How to Start an AI Startup

Key Takeaways

hello world it's Suraj and how are you supposed to start an AI startup I'm gonna go over the steps you need to take to do this using a combination of my own startup experience and analyzing some best practices that some recently successful AI startups have implemented the public's interest in AI will continue to rise in the coming years and we're just at the start of a larger AI revolution innovative startups continue to pop up every single week tetra for example uses recent advances in speech recognition to generate detailed notes from your phone calls hyper science extracts data from forms easily using optical character recognition and jet lore uses consumer behavior as input to a model able to output structured data so it's 2018 and you want to get in on this your first step is to study AI you can't create an AI solution if you yourself don't understand how this amazing technology works right I've got some great free playlists for you to help you get started with this if you're just getting started watch them in this order learn Python for data science then intro the tensorflow then intro to deep learning and lastly the math of intelligence additionally Andrew Aang's newer deep learning that AI course is a must-have and Jeremy Howard's fast on AI course is excellent as well my favorite book on deep learning is the deep learning book by Ian Goodfellow it's free online and where it really excels is in helping you get to terms with the math of deep learning deep learning has proven to help perform every other ml model almost every time on a wide range of tasks so the hype is justified however you should really know how the other machine learning models work because there are use cases where you don't have access to a lot of data and you just need to make a simple prediction in those cases support vector machines or even a simple decision tree would be a more useful approach than a deep neural network the only way to know what model to use and when to use it is to develop an intuition and that comes after studying using a wide variety of sources all the ones I've talked about will be in the description once you've got a grasp of the landscape of algorithms that make up AI you've got to find a problem that you feel passionate about some of the most successful companies in the world were started because the founders were trying to solve a problem that they themselves faced write down a list of problems that you or your friends have faced in your personal lives that you would like solved these will very likely be good candidates for problems that you are passionate about solving since those affected are people that you very much care about if you can't think of any good ones you could also use social media to search proactively for people looking for a solution search in quotes for words and phrases that indicate that the poster is frustrated by something or looking to solve a problem like how do I or can't seem to this will work on reddit Twitter just about any community where a target market might congregate on but perhaps the best way that I found to find problems is to travel and meet people I couch surfed around Europe for three months a few years ago and happened to stay with Alex macaw in London he gave me the inspiration necessary to change my trajectory in my early years of university from economics to computer science one of the most impactful changes I've made in my life it's what got me to eventually start a robotics startup lucid robotics while I was a third-year student at Columbia and more recently travelling through South and Southeast Asia for four months helped put things into perspective for me I left the bubble of Silicon Valley and got to see problems firsthand that real people faced in their day to day lives problems related to infrastructure banking communication and of course education that was the one that really affected me simple conversations I had with locals compounded into later experience is that resulted in a larger plan to educate the world on AI by making my content the product as you own musk says being an entrepreneur is like eating glass and staring into the abyss of death if you don't have the passion to solve a problem you're not gonna stick around when the going gets tough no matter how many negative comments I've gotten I've always persevered because it's not about me it's about solving AI once you have a problem to solve it's time to do some market research who are you selling to where are they going to buy it how much would they pay what are the competing products or services in this space what's the cost to deploy what does the history of this market look like usually Gove's and big data analytics firms have publicly available reports on this generally AI startups are classed into two varieties there are the horizontal AI startups these are working on one fundamental problem that would serve many different industries like say general natural language processing then there's the vertical AI startups these are solving problems for a very specific type of customer belonging to one specific industry just about every major tech company is working really actively on AI they're able to hire all the AI rock stars from hinton to Laocoon they've got the talent they've got the massive amounts of data they've collected from users over the years which remains closed only to their teams these tech giants have a huge advantage when it comes to building horizontal products that can apply to many industries like image recognition or language translation or infrastructure but the advantage that you have is that you can move fast on a single problem vertically they don't have the time to tackle every single niche problem but you do and as a whole they're focused on the consumer more than they're focused on the enterprise so a niche solution that would help companies would be a great choice create a landing page that describes your product with a simple email signup to get notified of release and share it on social media one way to raise awareness of your product is to raise your own personal profile establish yourself as an AI thought leader create blog posts or other content that answers fundamental questions about AI build an audience see if you get any signups if you do especially after you've stated what the price of your product will be it means there's a demand for what you could be offering this is crucial I went through the whole process of raising money building a product and the team my last startup only to find that the eldercare market didn't want a $5,000 robot to help pick things up off the floor people with ALS would much prefer a simpler tool I was too obsessed with the solution not the problem if you remove AI from the company but it still has a valuable product you're on the right track but if AI is your only thing then neither the customer nor investors will be excited about it once you've done your market research it's time to build a product but before you build a model you've got to collect organize and label your data as much as you can the quality of the data is usually the most important part of the machine-learning pipeline even more so than the architecture of your model the phrase garbage in garbage out comes to mind the easy way to get data is to search for public datasets there's an awesome list of datasets on github the University of California Irvine has a giant ML repository of datasets and of course Kaggle has some great ones as well if those don't work another way is to create data yourself using an existing data set like if you're trying to classify handwritten characters you could generate new data points from existing ones by adding some noise via a distribution if you've got unlabeled data crowdsourcing via Amazon's Mechanical Turk is a great option you could pay people to label it for you there are also data marketplaces out there like data circle where you can buy or exchange datasets directly from other people lastly if feeling creative you can scrape the data yourself using Python in a library like scrap PI or a web service that does this for you like Digger not calm when it comes to building models tensorflow is still the best most battle-tested machine learning framework out there and there's a whole host of tools in the tensorflow ecosystem like tensorflow serving that make it production ready build a model train it on your data via a service like AWS Google Cloud or Floyd Hub use your model to make a prediction and this is your product more data means smarter algorithms means better products means more users means better data and the cycle repeats raising money is a way to accelerate the scaling process the ico route may be hot right now but if you're in it for the long term and don't want to mess around with legalities I'd avoid this for now a friend of mine who was the first designer for aetherium didn't do an ICO for his company balance he instead raised money using a crowdfunding tool called we funder his reasons were really inspiring and I've got an article on it in the description venture capital is of course another option and while it's easy to convince some VCS that your product is cool by just repeating a bunch of buzzwords the best ones will understand the technology well so be prepared to explain the technicalities of it to them you're always in a better position to talk to VCS when you've already got to cash flow than if you don't that's why I got rejected by so many VCS in New York I hadn't proven my robot would sell now I've got a few offers to invest in what I'm doing since I built an audience so I can afford to be picky but I'm not focused on raising money right now when it comes to hiring there's a limited supply of talent and it's distributed globally so it makes sense to hire globally look at the leaderboard on Kaggle the first hires you make are the most important I worked really hard to find the right fit to help me with editing my videos if you really love something it's hard to ever let go control of but if you find the right person it becomes much easier as Steve Jobs said the neatest thing that happens is when you get a core group of you know ten great people that it becomes self policing as to who they let into that group at some point you'll be able to exit whether that be through an aqua hire or by one of the big tech companies or an IPO and if all else fails there's always the Consulting route it's not as sexy as building a product but hey it puts bread on the table find a couple of friends and start a consulting firm get some initial clients and slowly get more impressive brands on board as you prove yourself I hope my tips helped please subscribe for more programming videos and for now I've got a startup to grow so thanks for watching