What is AI?
Artificial Intelligence is a massive scientific field. According to the Merriam-Webster dictionary: Artificial Intelligence (AI) “is a branch of computer science dealing with the simulation of intelligent behavior in computers.” Therefore, it is not hyperbole to assert that the goal of AI is to create algorithms with the capability of imitating intelligent human behavior.
Under this general definition, there are different branches of AI, with several different methods that allow AI to learn and grow. These different components of AI can be applied everywhere and one of them is referred to as Machine Learning. If Artificial Intelligence hopes to create computer systems that mirror or exceed human intelligence, Machine Learning is the mechanism by which these systems are taught to think. In this article, we will discuss different forms of AI and their applications in the world today.
Machine Learning
While Machine Learning and AI are often considered the same thing, Machine Learning is merely a component of AI. At the most basic level, Machine Learning (ML) is a method that enables automated decisions to be made based on a set of information. ML finds patterns detected in data to infer a conclusion. It is called machine learning because it is teaching machines to make decisions in situations they have never analyzed before.
A common approach to ML is showing the algorithm a data set and telling it the right decision. Once the model has been trained, an individual can input new data into the algorithm. With the new data, the machine makes decisions based on the prior training data set to arrive at a data-driven conclusions.
Deep Learning
Deep learning is a branch of ML where artificial neural networks are created and modeled after the human brain. In Deep Learning, the algorithmic network develops layers of data. Each layer is an algorithm that processes the information on the data it has received. Deep Learning mechanisms act to discover patterns on and across layers. Each layer acts to figure out a different aspect of the larger inputted data set. The ability to learn abstract concepts increases as additional layers are added to the neural network.
For example, neural networks can learn how to recognize human faces or cars. The first layer of neurons takes pixels from the images. The next layers then learn the concept of how pixels form an edge. That knowledge is then passed to the other layers. By combining that knowledge of edges, with the other information gathered over each layer, the network can begin to learn the concept of a face or a vehicle. This process of layering knowledge continues until the neural network algorithms begin to recognize specific features, i.e. an ear or nose on a human face or tires on a vehicle. For a more detailed explanation, check out this white paper on Model Agnostic Contrastive Explanations for Machine Learning Classification Models.
Natural Language Processing
Another form of AI is Natural Language Processing (NLP). If AI intends to imitate human intelligence, it must be able to communicate with humans as well as humans communicate with each other. NLP is very useful but also presents a challenge because human communication is not always straightforward. Different inflections in a voice, like sarcasm, can be very complex for an algorithmic system to understand because they often depend on context.
What can you do with AI?
Ever wonder why Netflix always has a good show recommendation? Or how Amazon shopping suggests things to buy? These are just a few of the many examples of how AI can be leveraged to become an client’s individual assistant. IBM has developed different AI software to help with a variety of client needs.
An AI conversational computing platform behaves like an assistant. However, like a human helper, an AI assistant can only be as smart as the information to which it has access. As with an assistant, it is important to ensure the AI has access to all of the data it needs to make informed decisions. IBM has fine-tuned its AI assistant to a 95% accuracy rate, a 99% decrease in response time, and over 1000 responses to different queries.
Bradesco's AI assistant achieved an exceptional accuracy rate when responding to customer queries. For Bradesco, Watson (one of IBM’s AI platforms ) was trained, tested, launched, and got great results. Watson was able to reduce response times from 10 minutes to mere seconds by leveraging the NLP embedded in Watson. The first task was to teach Watson Portuguese, Brazil’s culture, regional accents, and the way people in different regions ask questions. This was to help Watson get the data and information it needed to be as helpful as possible. The results were outstanding: this use of Watson was trained on 62 products and now answers 283,000 questions a month at a 95% accuracy rate.
Autodesk, a company that worked with IBM, saw that shifting to a subscription model required real-time customer service and wanted to use an AI to help them make that shift. By introducing Watson technology to Autodesk, the company saw encouraging results. Autodesk drastically improved their customer response times with an AI assistant. By using the Watson Assistant service, Autodesk developed a virtual agent that interacts with customers quickly and effectively. They saw a 99% decrease in response times. Watson Assistant services leverage Deep Learning and NLP to provide quality and effective services to clients wanting to make a shift.
The Royal Bank of Scotland trained their AI assistant on 1,000 responses to over 200 queries. Using Watson to power their digital assistant, the Royal Bank of Scotland created Cora. Cora has helped their live agents by handling mundane questions and by directly addressing those customers. This allows The Royal Bank of Scotland's live agents to focus their energy on more complex concerns. Forrester, an American research and consulting company, named IBM Watson Assistant as a leader in the business-critical area of conversational computing in The Forrester New Wave?: Conversational Computing Platforms in 2018.
Government and AI
Sonoma County, CA, launched Project Access which is an effort of coordinated care to empower self-sufficiency. With IBM workshops and technology powered by Watson, Project Access was able to help people who had been devastated by wildfires and many others who need assistance.
North Carolina fights Medicaid fraud using IBM analytics to sort through millions of claims in minutes., The state identified nearly $200 million in suspicious claims from the first 250 providers targeted. IBM Advanced Analytics software can identify patterns in data to flag any abuse or fraud that was occurring in the system. This software led to a 90% reduction in fraud.
Bias
With all the amazing applications of AI and Machine Learning, there is a challenge at the forefront of our minds. It is to ensure that AI doesn’t inherit human bias when interpreting the world. The reason why these systems can inherit bias is because of the assumptions humans make about the world. If you feed biased data to an AI then it will produce biased results. For example, if an AI system was created to identify different types of shoes and the creators only fed it pictures of high-heels and sandals, the machine would probably be unable to classify a pair of sneakers correctly.
This example seems like a minor oversight, but the effects of having our internal bias unintentionally embedded into an AI that identifies faces or answers questions based on cultural assumptions diminishes the effectiveness of that system. Over the next five years, bias will increase in AI systems. IBM researchers are creating a rating system that evaluates the relative fairness of an AI system. This is to ensure IBM continuously creates and delivers exceptionally fair and effective AI.
IBM can also help anyone examine the fairness of their own models with the AI Fairness 360 Open Source Toolkit. It contains over 70 fairness metrics and 10 state-of-the-art bias mitigation algorithms developed by the IBM research community. It is open, free to use, and the community invites anyone to use and improve the toolkit.
Moving forward
As a way to promote continued discovery and progress, the MIT-IBM Watson AI Lab was founded in 2017. It is a unique academic / corporate partnership to spur the evolution and universal adoption of AI. The MIT-IBM Watson AI Lab focuses research on healthcare, security, and finance, using technologies such as the IBM Cloud, AI platform, blockchain and quantum to deliver research to any industry.
IBM Could Pak for Data
IBM is enabling enterprises and government organizations to take the leap into AI for themselves. IBM Cloud Pak for Data is a fully integrated data and AI platform. IBM Cloud Pak for Data modernizes how businesses collect, organize, and analyze data, allowing them to infuse AI throughout their organizations. Built on the Red Hat OpenShift Container Platform, IBM Cloud Pak for Data integrates market-leading IBM Watson AI technology to meet ever-changing enterprise needs. IBM Cloud Pak for Data runs across any cloud and is deployable in less than 4 hours. This platform enables organizations to easily integrate their analytics and applications to speed innovation.
There are many different components of AI, like Machine Learning and Natural Language Processing, but we have just scratched the surface when it comes to their applications. With the help of academic and corporate communities, working together to build checks and balances to root out bias, the future of AI is fascinating and something every person should be paying close attention to.