How does someone go from coding in their shed to creating a self-sustaining successful A.I startup?

Contrary to popular opinion, holding a PHD or being a ‘genius’ is not a pre-requisite. This article covers all the core learning resources, business strategies and best practices required to get an A.I company up and running. However it should go without saying, to maximise the chances of success, an extremely high pain threshold with an unyielding fanatic like commitment and frequent reality checks are essential.

The applications of A.I in business are extremely wide ranging, chances are if you have a business idea, you can apply A.I to it.

A.I applications are best illustrated with a few examples - Tetra has leveraged advances in speech recognition to generate detailed notes from your phone calls. Hyper science is automating remedial office work and is able to extract data from forms easily using optical character recognition. JetLore uses consumer behaviour as input to a model able to output structured data.

It would be an understatement to say there is plenty more to come, the list of new ways to leverage A.I continues to expand, from automating recycling to maximising crop yield. Lets dive into the principles and knowledge you need in order to make A.I work for you

It goes without saying you can’t create an effective A.I business if you don’t have a thorough understanding of the fundamentals of A.I and machine learning. Some quick fire tips for getting started below.

Learn Python

whilst you may have heard that machine learning runs faster in C++ or you can write genetic algorithms in javascript. Python is without a doubt the corner stone of modern day machine learning, the python A.I community is one of the fastest growing tech communities, with millions of python repos in GitHub and plenty of libraries this one is a no brainer. Some good python learning resources are as follows:

• CodeAcademy
• SoloLearn
• Udemy
• Udacity

Learn ML (Machine Learning)

whilst deep learning is the new kid on the block and the hype is justified, you should really learn how other machine learning models work. There may be use cases for your business model where you don’t have a lot of data or you just want to make a simple prediction, in this situation support vector machines or decision trees are the way to go. Machine learning is the bread and butter of A.I and covers a vast range of disciplines from statistics to networking. You don’t need to learn everything and all aspects, but you do need to be aware of them.

Learn Math

Its fair to say this part will most likely be the biggest turn off for those looking to pick up A.I, whilst it is unavoidable the challenge certainly isn’t unsurmountable for the average Joe. My advice is not to treat this as a tick box exercise, put aside some time weekly to maintain momentum, and try to cover these topics:

• Basics - brush up on high school math, from basic notation to standard approach i.e. BODMAS
• Brush up your Multivariate Calculus
• Brush up Linear Algebra
• Take a course in probability theory and statistical inference
• Take a basic course in Algorithms

DL (Deep Learning)

If you want to be on the cutting edge of A.I then this one is a no-brainer. Deep Learning (ML) is a subset of machine learning (ML) and has been shown to outperform all OTHER machine learning models 99% of the time when applied to a range of tasks. Whilst many of the principles are the same there are some core fundamental differences which is why it's important to understand what makes deep learning so powerful. You can check out some of my other blogs on deep learning to get the lowdown, but I strongly recommend heavy use of youtube, google and learning resources. When you are ready to start some practical examples, it is strongly recommended you start with tensorflow, this is a robust and easy to use Deep Learning framework. Tensorflow is still the best and most well tested framework out there, the community support is the fastest growing and there are endless examples and source code to learn from.

Picking your model

The key to being a strong A.I specialist is understanding what model to use and when to use it, this comes with experience and lots of study. There is no end point so don’t set your target too far in the future, you can start almost straight away if you have basic coding experience - the key is experimentation and developing your own process for taking feedback and improving your models.

Write down a list of problems that you feel passionate about , some of the most successful companies in the world were started simple because the founders were trying to find unique solutions to problems they faced.

• Automated gas and electric bill predictions are rubbish
• Dating apps require too much effort
• I want to automate my budgeting
• I can never tell when the job market is busy

Starting your own business is like ‘eating glass and staring into the abyss’ Elon Musk, you really need to be prepared to invest large amounts of time into your project and take the hits, simply put if you lack motivation, you just aren’t going to stick around when the going gets tough - and don’t delude yourself, it will get tough. Its important to think of the journey in your life so far, how have you got to where you are, what are the big issues you have noticed, where do you want to make an impact? Try to think of your A.I business as the culmination of all your struggles and work, you need to pour everything into it. If indeed your business does fail, you can leave in the knowledge that you went all in, as a result you have picked up a lifetime of experience…and then some.

Do your market Research

Once you have your problems defined, you need to start your long voyage of understanding your end customer.

• 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?

If you are doing a start up, or just want to make an impact in A.I you need to get out there. Create a simple clean website, with a to the point landing page, establish yourself as an A.I thought leader, create blog posts, try to solve problems proactively online, focus on expanding your network use linked In and Twitter to your advantage by sharing relevant and unique posts that you know resonates with your contacts. Create an email sign up on your site, then if you find you are getting even one or two sign ups after you have pasted your prices on the front page then it means there is a demand.

Now here is where the major stumbling block occurs. Elon Musk one of the greatest business men today generated a lot of his success in multiple endeavours, from scratch, by in his own words ‘reasoning by first principles’. Effectively, this is applying a strong fundamental physics principle by treating the problem as a completely new thing. To illustrate by example, when creating an electric car, Telsa first tried using the body of a Lotus Elise, whilst it worked, it was a vastly inferior product to their next generation Tesla S. Why? Because they assumed an electric car would be mostly the same structural design as a petrol car. When creating the S type however, they started from scratch, rather than asking ‘what works already’ they assumed no prior knowledge and explored all options to come to the optimal solution naturally. This allowed them to create novel solutions, such as realising electric cars don’t need much space under the front bonnet (which lead to significantly increased crumple zones) and the battery can be kept in the middle to add superior low centre of gravity, boosting performance ten fold.

Many failed business models have been so focused on the solution rather than the problem, because you will be in a vastly growing and competitive field - the product you create needs to be at least several times better than the nearest competitor, resist the urge to think optimistically or count on consumers taking risks. Your product should speak for itself.

Once you have done your market research its time to build a product, before you build a model you need to collect, organise your data as best you can. The quality of your data is the most important aspect of all, even more so than your model. The easy way to get data is to simply search public datasets, from sources such as Kaggle to UCI and github. If those don’t work, try to create data yourself by either optimise existing data or just get out there and make it happen, scrape data and explore.

Doing an ICO is all the rage at the present, but if you wish to avoid the legalities and are in it for the long run, it may be best to avoid this for now. Consider crowd funding is still a strong process for getting off the ground, and venture capital may also be a route - but be prepared to explain the technicalities to them as the best investors will have a high level understanding of the technology. It also helps to show that you already have existing inflow of capital or income as to show your product is proven.

Arguably the most important aspect, this will depend on the size of your company, the money you are willing to invest and other aspects related to the product. It is important to understand the trade off between hiring people that can do the job and people with the right mindset and belief in the product vision. Ultimately you want to maximise the passionate candidates and remediate for learning, as not only will this pay off in the long run, as Steve Jobs said ‘the beauty of having 10 great passionate employees, is they self police and they will decide the right people who get in’ the talent growth will become self perpetuating.