The ultimate guide to starting AI

Many teams try to start an applied AI project by diving into algorithms and data before figuring out desired outputs and objectives. Unfortunately, that’s like raising a puppy in a New York City apartment for a few years, then being surprised that it can’t herd sheep for you.You can’t expect to get anything useful by asking wizards to sprinkle machine learning magic on your business without some effort from you first.Instead, the first step is for the owner — that’s you! — to form a clear vision of what you want from your dog (or ML/AI system) and how you’ll know you’ve trained it successfully.My previous article discussed the why, now it’s time to dive into how to do this first step for ML/AI, with all its gory little sub-steps.This reference guide is densely-packed and long, so feel free to stick to large fonts and headings for a two-minute crash course or head straight to the summary checklist version..But if the person in charge doesn’t understand your business, you may as well just flush that cash now.The key thing is that ML/AI is not magic and it doesn’t solve every problem..How does it help the business?.It’s the first job.Just because you can do something, doesn’t mean it’s a good use of anyone’s time.Imagine your ML/AI system is operational and ask yourself if you’re happy you sunk company resources into making it..Try my drunk island exercise if you need a bit of help brainstorming.Some of you decision folk are gorgeously fluent with data..Tell them what it makes, not how it makes it.Ask yourself, “Is this the end or the means?” If it’s the means, don’t talk about it for now.The trouble with many fluent folk is that you think everyone shares your fluency..(Dear reader, if you’re not sure whether you’re fluent in this, really force yourself to slow down. Keep asking yourself, “Is this the end or the means?” Make sure you focus on the ends for now.) Stakeholders might just not be able to follow the thread of your argument, which means your pitch will fall flat and you’ll miss a chance to make the world more awesome with AI.Some folks have trouble figuring out which variable is the input and which one is the output… it all looks like one big confusing lump to them and they need your focus to appreciate why the outputs are worth having.Reason 2: Tacit agreementAs an engineer who has been around engineers for a long time, I’ve noticed that our kind loves latching onto details..Make sure you’re not thinking about labelling just one or two cups. ML/AI makes sense for automating many repeated decisions..It’s not for one-offs.ML/AI is not for one-offs, so make sure your business needs an impressive number of items labeled.You’re imaging labelling at least a few thousand of them?.And when this thing is live, you’re sure you can’t just look the answers up instead of predicting them?.Along with your UX specialist, their participation in the project will help you ensure that groups affected by your creation are given a voice.Once you can clearly articulate what labels you’re after, it’s time for a quick reality check: do you have data about this business problem?No access to data means no point in proceeding..You don’t have to analyze the data yet (that’s later in the project) but you should check that you will actually have something to analyze when the time comes..This is just a quick overview of questions for identifying an nonstarter.Once you’ve cleared the reality checklist, it’s time to start recruiting, which you can do in parallel with wading through the rest of this guide.. More details

Leave a Reply