author profile pic

By Eleanor Bennett

Interview

5 min read

In the latest instalment of our interviews speaking to leaders throughout the world of tech, we’ve welcomed Alberto Rizzoli, CEO and co-founder of V7, an AI computer vision platform.

Alberto holds a degree in Management Science from Cass Business School and attended Singularity University’s Global Solutions program in 2015, where he founded Aipoly— an AI-driven app for blind and visually impaired people.

Tell us about the business you represent, what is their vision & goals?

V7 is a training data platform to automate any visual task with AI. Users load images or videos onto the platform, identify objects using their mouse, and V7 learns to automate this identification as they complete tasks, saving up to 90% of the time needed to create training data.

The AI industry relies on training data to function - all machine learning models learn to replicate the training samples they have previously seen when given a new challenge. At V7 we handle the training data of hundreds of AI companies and enterprises solving anything from smarter cancer screening, to robots that learn to harvest fruit, or drones that spot cracks in bridges automatically.

In 10 years, we believe every business in the world of software will need a platform like V7 to power its AI components.

Can you share a little bit about yourself and how you got into the field of artificial intelligence?

My co-founder Simon Edwardson and I were fascinated by a 2015 paper by Google researchers showcasing how modern neural networks could describe images with sentences.

For a computer scientist, this is wizardry. Where are the rules? What determines a sentence or another? It’s all in one big neural network that ingests lots of images and learns small parts of it, like a hologram.

We were hooked.

In our first startup together, Aipoly, we wrote the first engine to run convolutional neural networks on smartphone CPUs to help blind and visually impaired people identify over 5,000 objects in real-time, and listen to their descriptions . In 2018 we founded V7 to bring computer vision to every industry, and become a platform that ML engineers could use to turn the tech world from one running on software to one running on neural nets.

What do your day-to-day responsibilities look like at your organization?

9:30 - Team standups 10:00 - 4:30 - sales calls, demos, interviews, and anything that faces outwards. 4:30 - 7:00 - internal syncs, planning, and anything that faces inwards 7pm - late - Actual work, this is where you build stuff with the remaining time you have!

Once you reach a headcount of 15 or so your responsibility starts shifting into empowering your team members to execute.

I still get the time to craft something on my own - I currently take care of all of the design at V7, but it’s a disservice to my company to spend too much time on it. The best managers are those who really want to build stuff but can’t because they have to manage.

What are the key differences between computer science, machine learning, and AI?

Computer science deals with the deterministic nature of code and data. There’s much more to it, but in this comparative context, we can leave it at that.

Machine learning is when data is used to make conclusions about other data, using an algorithm, often run in the form of code.

AI, or at least the dark arts of deep learning, uses neural networks instead of code - these bundles of interconnected features are not deterministic but probabilistic, as there are far too many parameters to tell how they think. Instead, they model huge amounts of data to make a conclusion.

Think about your sense of smell - you can’t really explain how chocolate smells, but you can tell something smells like chocolate from experience. That’s deep learning. Hearing is a lot more like code - you can break things down into notes and pitches to explain how a word sounds.

What are some misconceptions that you believe the average person has about AI?

AI’s progress is led by a few thousand researchers; everyone else re-implements their work. Large tech companies went all-in on AI because AI-first versions of their products are the biggest threats for their existence, and continue to be. AI is far more about training data than algorithms. Neural network architectures have much less of an impact on AI performance than training data does. Nobody in the technical AI community uses the word “AI” with each other. They’ll refer to the specific branch they work with, like computer vision, or NLP, and leave the two-letter word to marketers. The biggest danger in AI is how poorly human decision-makers understand how it works.

What advice would you give to someone wishing to start their career in artificial intelligence?

Become good at one subfield, and narrow it down, much of it is transferrable. Don’t waste your time with classical machine learning if you’re looking for a job. If you think you’ve figured it out, you haven’t. Stop thinking of data as a big table. Atoms are data, cake batter is data, and everything can be learned with the right approach.

Would you like to share any artificial intelligence forecasts or predictions of your own with our readers?

Hundreds of startups will emerge in the next 5 years through products that mainly rely on AI models to solve conventional tasks, but better.

AI will start becoming more boring in 5 years as it cements itself as a de-facto branch of computer science, and we will all rely on generative models to produce creative work. Robots will take longer than we think to roam the earth, and AI applied to biology will be the true paradigm shift of the 21st century. There might be a silicon vs carbon debate in 10 years to settle.

What is your experience of using AI-backed data analysis or log management tools? What do you think is the benefit of using a log management tool that has machine learning capabilities for an organization?

Deep learning is still an emerging field, which means it comes with several emerging bugs. Log management tools can help keep order among hundreds of parallel processes. ML in particular can be useful to help document errors and provide descriptive responses to users based on previous resolutions of the same error.

Are there any books, blogs, or any other resources that you highly recommend on the subject of AI?

Play around at Distill.pub until you’re confused enough to want to embark on a few hours of studying.

If you enjoyed this post and want to keep reading our best articles then why not check out our list of software deployment tools or our resource comparing different Kibana dashboards?

Get the latest elastic Stack & logging resources when you subscribe

backReturn to Blog

© 2022 Logit.io Ltd, All rights reserved.