Nexttech

Nexttech
Creating Generational Legacies

Sunday, March 11, 2018

AI at SCSW was VG while I was at BBG in the offices of BSI - it was VG

Learnings from the AI session at #sxsw - hosted by Loïc Le Meur - at leade.rs

Loic just hosted a session at South by Southwest on Artificial Intelligence with fantastic speakers.

Adam Cheyer — co-founder of Siri now building “the Siri of Samsung” as his new company got recently acquired

Daphne Koller — co-founder of Coursera, McArthur Genius award, Stanford computer science professor

Nell Watson — entrepreneur, engineer who wrote two books on philosophy and focusing on machine ethics

A few learnings from Loïc from the session

Siri doesn’t understand my “frenglish”

The highlight for me was definitely when I complained to Adam that Siri never understood my accent and he replied “I never understand your accent either what do you expect”. The entire room laughed (photo). So good.

How can we be scared by the creation of a Terminator like AI+robot when we can’t even tell Alexa I have eggplant and parmesan in the fridge give me a recipe? That doesn’t work well yet while Google or YouTube will just return a good recipe of… eggplant parmesan.

I also learned that the top two questions asked to Siri according to Adam are

1. will you marry me?

2. do you love me?

Humans apparently really need love from machines!

The Terminator is a 1,000 years away

Adam Cheyer said we might be hundreds of years away, maybe 1,000 years away of any AI exceeding human intelligence. For him there is absolutely no reason to have any fears of an AI Terminator. He said that Elon Musk’s Open AI to work on a “safe AI” with up to $1 billion funding committed is a worthy exercise but the near threat isn’t real.


What is human intelligence?

How far did we understand our own intelligence, what makes it different to machines?


Daphne Koller told us that this was a totally irrelevant question to understand or create artificial intelligence. Point. You do not have to understand human intelligence to create AI. It goes even further for Adam, it is better to NOT teach human intelligence to a machine to design an AI. Google/Deepmind managed to repeatedly defeat the best Go player in the world by letting the machine teach itself how to play without any knowledge of how a human would play.


What makes human intelligence different is the capacity to learn from a single experience. A kid burning himself touching a flame for example. A machine needs many repeat experiences to learn, we only need one.

The line between reality and AI created fiction blurs


There are now videos of Obama that are generated by machines making him say things he did not in a remarkably “real” way. It is so well done that humans cannot figure what’s true or not and it will only accelerate. We should be concerned that fake news will look more real than reality now. Machines also learn from news and social feeds and deliver ultra targeted ads. It is easy to imagine how the story of Russian Facebook ads during the Trump campaign is only a start. We won't tell the difference between what's real and fiction anymore.


Loading human values into machines


Nell Watson made really interesting points on how we need to teach machines human ethics so that they do not destroy us. I wasn’t convinced how to be honest but then this was a very short panel. I want to understand it better. If a machine learns from the human atrocities it can find on the internet it will be difficult to teach it “human ethics”. We don’t even agree on these ethics ourselves.


What are the 8 million americans making a living from driving a car or a truck going to do when they lose their jobs?


I got the same answer as always. The agriculture and industrial revolutions also destroyed millions of jobs and we figured it out. Nell added that in the near future she believes that millions of people will just be directly employed by machines. Yes, she believes we will more an more work for machines managing businesses on their own without human management.


Are we going to remain the most intelligent specie on the planet in the future?


The AI experts all agreed that they would have never thought any of the major AI achievements in the last decade would have happened. They urged us to focus on the positive and the incredible applications starting to show results in biology, agriculture or energy. Curing major diseases with AI and machine learning seems to be the most promising for the future of humanity.



No comments:

Post a Comment