Scoping Chatbots for Safe and Effective Experiences – Source: securityboulevard.com

Mark McNasby, CEO and co-founder of Ivy.ai, co-authored this article.

Artificial intelligence has turned our world upside down, and organizations are rapidly realizing AI tools can change the way all industries operate. Whether it be increased productivity or streamlining revenue operations, AI has already proven its positive effects. But the development and adoption of chatbots is just as much an art as it is a science, in that you shouldn’t automatically assume the “smartest” bot is the best fit for your business. Chatbots that aren’t built thoughtfully have the potential to cause harm and confusion if we don’t regulate their construction. Additionally, the most effective chatbots are customizable to align with an organization’s structure and specific use cases.

So, what features are most critical for businesses as they look to drive productivity and revenue with AI? Let’s dive in.

The Shared Brain: A Comprehensive Knowledge Base

Among the various types of AI is the ‘shared brain’ concept. A shared brain is an advanced AI system designed to accumulate and utilize a collective knowledge base, enabling the bot to provide contextually relevant responses based on the specific source from which a user accesses it. As information is added to a bot’s shared brain, it begins to resemble a web, allowing for a complete understanding of various topics. By using a shared brain, users are getting the most comprehensive and accurate answers to their questions, straight from the source.

Consider a university financial aid office that wants to implement a chatbot to alleviate its staff’s workload. A chatbot can pull web addresses, PDFs, schedules, documents, etc., so when a student asks the bot on the financial aid website, “How do I apply?” it could pull from the admissions site and offer guidance for applying to the school instead of a financial aid plan, causing frustration for end users.

To optimize the shared brain concept, organizations can create multiple chatbots that function as experts in their areas, breaking apart the “web” of chatbots and allowing the “active brain” to take over as needed. Now, if a student asks the financial aid chatbot “How do I apply?” it will prioritize answers most relevant to the financial aid department. The bot is still operating under the shared brain structure and can continue to pull details from other pages, but now the active brain is given preference to answer the prompted question.

Web Crawlers: The Essential Knowledge Librarians

At a company where a shared brain might be used, it’s likely its information is being updated frequently, requiring staff to ensure the chatbot is delivering the most current information. This is where a web crawler comes in.

A crawler is a program designed to navigate an organization’s website or knowledge base by visiting web pages and indexing the content. It operates by starting from a seed URL and then following links from that page to other pages, creating a network of interconnected web pages. As it moves from one page to another, it collects information about the content, structure and metadata of each page it visits. This data is then processed and stored in a vector database, making it searchable by the chatbot.

This is a particularly critical tool for a university as there are multiple departments within a school, and the information on the website is ever-changing. This includes updated deadlines for applications, new event information, campus maps for high-traffic events, etc.

Ethical AI Usage

We know that technology comes with risks, and AI has raised a lot of questions about how to ensure safe deployment without putting users at risk. There are a few ways that leaders can deliver on their promise of safe AI. Companies actively implementing AI in their organization must take the time to educate all stakeholders – including students, professors and administrators –  as their products are developed. This includes being transparent about both the algorithm and approach.

In addition to practicing transparency, companies building AI-powered technology must also analyze their algorithms, which should be sanitized before they are sent to GPT to remove any personally identifiable or financial-related information. This means putting a fence around GPT so that it can only answer questions with the authorized data provided by your customers, ensuring your chatbot is built solely from your data.

Without proper guardrails in place around generative AI, someone could write a prompt revealing personally identifiable information (PII). For this reason, companies leveraging AI should be subject to audits and other security controls demanded by their customers or regulators. For higher education institutions, they must ensure the bots they use are FERPA compliant. Training and education on AI are essential to successful deployment and should be included in internal risk management strategies.

When choosing which AI your organization will implement, it’s important to center your decision on what information your organization holds and wants to distribute, as well as the potential risks associated with various bots. If companies keep ethics and efficiency at the forefront of AI implementation, they will gain customer trust and buy-in of AI, and when implemented correctly, everyone will feel the positive impacts.