The Future of Artificial Intelligence

August, 5 2016

Subscribe and stay up to date

No spam, we promise! You will only 
receive essential emails.


The White House recently requested information on the future of artificial intelligence. Since it's our job to think about these things, as builders of AI-powered assistants for businesses, our Chief Data Scientist and Co-founder, Byron Galbraith, was an awesome person to provide insights on the topics they're asking about. Below are his answers.

Preparing for the Future of Artificial Intelligence

AI will have a major, transformative, and positive impact on Education.

This response addresses three of the topics posed in the White House’s Request for Information on Preparing for the Future of Artificial Intelligence: (1) the use of AI for public good, (2) the safety and control issues for AI, and (3) the scientific and technical training that will be needed to take advantage of harnessing the potential of AI technology. These topics are all addressed in the context of Education, with a specific focus on the upcoming role of AI in K-12 classrooms.

Topic 1: AI for public good


AI will enable and empower students and teachers, especially in low-income and underserved areas.

AI embedded in learning software and accessible via natural language interfaces will simultaneously empower students in learning traditional topics through individualized custom learning curricula, but also drive focus on 21st century skills seen as vital to modern education. AI will not, however, replace teachers in the classroom, just as B.F Skinner’s teaching machine, radio, cable television, personal computers, the Internet, and smart devices have also all failed at replacing teachers. AI will augment teachers, allowing them to focus more on roles as instructional coaches and facilitators of learning.

We are already seeing examples of this in the form of Intelligent Tutoring Systems (ITS), which can be seen as simple AI that reside inside learning software. ITS monitor student behavior such as analyzing how they answer questions or watching what choices they make inside a game-based learning environment. This analysis is used to adaptively adjust the scaffolding around the intended learning objectives, e.g. presenting additional guidance prompts or increasing the level of question difficulty. It can also be relayed to teachers in order to track how students are doing and alerting them to when students need additional assistance.

While these are very promising, a significant limitation of these AI is that they require the usage of some form of specialized computer software, such as a reading comprehension game. Access to computer and Internet resources by low-income families is frequently not guaranteed, either at home or at the schools predominantly serving them. Non-profit institutions are making excellent progress here, validating these methods, but as they largely rely on government and foundational grants, those solutions have difficulty scaling beyond the lifetime of the particular funded project.

An exciting potential solution to the problem of computer resource access is coming in the form of AI-powered chat bots – intelligent agents that you can communicate with via natural language. While these AI can also have great power embedded inside a domain-specific application, such as an interactive helper inside a physics simulation environment, they can also be accessed via standard messaging systems, such as SMS. Mobile phones and smart devices are much more prevalent than desktop or laptop computers. Having access to a tutor you can chat with at any time of day about various topics will have a much greater impact on the underserved populations, as access is not gated as much by privilege.

This potential boon to Education does come with some real concerns about security and privacy.

Topic 2: The safety and control issues for AI


AI-initiated decisions must be transparent to students, parents, and teachers, while administrators must be confident that proper student privacy regulations are followed.

AI are trained by practitioners and on existing data sources, both of which have inherent, systemic bias. Understanding those sources of bias and how they affect the decisions that AI make are important, especially when considering deploying those AI to a diverse set of socioeconomic and demographic populations. If an AI in a learning software application or chat bot interface collects data and makes decisions about students, it is very important that those individuals affected by an AI-initiated decision have a way to understand what reasoning led to that outcome, especially if it is negative or deleterious. This can be a complex and difficult undertaking, as many of the algorithms used in AI are difficult for even practitioners and experts to fully elucidate.

On the other side, both school administrators and vendors of products targeting K-12 must have guidance on how interactive AI, especially chat bots, are exposed to child online protection regulations like the Children’s Online Privacy Protection Act and California’s Student Online Personal Information Protection Act. Vendors must ensure that if their AI products are collecting data from children, that data is stored, encrypted, and access restricted according to all necessary guidelines.

The future prospect of AI in the classroom is bright, but there is still quite a long way to go in developing the technologies needed to realize that potential.

Topic 3: The scientific and technical training that will be needed to take advantage of harnessing the potential of AI technology


The future of educational AI is massively multidisciplinary.

In order to fulfill the vision and potential of natural language-powered AI for Education, we must have greater expertise with the computational processing of written language, especially in conversational settings. This includes understanding and anticipating user intent, translating that intent into actions, and generating acceptable natural language responses in return.

Built largely from the domains of computational, mathematical, and linguistic training thus far, these fields are required more than ever, but by themselves are insufficient for developing advanced AI, as a tremendous amount of natural language interaction in conversation relies on multiple levels of context and assumed shared experience. Psychology, sociology, and anthropology, therefore, are also going to need representation in the development of AI. Finally, expertise in Education and Learning are needed to craft the roles AI will place in the classroom and ensure the affordances of AI are exploited to maximize their effectiveness.

View all posts

Subscribe and stay up to date

No spam, we promise! You will only 
receive essential emails.

Subscribe and stay up to date