can identify faces from pixelated photos.
🐮 Google Maps protects a cows identity by pixelating its face.
Kate Crawford: We need a new measure to work out when AI is working or not. Worth mulling over.
Technical details on NVIDIA’s new neural network inferencing cards
Short morsels to appear smart at dinner parties
The Neural Network Zoo: great visualisation.
😀 Stem cells have restored the use of arms in a paralysed man
Brain sensing transcribes intent at 12 words a minute.
How well does brain structure predict behaviour? Very well for Alzheimers, less so for Parkinsons.
😠 How the sugar industry paid scientists to play down links with heart disease. MUST READ
🍆 Sex toys and the internet of things. Swipe left.
The Philippines:South Korean remittance corridor is now apparently 20% in bitcoin. Signs of life!
🐬 Dolphins recorded having a conversation with each other.
💥 Hillbilly Elegy: Review of JD Vance’s relevant memoir on Redneck man
Will the iPhone 8 be made out of advanced ceramics? ZrO2
Reader's corner
I came across an excellent paper on Accountable Algorithms in the University of Pennsylvania Legal Review (h/t @milesbrundage). Once summarised & critiqued, it will make a great read for Exponential View subscribers. If someone is prepared to take that task on, I’ll be happy to feature it in the newsletter.
Regular EV reader Gerd Leonhard asks how we find a sustainable balance between technology and humanity in his new book.
EV reader Rodolfo Rosini, founder of Weave.AI, is featured in the Financial Times. (Disclosure: I advise Rodolfo on stuff from time-to-time).
End note
Thanks to readers who pointed out a couple of horrible howlers in last week’s issue. Yes, Hobson’s choice is no choice at all. Facebook doesn’t face a Hobson’s choice over editorialisation, it faces a difficult one and it can choose either route. Apologies for manhandling the language.