The Scariest Thing About AI Is the Competitive Disadvantage of Being Slow to Adapt
Grow Your Business, Not Your Inbox
“If you don’t have an AI strategy, you are going to die in the world that’s coming.” -- Devin Wenig, CEO of eBay
Whenever there is talk about artificial intelligence, Terminator-style robots that are stealing our jobs and will eventually take our lives come to mind. This picture portrays a false image of AI; AI is much more than just robotics and war machines.
The tools we use in our daily lives are getting smarter every day, from spam filtering in our inboxes to voice assistants in our homes. Google New uses AI to cluster similar news, and Nvidia uses it to create slow motion videos that outperform the Slow Mo Guys.
So, what is AI, and how can you use it?
What is AI?
AI has existed for a long time, with the earliest work dating back to 1940. But the most recent surge is thanks to two parameters -- increased machine performance and new algorithms especially in the field of deep learning, a practice that tries to mimic the human brain to make decisions. Of course, AI is not limited to dDeep learning. Essentially the paradigm is that when the task is too difficult to program, we let the machine learn by itself; we let it program itself. While AI has delivered stunning results in some fields, it has also showed its limitations -- to an extent that some experts believe a new “AI winter” is looming.
It is very important to know what AI is actually capable of -- and what it’s not. AI does not “understand” what it is doing, but when it looks at tons of data, it can detect patterns which have proven to be exceptionally efficient. Knowing AI’s sweet spot lets you make best use of it.
What AI can do.
Essentially, AI can help you with automation and prediction. Automation is not only about taking cumbersome, repetitive tasks from humans and assign them to tireless machines; we could do that even before AI. AI takes automation to a new level.
For instance, AI could do a quick analysis of your customer and give you a 360 degree view on their preferences and needs. This information could mark the difference between closing a sale and losing a sale as the customer interacts with your business. Also consider that AI can take care of this automatically, without any human intervention.
Many websites are currently using chatbots. While definitely not as smart as humans, they are 24/7 present assistants that take no breaks, respond immediately (no matter how many clients talk to it), remember customers’ preferences and can offer rich responses with links and images that a human would need significant time to compose.
Prediction is another fun topic, boosted by AI in recent years. Previously, we used “customers who bought X also bought Y” to up-sell or cross-sell, but AI disrupted that too. Take Netflix as an example. AI is able to look at much more data than just single products, and it can find patterns that lead to much more accurate results and increased sales.
AI lets you focus on what matters most, and this is important not only from a time and budget perspective but also because people will have much less legwork to do, and that increases their engagement and creativity.
This brings us to our next topic -- how you can use AI in your business.
To start using AI, you have two options; either rely on a company with an established AI service to automate your business, or create your own. The latter option would require tons of data to train the AI on, and you would need to have some development skills. It is the harder route, but it can really pay off, especially if your business model is very customized.
Whatever your choice, there are fundamentals you need to pay attention to.
Know what you need.
This is the first important step as it decides the direction you take and the algorithms you use. For instance, let's assume we need to find reporters who would be interested in covering our services. One approach would be to create an AI that would study the articles the reporters have written, classify them by topic, and find other reporters who write about similar topics.
This idea has two flaws:
- The number of topics was few, compared to the number of journalists. As such, the recommendations would always be the same (i.e., if a reporter had written about blockchain the system would always recommend them for that topic). We will end up with hundreds of recommendations for a single topic.
- There are new topics added each day, so we cannot not rely on a list of predefined topics.
As you see, while the idea looks good at first, we actually need a system that’s much more granular and precise. A better approach would be to let the system detect the patterns by itself, in the same way Google News works. This is the subject of “unsupervised learning,” where we share the data with the system and let it come up with suggestions on the similarities it finds there.
Knowing what you need is an important step, both for getting the results that work for you and for not wasting budget and time on methods that don’t.
Purify your data.
The myth is “the more data, the better.” Companies tend to gather as much data as they can and then throw it at the poor AI, which is supposed to make sense of it. In reality, AI does not understand a word of what it sees (or says). It only sees numbers and correlations between those numbers. Purifying the data before feeding it to the algorithm drastically improves the results.
If the data contains HTML tags, or if it contains text from several different languages, these can all throw AI off balance. This might seem obvious, but it’s not; just Google for “AI bias” to see how facial recognition AI failed to identify women of color because the AI was not properly trained on them.
Visualize your data.
This might seem trivial, but take my advice on this. Visualizing the data in the form of charts and graphs really helps you to see patterns as well as trash data, especially in the early stages. Even before feeding it to the AI, visualizing can help you see abnormalities as well as outliers, which in turn could help you with the data purification process.
Visualizing your data can be helpful, even if you don’t use it for AI. Making a chart of the weekdays and hours someone tweets can easily show you what hours and days they are most active on -- something difficult to see if you only have date and time numbers in front of you.
If you made it this far, you should know what AI is, what it is good for and what you need to get started. AI is a trend Andrew Ng claims to be widespread in every industry, “like electricity.” So the sooner you can get started with your own AI, the higher chance you will have to stand out and compete with those that have already gotten there.