The Basics of Artificial Intelligence
Researchers have been developing artificial intelligence since the 1950s and it gained worldwide popularity when Siri and Alexa came to market. Some of the popular applications of AI include search, personalization, translation and semi-automated cars.
There are three subtypes of artificial intelligence: machine learning, natural language processing and robotics. In this blog, we'll dive into each one in more detail.
This blog post was authored by Suman Seshadri, lead product manager at DocuSign.
Three types of artificial intelligence
Machine learning (ML)
The primary usage of machine learning today is to help with prediction. In order to come up with an accurate prediction using AI, analysts would need to build mathematical models which mimic the real thing. Machines would then need to be trained using a vast amount of unbiased labeled sample data covering a wide range of scenarios. For example: if we are attempting to predict whether a person would like a movie or not, the mathematical model could use parameters such as the movie title, description, cast and genre. The model would then be fed with a set of example training data which includes an indication of whether the person liked those movies or not.
There are three types of machine learning: supervised machine learning, unsupervised machine learning and semi supervised machine learning. Supervised machine learning is where the model takes in specific inputs and produces specific outputs. Unsupervised machine learning is about building a model where there are no explicit inputs and the machine learns by itself using methods such as pattern matching. Semi supervised machine learning is where the model uses sample data and actively uses new data to predict better.
Another way of classifying machine learning is shallow learning versus deep learning. As the name suggests, shallow learning is about the machine doing just a few levels of processing while deep learning is about the machine doing multiple levels of processing to arrive at the specific answer, for example: face-recognition. As the number of layers of processing increases in deep learning, the processing complexity increases as well.
Natural language processing (NLP)
NLP is about machines being able to understand structured human languages such as English. Syntactic parsing is possible today and machines can analyze sentences and can recognize nouns, verbs, subjects, objects, etc. However, the challenging part of NLP is about machines being able to understand the context of the real world. Back in 2011, a blogger funnily noted that every time the movie actress Anne Hathaway was in the news, Berkshire-Hathaway's stocks went up! This happened because the machines found a correlation without context. Therefore, the context needs to be fed into the data or the machine needs to be trained with tons of data which will enable more accurate pattern recognition. Therefore, machine learning and NLP go hand-in-hand most of the time. Common examples of NLP applications include language translation, search, smart assistants, chat bots, email filters and text generation, to name a few.
There are two methods by which machines can generate natural language. The first method is to generate the natural language from scratch which tends to be very hard as the grammar and sentence structure could be very different between languages. The second and the easier method is to use templates. For example: let’s say a system aggregates user feedback from a number of sources such as social media, support tickets and surveys, NLP can be used for sentiment analysis of the feedback and can be used to automatically respond to customer complaints for certain scenarios. Machine learning can be used to classify and choose the right template to be used. An example of a very simple template based response could be: “Dear <customer name>, we have received your valuable feedback on <product name>. A customer support representative will contact you within the next <hours> hours.”
Robots are machines that move using motors and sensors. They come in a variety of shapes and sizes and are used by high-tech companies for jobs such as stocking shelves, packaging and delivering boxes. Another major application of robots is in autonomous vehicles where robots can do a much better job than humans in certain scenarios where they can use data from several sensors.
Questions to answer before adding AI to your business strategy
- What problems are you trying to solve?
- Can machine learning help solve these problems?
- Are there off-the-shelf products with AI that solve these problems?
- What is the role of humans and machines?
- What human resources are available to dedicate to an AI project?
- Is sufficient data AI-ready: clean, well organized and labeled?
- What acceptance criteria are required and available?
- Does the company have the technical and managerial talent to adopt AI-enabled solutions?
Artificial intelligence at DocuSign
Several DocuSign products use artificial intelligence to deliver exceptional customer experiences. Machine learning and data extraction is used for searching, analyzing agreements and for providing proactive notifications to customers. AI also helps classify agreements into logical groupings based on specific clauses and is especially useful in cases where businesses are dealing with a large volume of agreements. Thus, AI helps remove manual processes enabling our customers to utilize their time more efficiently.
Future of AI
As AI becomes more and more prevalent there is the concern that human jobs will be replaced by machines. So, the question boils down to “What value is the human adding?”. The greatest value that humans bring is in the last-leg of problem solving and in the final decision making, taking into consideration the real-world context and common sense. For example: in the future, with advances in AI, creating, managing and maintaining agreements could become far simpler. Systems will be able to smartly recommend sections of the agreement to be drafted and will be able to analyze risks better. AI will also be able to streamline processes reducing the level of effort and the time spent on routine tasks while key decisions will still be made by humans.
The future of AI is very promising with researchers aiming to build machines that can truly understand humans in the context of things around them. AI experts are also working to make machines more “self aware” to bring down error rates. Last but not the least, researchers are continuing to push the boundaries of AI where they are figuring out ways in which machines can explain the rationale behind their decisions to clearly articulate why they predicted what they did.