A running joke with our Master's cohort was the business student who asked a group specializing in voice recognition software how quantum computing would affect their business model. It's like asking a doctor how finding the cure for cancer will affect their practice. It'll be great, but a 'cure' for 'cancer' is bizarrely vague and has little to do with how the doctor's daily practice works. It's a concept far out of left field, so far away from being a ubiquitous reality, and doesn't have any direct application to what they're working on that the question is immediately taken less seriously. It's a topic best suited to a stoner's living room musing about the future than a professional environment. One of our group chats is called Quantum Blockchain Learning, a joke emphasizing that we constantly hear pitches that focus entirely on the 'revolutionary' technologies that make headlines but don't do much beyond illustrating that the writers only know the barest details about what they're describing. Sure - this sounds condescending and conjures up images of the nerds getting together laughing about someone's ignorance of a niche subject. The point of this article is to emphasize that this mistake has very real implications for your ability to actually communicate your brilliant idea. Think about it: your investors and engineers ARE the nerds who laugh about others' ignorance at happy hours. They have all the financial incentive in the world to be as critical as possible of your idea. Otherwise they wouldn't be rich or talented in the first place. There's no reason to worry - you're reading this article! Just having a bit more idea of what the concepts you're espousing actually represent can mean the world to whoever you're speaking to.

Using vague buzzwords to describe your product has very real consequences for what you do. Besides the skepticism that investors will have in deciding whether or not to fund your idea, your engineers actually designing the thing will be completely lost on what you want to achieve. Let's start with the headliner: Artificial Intelligence. Most serious investors and business executives have caught on to the reality that Artificial Intelligence is an entire field, not a single technology. Using AI to describe how your product works is like saying 'physics' allows your delivery company to function. It's a detail that gives nothing away about your product. Happily, those who use 'AI' to describe what they're working on have largely stopped doing so. Instead, new buzzwords have appeared: machine learning, algorithms, robotics, NLP, blockchain, cryptography, you name it. These aren't much better. All of them are titles of introductory course names at your university of choice - not robust descriptions of what you do or how you do it.

Artificial Intelligence isn't a single technology. Instead, it is the merging of hundreds of technologies to achieve one goal: mimicking human behavior while multiplying its potential. Here's a short history of it. AI started out as philosophy - understanding that the mind and ideas are separate from the body and have the ability to 'learn' new things through abstraction. It grew into studies of logic and mathematics, eventually turning into computing. As computers became more commonplace and the potential data to compute became massive, we began bringing in aspects of biology, neuroscience, psychology, and cognitive science to solve problems. Around this time, Artificial Intelligence became a real discipline, encompassing many subfields that in turn, contained many subfields. Modern artificial intelligence practitioners often have expertise in one specific subfield, with a general idea of how the rest work. These subfields include:

1. Neural Networks - a combination of computing and biology that seek to model how the brain processes information.
2. Machine Learning - the use of statistics and algorithms to engineer automated systems.
3. Natural Language Processing - teaching machines how to read and how to pick up on task-relevant knowledge.
4. Robotics - use of intelligent control to promote autonomous exploration (i.e. using a computer program to interact with the world).
5. Computer Vision - enabling a computer to pick up on image features to begin identifying what it contains.
6. Speech Processing - same as above, but with human voice as the medium.

Now, onto the main issue presented in this article: how to talk about Artificial Intelligence in the best way to properly describe your idea so investors remain confident and engineers know what to do/provide recommendations. The best thing you - as the idea generator - can do is properly define your data. What medium does it take place in? Is it in pictures? Video? Sound? Text? Excel spreadsheets? Determining your data will determine what branch of AI you're using, but does not tell you what approach you're using. You cannot begin a meeting with a VC and say, 'oh I'm using NLP to do X, Y, and Z.' That is still the field, not the mechanism at work. You need to go beyond the field and state the specific class of algorithms or approach you'll use, in addition to how you use your data within that mechanism. Let's say you're a tech startup dealing with computer vision. Don't use computer vision to describe what you do - describe your algorithm! You don't have to go into detail, just reference ideas like 'Convolutional Neural Networks' to describe your approach. CNNs are a class of deep neural networks that are commonly applied to recognizing images by analyzing their pixels. Does it sound more intense than the buzzwords you're used to? Good. Nobody's investing in half-baked ideas anymore, you need to wow them with actual know-how before they'll hand over money or want to work on your project.

Onto Machine Learning. Always refer to it in relation to the medium and specific use-case you're working with. Machine learning is essential just informed automation, so what are you automating? Are you automating the extraction of specific text? You're using a subclass of NLP called Named Entity Recognition. Are you trying to get a mechanism to learn about its environment? You'll want deep reinforcement machine learning using a Q-learning algorithm. This is the kind of information that is most relevant to important players in the field and the engineers that will be working on your project. You may have noticed that many of these fields tend to overlap - that's exactly why it's weird to refer to anything less than model specific in your pitch. Machine learning can have a lot to do with NLP or nothing at all. If you keep things vague, the more likely it is that it will be met with skepticism.

I understand the importance of keeping things short and sweet to the point that you're able to make a coherent elevator pitch. But getting the right name and approach of your design right and keeping things short aren't mutually exclusive! Being able to talk about your model explicitly can take the same time as referring to the inner workings as 'AI magic,' with the added benefit of being much more descriptive. Explaining that your model uses pre-trained word embeddings to minimize the overhead of a machine summarization system allows you to convey that you know what you're talking about while being relevant enough for engineers to get interested in your product. It doesn't matter that you don't know how these things work under the hood - what does matter is that you know how your product will actually be built and is based on a feasible strategy, and you're not just hitching a ride on the shiniest new thing. Once you have the make and model of what you're trying to achieve, that can point your engineers and investors down a rabbit hole they'll be happy to dive into.

My last point is that people often frame their problem as if a single solution comprises the entire product. Many assume that once they start talking about NLP, that's all there is to their project. They think that they just have to cover the creation of the model and the rest will take care of itself. You really need to think long and hard about all of the additional facets of your creation. How are you going to house your results? Did you consider your database strategy - and how to allow that database to handle the network of information that your model will create? How are your results displayed? Did you consider how to properly package a ML-based application so it's insights aren't limited to field-specific professionals trained to handle that kind of data. These questions are just as important as defining the model you want to use, and can go the extra mile in reassuring your investors and programmers that you have a defined idea of what you want to achieve and how to get there. Additionally, multiple models might be required to clean, normalize, or even create your data constraints. Make sure your model can do everything you say it can by itself before pitching it.

An even bigger problem that a lot of these startups and big companies seem to have is understanding their data, which will be the focus of my next post. Understanding what data you have available, how to format it to be useful, and data maintenance and cleaning strategies can make or break a startup or an innovation team. Knowing how to describe your analytical approach through the lens of more specific descriptions of your product will enable this process to happen. Unless you specifically understand the model you want to use to tackle your data, you won't be able to pitch your product and you'll be stuck with a 'Hello World' application forever.