What Lies Ahead for AI

Any company looking to make investments in AI should be aware that, across the next five years, both machines and applications will be losing artificiality and gaining intelligence. The reliance on big data from bottom-up will decrease. Instead, they will be taking a logical approach to tasks and problems in the same way that humans do. Because AI will be more capable of reasoning, it will have broader applications than it did in the past. This presents an opportunity for early adopters. AI may soon be suitable for industries and applications that it couldn't handle in the past.?

In recent years, it's been standard to feed AI large amounts of data. This allowed it to create systems via deep machine learning. One example of this is driverless cars. They are trained on countless traffic situations. These neural networks are always ready to be fed more data, but they have clear limits. AI often struggles when there isn't much data to draw upon. A vehicle without a driver knows how to handle traffic, crosswalk, and ordinary pedestrians. However, it may not know how to response to kids running around in costumes on the night of Halloween.?

It's easy to stump many systems. For example, the facial recognition system on the iPhone can struggle to recognize people in the morning, when they look more rundown than they usually do. While AI has managed to defeat chess masters and the top players of Go, an ancient Japanese game, it may not be able to correctly identify an image if it's turned upside down. Data may also cause neural networks to inaccurately identify objects.?

There are also ethical and business concerns that must be taken into consideration when feeding data to systems. Businesses don't always have enough data to create neural networks that have distinctive abilities. There are privacy concerns associated with using large amounts of data from citizens. This is something that governments are taking action against. As an example, the General Data Protection Regulation from the European Union has strict requirements on how personal data can be used. It's not always clear how systems use data to arrive at conclusions. Systems can easily be exploited, as the Russians did during the United States presidential election in 2016. When something does go wrong, it isn't always easy to figure out what happened.?

Thankfully, as technology continues to develop, systems will take a top-down approach to problem-solving. They'll be less reliant on data and will be more adaptable. Not only will these systems be quicker, but they'll have real intelligence, as humans do. There are organizations and businesses that are already starting to use these types of natural systems. If you're trying to decide where you should invest over the next few years, you'll need to have a clear picture of what lies ahead for AI. These are the areas you should be focusing your attention on:?

Robots that Think Logically: When robots are able to think conceptually in the same way that humans do, they require far less data to learn new concepts.?

As an example, look at the Completely Automated Public Turing tests to tell Computers and Humans Apart (CAPTCHAs) that sites use to confirm that you're a human. These combinations of numbers and letters are made to be simple for people to figure out, but difficult for robots. Vicarious has used computational neuroscience to create a model capable of solving CAPTCHAs at a much higher rate than neural networks. Furthermore, it requires less data. The model from Vicarious was able to reach a near 67% accuracy rate for CAPTCHAs after being trained on only five examples for each character. A cutting-edge neural network, however, needed to use a CAPTCHA string training set that was 50,000 times larger.?

Top-down AI is able to mimic the approach that a human would take when they don't have enough information. It's more efficient than systems that need to be fed a lot of data, and it's capable of responding to many different situations. As an example, top-down AI is being utilized by Siemens to keep the gas turbine combustion process under control. This process is very complex. Gas and air both enter a chamber where they ignite and then burn at temperatures of up to?1,600 degrees Celsius. The amount of time that the turbine operations and the total volume of emissions produced vary on many interconnecting factors, including air flow, gas quality, and both internal and external temperatures.?

There are many companies that are teaching machines to approach the world logically. This makes it easier for AI to handle new situations, learn, communicate, and understand actions and objects. Many of the things that are easy for humans to do without data or training can be incredibly difficult for a machine. There are many easy questions that AI systems cannot consistently answer. For example, an AI may not be able to tell you if the clothes you hang in your closet will still be there the next day.?

Even though the latest advances in AI are fairly new, they're reminiscent of the AI developments made back in the 1950s. Many researchers in this period used top-down methods to teach machines to think the way that humans do. However, when there was no significant progress, and it was clear that bottom-up machine learning had a future, most researchers cast top-down learning methods aside. Now, new advances and techniques have breathed new life into top-down AI. The early promises of AI are finally coming true. Many companies will be smart enough to invest in this.