There seems to be a glaring ambiguity as to exactly what artificial intelligence (AI) is, and how the discipline of AI should be categorized. Is AI a form of analytics or is it a totally new discipline that is distinct from analytics? I firmly believe that AI is more closely related to predictive analytics and data sciences than to any other discipline. One might even argue that AI is the next generation of predictive analytics and is born out of sophistication of analytics . Additionally, AI is often utilized in situations where it is necessary to operationalize the analytics process. So, in that sense, AI is also often pushing the envelope of prescriptive, operationalized analytics. It would be a mistake to say that AI is not a form of analytics.
I’ve seen AI applied to some of the most obscure topics you can imagine, ranging from industrial energy usage all the way to finding the right GIFs. Using Artificial Intelligence to improve and create solutions to today’s pressing business and social problems is one of the defining trends of the tech world for me.
So, if you are a PE / VC entity and are looking for investment opportunity in AI space, you will have to understand what kinds of AI companies exist and how this AI practice has evolved from Analytics practice.
There are three types of AI companies — core, applied, and industry
1. Core AI Companies
Core AI companies develop technology that improves parts of the AI creation or deployment process itself. Here are a few selected parts of that process and a few companies that are innovating in each:
Data scrubbing and cleaning: Trifacta, Paxata, Wealthport, Datalogue
Modeling: Sentient, Petuum, MLJar
Deployment: Yhat, Seldon
These companies all innovate in some specific, industry-agnostic part of the AI pipeline. Some of them are specific tools, while others purport to have an entirely new approach to AI that will revolutionize how it’s done (see Geometric Intelligence circa 2015).
If you’re investing in Core AI companies, you should probably have a good understanding of how this pipeline works. If you’re founding one of these companies, you should probably have experience deploying Machine Learning and AI at scale.
2. Applied AI Companies
A bit on the more specific side, Applied AI start-ups develop technology that helps companies across different industries perform a specific task using AI. As with the above, here are some examples of those applications and a few interesting companies in each:
Analysing and understanding text: Indico, Synapsify, Lexalytics
Analysing and understanding images and videos: Clarifai, Kairos, Imagry, Affectiva, Deepomatic
Bots / Voice: Init.ai, MindMeld
While investors can get away with not having experience in one of these specific applications, founders will likely have done projects involving this stuff in the past.
The implementation of AI in this scenario corresponds to the implementation of predictive analytics. At its core, predictive analytics is, naturally, about predicting something. Who will buy? Will certain equipment break? Which price will maximize profits? Each of these questions can be addressed by following a familiar workflow – First, we identify a metric or state that we want to predict and gather historical information on that metric or state. Next, we gather additional data that we believe could be relevant to predicting our target. Then, we pass the data through one or more algorithms that attempt to find a relationship between the target and the additional data. Through this process, a model is created that produces a prediction if new data is fed to it. If a customer had this profile, how likely would she be to respond? If we priced at this point, how much profit might we expect?
The goals and steps followed within an AI process are the same. Let’s look at two examples:
Take image recognition. First, we identify a bunch of cat pictures. Then, we grab a bunch of non-cat pictures. We pass a deep learning algorithm over the images to learn to accurately predict whether an image is a cat. When provided with a new image, the model will answer with the probability that the image is a cat. Sounds a lot like predictive analytics, doesn’t it?
Let’s now consider natural language processing (NLP). We gather a wide range of statements that have specific meanings we care about. We also gather a wide range of other statements. We run NLP procedures against the data to try to tease out how to tell what is important and how to tell what is being asked. As we feed a new line of text to the process, it will identify what the point of the statement is in probabilistic terms. The NLP process will assign probabilities to various possible interpretations and send those back (think Watson playing jeopardy). This also sounds a lot like predictive analytics.
3. Industry AI Companies
The final category of ML/AI companies apply these techniques to specific business problems in specific verticals. This is undoubtedly the lion’s share of the actual number of companies being founded, and in many ways, represents the true promise of AI — solving actual and immediate problems with new techniques. Here, it’s easier to give companies as examples. The format is always “AI for ________”:
DigitalGenius: AI for customer support
Cylance: AI for cyber threat prevention
X.ai: AI for scheduling meetings
Drive.ai: AI for autonomous vehicles
The implementation of AI in this scenario corresponds to industrialized embedded analytics. A major trend today is to embed analytics into business applications so that the models are utilized in an automated, embedded, prescriptive fashion at the point of a business decision. For example, as a person navigates a web page, models are utilized to predict what offers should appear on the next page. There is no human intervention once the process is in place. The process makes offers until told to stop.
Many applications of AI today also require industrialization. For example, as an image is posted on social media, it is immediately analysed to identify who is present in the image. As I make a statement to Siri or Alexa, it attempts to determine what I said and what the best answer is. While this qualifies as a more advanced application of predictive analytics that moves into embedded, prescriptive, automated processes, it is still very much in line with how industrialized embedded analytics are being used today.
The common theme among these companies is that they take Machine Learning / AI and use it on a specific problem or space. When researching investments like this, investors should look at both the AI itself (if it works well) and the business case (whether it’s compelling). In x.ai’s case, investors need to know if the AI works, but they should also consider whether AI is the best way to solve the scheduling problem, and whether scheduling is a problem worth solving at all. With the other two types of companies, this is rarely a consideration. Founders of these types of companies can often not have AI experience and can even be non-technical (with the right supporting team and CTO, of course).
Your journey to a fruitful AI investment will be far easier if you recognize and embrace AI as sophistication of analytics and understand the true categorization, and then task your analysts with leading the charge. Don’t cause confusion and redundancy by considering AI to be something completely different.