Radical ventures’ organized a four-part Master class for the AI founders. Its aim is to provide key insights to AI researchers on how to venture into the AI startup space and build successful AI-driven products and services. More information about this can be found [HERE].
Pieter Abbeel, a professor at UC Berkley and a co-founder of Gradescope and Covariant, presented some interesting ideas on this topic.
Covariant applies Deep Reinforcement Learning and Imitation Learning to design robots that can be reliably used to solve real-world planning problems.
In this article we present a summary of this first session.
Getting started
You should be interested in solving the problem. This should be a self-drive. This is likely to help attract the like-minded who will make up your team.
While being unique has been severally repeated in many startup founding scripts, it is not an exception here. Value and uniqueness are key!
What’s the future like?
Ask questions about the future. Will we still have the same problem 10–20 years from now? Will the technology you are proposing increasingly lead to better ways of living?
We will be still searching for the next 20 or so years to come (Google/Internet) and we will still be interested in watching movies from home in 10+ years to come (Netflix) and probably crashing at strangers’ places at a cost in the next 20+ years (Airbnb). This Sanity check is key.
Will we still have the same problem 10–20+ years from now?
Building a team
If you are in research, you will likely be able to easily access other researchers, especially graduate students. These are the most qualified members of your team. The founding members of Covariant were graduate students.
Reach out!
Pull up your email contact list and send out an invite. Perhaps one or two would be interested in the same problem your startup is trying to address.
Pieter did the same thing while founding Covariant and two people ended up showed interest. They later became co-founders. This echoes the sentiments about “knowing people in your space”
Its time to jump In
If you have been thinking about building an AI-driven startup then this is the right time to do so. AI research has active researchers and so many “reachable” like-minded people.
AI technologies can be disruptive — solve problems in a way it has never been solved before — if well engineered, akin to the “Uber in taxi business” or the next Airbnb or Lyft. Some industries are stuck and are just waiting for AI. Jump in then the rules will follow!
Define the role
This is important. What is the role of every team member? The CEO should be able to make decisions which work fast. The CTO should also be in a position to make decisions about things which will work.
You need to know your team, including yourself, to be able to settle on these roles. However, it is important to “not think forever”.
The CEO should be able to make decisions which work fast
It’s gonna take some time
You should have a long-time excitement. Creating a reliable AI product can take time. You need to collect data, validate, run simulations, implement simulations in the real world, gather feedback and iterate again. You should be willing to spend at least seven years for a successful product design.
90+ accuracy is not good enough
If a robot has a 90% accuracy in picking and placing objects then it means it would misplace approximately 1 in 100.
With many robots working in parallel and sheer volume objects, the number of misplacement would start increasing slowly but significantly. This behavior can also be extended to other AI models.
Rolling out the first commercially viable product at Covariant for instance took 4 years!
Veto power
Have some control. The new team members or co-founders may come with radical ideas which may throw the vision or mission or core values or even product off the edge. However, it is important to exercise your power indirectly [carefully & mindfully] without too much friction.
Team members are the ones working on the product and making product-level decisions every day. They are the builders.
Talk to other entrepreneurs
Every successful entrepreneur is always willing to help another one (upcoming) albeit if you’re a competitor. Write them an email — there is a good chance they will get back to you.
In addition, it is also important to follow up if you have not received any feedback. People can be busy yet still willing to help!
Every successful entrepreneur is always willing to help upcoming one
Customers
Watch customers using your product and be comfortable with the complaints raised. AI product design is also a Software Engineering Product. The Software Engineering rules still apply.
You need a product which has error handling ability and would restart after some time. End users always appreciate such use cases and such simple things will help you get it right.
Investors
If you cannot reach investors then reach out to people who reach them. As previously mentioned, established entrepreneurs are always willing to help new ones.
However, it is important not to lie because Investors always carry out their independent research and you do not want to get started with them on the wrong foot.
If you cannot reach investors then reach out to people who reach them
Good Luck you all! See you soon!
-Stephen.
