“The development of full artificial intelligence could spell the end of the human race,” Stephen Hawking told the BBC a few years ago. Elon Musk has delivered even more dire warnings, while Bill Gates recently suggested “we shouldn’t panic” because potential problems are not imminent.
Our sci-fi visions of the potential impact of artificial intelligence (AI) have always had a tinge of doomsday about them, but the truth is that we’re a long way from Skynet-style sentience.
Earlier this summer a Google Duplex demo showed off Google Assistant calling businesses to make appointments for people. It spoke in a natural, human-like way, as it made an appointment at a hair salon. The attendant on the other end of the call clearly had no idea they were talking to a digital assistant.
Passing the Turing test is impressive, but this demo is more about the potential of machine learning. It was able to learn natural speech patterns and master how to react to different responses, but it wasn’t thinking for itself. It’s an expert system trained and built on rules about making appointments. It’s exactly this kind of AI or machine learning that can help you to leverage more value from your data, improve customer service or unearth solid leads.
The Benefits Of Machine Learning
Once trained with enormous data sets, AI programs can outperform humans for some pattern recognition tasks. AI doesn’t get tired and it doesn’t need breaks -- it can perform with a very high level of consistency.
Asked to sort images of dogs into categories based on breed, the average human has a 5% error rate. This has already been surpassed by software from Baidu (4.58%), Google (4.8%) and Microsoft (4.94%).
There are a lot of possible applications for this in the business world. Quality control in manufacturing is an obvious place to start, but AI could also be trained to recognize wear and tear in parts and highlight when a Boeing turbine engine needs something replaced or to pick out patterns in financial records that might indicate fraud. It can also be used to answer basic queries from customers, reducing the load on human operators. The basic principles and algorithms can be conceivably applied to unlimited applications.
It’s More Accessible Than Ever
The days of requiring a supercomputer and a team of experts to write programs and devise algorithms are reaching an end. Employing AI doesn’t have to mean making an enormous investment. Businesses can benefit from the research and development that has already been done and buy configurable programs that enable them to deploy AI.
Platforms like Amazon’s SageMaker are offering building blocks to make it easier and faster for companies to create machine learning models and train them up. You can select the best algorithm for you and tune it to your requirements, then feed it data and train it. When you think it’s ready, test it out with some live data and assess how it performs, then retrain as necessary until you’re happy with the results.
We’re likely to see more and more choice in this arena over the next few years with increasingly specialized machine learning or AI programs available to buy off the shelf.
Preparing Your Data
One important prerequisite for the successful application of AI programs or machine learning models is data preparation. The more relevant data you can feed in, the better results you should see. But that means correctly identifying, aggregating and then formatting data that might be spread across many different systems.
You need to start by mapping out your existing data and deciding what is pertinent. You’ll need to create training sets and validation sets and then, when it’s working smoothly, figure out how to feed in live production data. Integrating your data and making it accessible without exposing it to unnecessary risk can prove tricky, so make sure you have a solid cloud data management strategy in place.
The quality of the data you feed in is crucial -- that old programming adage, “garbage in, garbage out” applies. It’s a smart idea to get your data in order and think about how to process it, leaving room for additional parameters in the future, so that you have a solid foundation to build your machine learning model on.
(Full disclosure: SoftNAS Inc. offers a cloud data platform.)
Building For The Future
By their very nature, machine learning and AI programs will continue to improve as they are exposed to more data, trained and corrected. Expert guidance and careful review will give way to further automation as we begin to trust the results, with experts only being alerted in situations where the AI is uncertain or some greater perceived risk is present.
While we may soon trust virtually autonomous systems to identify and cold call or message leads, deal with simple customer inquiries, identify broken parts or highlight and act upon insights from swathes of data, there’s still a leap between that and true AI capable of thinking for itself.
As AI gets cheaper and more accessible, companies that fail to adopt it may struggle to compete. Get your data in order and embrace it today to gain an edge.
