May 22, 2018 · 26m · mad

A New Approach to Machine Intelligence // Ben Vigoda, Gamalon (FirstMark's Data Driven)

Ben Vigoda · 18m spoken Matt Turck · 1m spoken Mark Sear · 33s spoken
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

At Data Driven NYC, Gamalon Founder and CEO Ben Vigoda presents a novel approach to natural language processing that replaces fragile deep learning models with editable, probabilistic Idea Trees capable of structuring enterprise text at scale.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 8% of the talking time here. How this is scored →

Matt as informed peer 0.9 Guest teaching 2.9 Guest disagreement 0.7 Matt pushing back 0.6
05100:0010:0020:000:29–3:02 · Matt as informed peer 0/10 Ben Vigoda's Background & Machine Learning Career Ben opens with a talk introducing his background in neural networks since 1989 and outlines the enterprise challenge of structured natural language understanding. Because this is a monologue presentation, host activity scores are zero.3:02–5:32 · Matt as informed peer 0/10 Flaws in Deep Learning & Industry Criticism Ben criticizes modern deep learning, calling it Pavlovian conditioning that produces uninterpretable black boxes and requires manual, wide matrix data labeling. The host is not involved during this monologue segment.5:32–8:34 · Matt as informed peer 0/10 A New Vision for Natural Language Machine Learning Ben outlines his vision for probabilistic programs over simple neurons, emphasizing explicitly tracked uncertainty error bars and interactive models. Host scores remain zero as the talk continues uninterrupted.8:34–10:59 · Matt as informed peer 0/10 Live Demonstration: Gamalon UI & Idea Trees Ben demonstrates the Gamalon UI showing live idea trees classifying credit card queries and managing atomic ambiguity thresholds. The presenter speaks exclusively to the audience without host participation.10:59–14:14 · Matt as informed peer 0/10 Case Study: Fixing Voice Assistant Failures Ben highlights voice assistant failures like Alexa getting stuck when users change their minds and showcases how Gamalon enables direct copying and pasting of subtrees. Host engagement is absent.14:14–16:41 · Matt as informed peer 0/10 Unsupervised Idea Tree Learning from Raw Text Ben displays unsupervised idea tree generation from unlabelled raw text streams, emphasizing major speed and cost benefits for enterprise customers. Host interaction remains zero during the live demo.16:41–26:26 · Matt as informed peer 6/10 Enterprise Dashboard Analytics & Presentation Conclusion Host Matt Turck joins to ask informed questions about AI historical waves, probabilistic programming, and potential trade-offs like compute demands and ontology setup time. Ben responds constructively while taking additional audience questions regarding grammar rules and shared subtrees.0:29–3:02 · Guest teaching 2/10 Ben Vigoda's Background & Machine Learning Career Ben opens with a talk introducing his background in neural networks since 1989 and outlines the enterprise challenge of structured natural language understanding. Because this is a monologue presentation, host activity scores are zero.3:02–5:32 · Guest teaching 3/10 Flaws in Deep Learning & Industry Criticism Ben criticizes modern deep learning, calling it Pavlovian conditioning that produces uninterpretable black boxes and requires manual, wide matrix data labeling. The host is not involved during this monologue segment.5:32–8:34 · Guest teaching 3/10 A New Vision for Natural Language Machine Learning Ben outlines his vision for probabilistic programs over simple neurons, emphasizing explicitly tracked uncertainty error bars and interactive models. Host scores remain zero as the talk continues uninterrupted.8:34–10:59 · Guest teaching 2/10 Live Demonstration: Gamalon UI & Idea Trees Ben demonstrates the Gamalon UI showing live idea trees classifying credit card queries and managing atomic ambiguity thresholds. The presenter speaks exclusively to the audience without host participation.10:59–14:14 · Guest teaching 2/10 Case Study: Fixing Voice Assistant Failures Ben highlights voice assistant failures like Alexa getting stuck when users change their minds and showcases how Gamalon enables direct copying and pasting of subtrees. Host engagement is absent.14:14–16:41 · Guest teaching 3/10 Unsupervised Idea Tree Learning from Raw Text Ben displays unsupervised idea tree generation from unlabelled raw text streams, emphasizing major speed and cost benefits for enterprise customers. Host interaction remains zero during the live demo.16:41–26:26 · Guest teaching 5/10 Enterprise Dashboard Analytics & Presentation Conclusion Host Matt Turck joins to ask informed questions about AI historical waves, probabilistic programming, and potential trade-offs like compute demands and ontology setup time. Ben responds constructively while taking additional audience questions regarding grammar rules and shared subtrees.0:29–3:02 · Guest disagreement 0/10 Ben Vigoda's Background & Machine Learning Career Ben opens with a talk introducing his background in neural networks since 1989 and outlines the enterprise challenge of structured natural language understanding. Because this is a monologue presentation, host activity scores are zero.3:02–5:32 · Guest disagreement 2/10 Flaws in Deep Learning & Industry Criticism Ben criticizes modern deep learning, calling it Pavlovian conditioning that produces uninterpretable black boxes and requires manual, wide matrix data labeling. The host is not involved during this monologue segment.5:32–8:34 · Guest disagreement 1/10 A New Vision for Natural Language Machine Learning Ben outlines his vision for probabilistic programs over simple neurons, emphasizing explicitly tracked uncertainty error bars and interactive models. Host scores remain zero as the talk continues uninterrupted.8:34–10:59 · Guest disagreement 0/10 Live Demonstration: Gamalon UI & Idea Trees Ben demonstrates the Gamalon UI showing live idea trees classifying credit card queries and managing atomic ambiguity thresholds. The presenter speaks exclusively to the audience without host participation.10:59–14:14 · Guest disagreement 1/10 Case Study: Fixing Voice Assistant Failures Ben highlights voice assistant failures like Alexa getting stuck when users change their minds and showcases how Gamalon enables direct copying and pasting of subtrees. Host engagement is absent.14:14–16:41 · Guest disagreement 0/10 Unsupervised Idea Tree Learning from Raw Text Ben displays unsupervised idea tree generation from unlabelled raw text streams, emphasizing major speed and cost benefits for enterprise customers. Host interaction remains zero during the live demo.16:41–26:26 · Guest disagreement 1/10 Enterprise Dashboard Analytics & Presentation Conclusion Host Matt Turck joins to ask informed questions about AI historical waves, probabilistic programming, and potential trade-offs like compute demands and ontology setup time. Ben responds constructively while taking additional audience questions regarding grammar rules and shared subtrees.0:29–3:02 · Matt pushing back 0/10 Ben Vigoda's Background & Machine Learning Career Ben opens with a talk introducing his background in neural networks since 1989 and outlines the enterprise challenge of structured natural language understanding. Because this is a monologue presentation, host activity scores are zero.3:02–5:32 · Matt pushing back 0/10 Flaws in Deep Learning & Industry Criticism Ben criticizes modern deep learning, calling it Pavlovian conditioning that produces uninterpretable black boxes and requires manual, wide matrix data labeling. The host is not involved during this monologue segment.5:32–8:34 · Matt pushing back 0/10 A New Vision for Natural Language Machine Learning Ben outlines his vision for probabilistic programs over simple neurons, emphasizing explicitly tracked uncertainty error bars and interactive models. Host scores remain zero as the talk continues uninterrupted.8:34–10:59 · Matt pushing back 0/10 Live Demonstration: Gamalon UI & Idea Trees Ben demonstrates the Gamalon UI showing live idea trees classifying credit card queries and managing atomic ambiguity thresholds. The presenter speaks exclusively to the audience without host participation.10:59–14:14 · Matt pushing back 0/10 Case Study: Fixing Voice Assistant Failures Ben highlights voice assistant failures like Alexa getting stuck when users change their minds and showcases how Gamalon enables direct copying and pasting of subtrees. Host engagement is absent.14:14–16:41 · Matt pushing back 0/10 Unsupervised Idea Tree Learning from Raw Text Ben displays unsupervised idea tree generation from unlabelled raw text streams, emphasizing major speed and cost benefits for enterprise customers. Host interaction remains zero during the live demo.16:41–26:26 · Matt pushing back 4/10 Enterprise Dashboard Analytics & Presentation Conclusion Host Matt Turck joins to ask informed questions about AI historical waves, probabilistic programming, and potential trade-offs like compute demands and ontology setup time. Ben responds constructively while taking additional audience questions regarding grammar rules and shared subtrees.

speaking balance: gold is Matt, purple is the guest (3 minute bins)

0:00 · Matt 0% · guest 100%0:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%3:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%6:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%9:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%12:00 · Matt 0% · guest 100%15:00 · Matt 0% · guest 100%15:00 · Matt 0% · guest 100%18:00 · Matt 52.8% · guest 47.2%18:00 · Matt 52.8% · guest 47.2%21:00 · Matt 16% · guest 84%21:00 · Matt 16% · guest 84%24:00 · Matt 2.2% · guest 97.8%24:00 · Matt 2.2% · guest 97.8%
Sharpest disagreement ▶ 3:50 Ben calling deep learning Pavlovian conditioning

Ben forcefully critiques the dominant paradigm in machine learning, describing deep learning as crude Pavlovian conditioning that creates opaque black box models.

Hardest push from Matt ▶ 22:02 Matt Turck questioning trade-offs and ontology time

Matt Turck challenges the pitch by asking directly whether high compute demands and upfront ontology creation time represent major practical trade-offs.

Biggest teaching moment ▶ 21:01 Ben explaining Bayesian communications theory in cell chips

Ben educates the host and audience by explaining how Qualcomm cell phone chips use Bayesian communications theory to prevent dropped calls, relating it to human conversation.

Matt holds his own ▶ 18:02 Matt Turck placing the AI wave in historical context

Matt Turck demonstrates strong technical knowledge by contextualizing the deep learning wave's historical origins from previous decades and contrasting research pivots with industry adoption.

the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Ben Vigoda's Background & Machine Learning Career 0200 Ben opens with a talk introducing his background in neural networks since 1989 and outlines the enterprise challenge of structured natural language understanding. Because this is a monologue presentation, host activity scores are zero.
Flaws in Deep Learning & Industry Criticism 0320 Ben criticizes modern deep learning, calling it Pavlovian conditioning that produces uninterpretable black boxes and requires manual, wide matrix data labeling. The host is not involved during this monologue segment.
A New Vision for Natural Language Machine Learning 0310 Ben outlines his vision for probabilistic programs over simple neurons, emphasizing explicitly tracked uncertainty error bars and interactive models. Host scores remain zero as the talk continues uninterrupted.
Live Demonstration: Gamalon UI & Idea Trees 0200 Ben demonstrates the Gamalon UI showing live idea trees classifying credit card queries and managing atomic ambiguity thresholds. The presenter speaks exclusively to the audience without host participation.
Case Study: Fixing Voice Assistant Failures 0210 Ben highlights voice assistant failures like Alexa getting stuck when users change their minds and showcases how Gamalon enables direct copying and pasting of subtrees. Host engagement is absent.
Unsupervised Idea Tree Learning from Raw Text 0300 Ben displays unsupervised idea tree generation from unlabelled raw text streams, emphasizing major speed and cost benefits for enterprise customers. Host interaction remains zero during the live demo.
Enterprise Dashboard Analytics & Presentation Conclusion 6514 Host Matt Turck joins to ask informed questions about AI historical waves, probabilistic programming, and potential trade-offs like compute demands and ontology setup time. Ben responds constructively while taking additional audience questions regarding grammar rules and shared subtrees.

Statements from this episode (8)

Assertion Not publicly verifiable
Ben Vigoda began working on deep learning in 1989 at age 14
“I started doing deep learning in about 1989 with David Rumelhart when I was 14 working at Stanford.”
Ben Vigoda May 22, 2018 ▶ 0:39
Assertion Not checkable as stated
A major New York bank receives 1.2 billion text messages annually
“One big bank in, in New York City that we've been talking to gets 1.2 billion of these little text messages a year.”
Ben Vigoda May 22, 2018 ▶ 2:16
Assertion Not checkable as stated
Changing deep learning target categories requires retraining models from scratch
“But if you want to change your columns, your categories, then you have to redo all your multiple choice tests and then retrain the system. And if you want to change the categories, you have to start over.”
Ben Vigoda May 22, 2018 ▶ 4:38
Insight
Ben Vigoda: Neurons and synapses are the wrong abstraction for AI
“Neurons and synapses is not the right abstraction. Models should be programs, specifically programs that simulate the system that generated the data.”
Ben Vigoda May 22, 2018 ▶ 7:46
Insight
Ben Vigoda: Machine learning variables must include uncertainty and error bars
“Every variable that you're trying to infer in a program should come with an uncertainty. Neural networks today are just, they're just a number in the neuron. It's like an activation level. But you need error bars around those numbers.”
Ben Vigoda May 22, 2018 ▶ 8:03
Assertion Not checkable as stated
Gamalon reduced an automaker's text processing costs from $1.25 to 10 cents
“We're able to work with them for a few weeks and generate a model, and they're actually, we're paying a dollar 25 per utterance to get them read. It took months to do it every year, and now it takes 25 milliseconds, and they pay 10 cents.”
Ben Vigoda May 22, 2018 ▶ 17:18
Insight
Ben Vigoda: Deep learning is excellent at instinct but poor at thought
“AI is really good at instinct. I mean deep learning, it's not so great at thought.”
Ben Vigoda May 22, 2018 ▶ 19:13
What-if
Cell phones would drop calls 1,000x more often without Bayesian probability bounds
“You would drop calls a thousand times more often if there were, if cell phone receivers didn't found uncertainty in one patient.”
Ben Vigoda May 22, 2018 ▶ 21:46
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.