Jul 16, 2017 · 46m · a16z
The Promise of AI
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this Andreessen Horowitz presentation, Frank Chen illustrates how artificial intelligence makes complex computational tasks inexpensive and accessible, driving widespread transformation across six major commercial categories in a pattern echoing the historical adoption of relational databases.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 0.0 Guest teaching 6.6 Guest disagreement 0.0 The host pushing back 0.0
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 34:35 Human Limitations in High-Dimensional Thinking
In a non-combative solo presentation, this moment represents the guest's strongest critique of human cognitive limits, pointing out that human minds give up beyond three or four dimensions.
Hardest push from the host ▶ 0:28 Monologue Presentation - Zero Host PushbackThe host does not speak or offer any pushback during this monologue presentation.
Biggest teaching moment ▶ 34:00 Google DeepMind Data Center Cooling OptimizationChen educates listeners by detailing how DeepMind analyzed 120 variables in data centers to cut cooling power costs by 20-25%.
The host holds their own ▶ 0:28 Monologue Presentation - Zero Host Hits BackBecause the host is silent throughout the monologue recording, no host expertise was demonstrated.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Historical Analogy: The Relational Database Paradigm Shift | 0 | 6 | 0 | 0 | Because this segment is a solo monologue presentation by guest Frank Chen, host-side scores are zero. Chen provides an educational overview comparing AI ubiquity to the historical rise of EF Codd's relational database and details autonomous vehicle deployment like Zipline blood delivery. | |
| Category 2: Computer Vision, Recognition, and Environmental Sensing | 0 | 7 | 0 | 0 | In this monologue segment, Chen educates the audience on computer vision advancements, breaking down Generative Adversarial Networks, ImageNet error rates, and practical applications like TensorFlow cucumber sorting. | |
| Category 3: Automated Content Generation Across Media Types | 0 | 7 | 0 | 0 | Chen delivers a solo breakdown on AI content generation across text, photorealistic image synthesis, music generation, and IBM Watson movie trailer curation. Host scores remain zero due to the monologue format. | |
| Category 4: Predictive Analytics, Biometrics, and Health Diagnostics | 0 | 7 | 0 | 0 | Chen presents case studies on predictive biometrics and health diagnostics, including gait recognition with Unify ID and cell-free DNA cancer detection with Freenome. Host participation is non-existent. | |
| Category 5: Automated Optimization of Complex High-Dimensional Systems | 0 | 7 | 0 | 0 | Chen explains high-dimensional system optimization, highlighting Google DeepMind's reduction of data center cooling electricity by managing 120 variables simultaneously. | |
| Category 6: Natural Language Understanding and Human Interaction | 0 | 7 | 0 | 0 | The guest details natural language breakthroughs such as Google Smart Reply and Stanford crisis counseling text analysis for suicide prevention. | |
| Conclusion and Strategic Action Plan for Organizations | 0 | 5 | 0 | 0 | Chen concludes his monologue with a three-part strategic framework for organizational adoption of AI tools and training. |
Statements from this episode (25)
Frank Chen: AI will be integrated into every important piece of software
“And I think AI is going to be like that as well, which is it's going to get into every important piece of software.”
Chen: Otto demonstrated autonomous commercial beer delivery in October 2016
“When it becomes cheap to drive, you'll have beer drive itself from the brewery to the store, which our friends at Auto already showed in October of 2016.”
Chen: Zipline's medical delivery drones can fly 90-mile round trips
“The blood gets loaded up at the collection center, put onto a zipline drone, which can fly up to 90 miles round trip.”
Chen: Non-autonomous vehicles will soon seem odd to future generations
“The technology will become pervasive and it will seem odd to our children or our grandchildren that something moves but can't move itself and can't get itself from point A to point B all by itself.”
Frank Chen: AI image classifiers are now more accurate than humans
“In fact, the most accurate image classifiers, as these are called, are basically better than humans at classifying the objects in a picture.”
Frank Chen: Tens of thousands of niche vision AI apps will emerge
“So I love this very inexpensive, super creative use of AI, and I think we're going to see 1010 of thousands of applications like this over the next decades.”
Frank Chen: Vision AI will shift farming to plant-by-plant fertilization
“And so when AI makes it cheap for us to see and understand what's going on in the world, we won't fertilize fields anymore. We will fertilize individual heads of lettuce.”
Chen: Washington Post and Toutiao used AI to write Olympic sports coverage
“During the last Olympics, two professional journalism outfits, one, the Washington Post, and then another, the Chinese news aggregator called Toutiao, both ran experiments where they had AIs write coverage about sporting events in the Olympics.”
Chen: AI algorithms can generate written recipes by watching cooking videos
“Algorithms are now able to watch the exact same videos that you're watching on YouTube or Facebook, and they're basically able to retroactively create the cookbook instructions for that recipe.”
Chen has met roughly six startups building real-time AI music generation
“I've already seen maybe half a dozen startups that are working on this very problem that will be able to generate music on the fly”
Chen: IBM Watson selected all scene edits for the 'Morgan' trailer
“So this trailer was created by IBM Watson, and it selected all of the scenes that you saw in the trailer.”
Chen: Microsoft DeepCoder builds software by remixing code samples
“Microsoft has a system called deep coder that creates software in much the same way that human developers do, which is they go find some sample code and they sort of remix it for their purposes.”
Chen: Biometric algorithms will eventually replace passwords entirely
“We all know that eventually the algorithms will get so good at using biometric factors like the way you walk or the way you type or the way you swipe and predict that you are you and I can't wait for the passwordless future.”
Chen: Freenome is developing a cell-free DNA blood test for cancer diagnosis
“The company Freenome is working on a cancer diagnostic by reading the DNA that's free floating in your bloodstream.”
Chen: Cardiogram uses Apple Watch data to detect abnormal cardiac events
“Using the data that comes off in Apple Watch, they can predict whether you're having one of these abnormal cardiac events.”
Chen: AI compiler optimization yields 1.6x execution speedup
“Not only is the instruction set much, much shorter, but the resulting code runs 1.6 times faster than the code that's spit out by the compiler today.”
Chen: Google DeepMind cut data center power use by 20 to 25%
“All of these variables fed into a system, and they were able to take 20 to 25% of the electricity out of the equation, serving the exact same workload.”
Chen: Humans fail at optimization beyond four dimensions while AI excels
“And after about three or four dimensions, your mind just kind of gives up. Your brain isn't programmed or optimized for that type of mathematical optimization, but machine learning algorithms love lots of data and are able to do this in a way that human brains…”
Chen: Instacart cut order fulfillment time by 8% via machine learning
“So using a variety of machine learning techniques, they were able to shave eight percent of the time that it took for a shopper to get through a supermarket and to your house.”
Chen: Voice input allows users to message three times faster than typing
“Accuracy rates have gotten so good that people who talk to their phones can communicate messages three times faster than typing.”
Chen: Google Smart Reply generated 10% of mobile Inbox replies in 2017
“In February, when I heard Jeff Dean talking about it, he pointed out that Smart Reply was generating 10% of all mobile inbox replies.”
Chen: Emotional understanding and simulation will become essential to AI UX
“As AI systems get increasingly sophisticated at understanding us, I think one of the things that will become important from a user experience point of view is systems that understand our emotions and can simulate emotions in interacting with us.”
Chen: Real-time earpiece language translation is near commercial deployment
“We're very close to the time that we can just pop something into our ear and hear another language being translated in real time.”
Chen: AI models predict suicide risk years ahead with over 80% accuracy
“Study one basically looks through electronic health records and based on the data in electronic health records is able to predict people who are likely to commit suicide. And these people might be two or three years away from that event, but they are able to p…”